Author: Pete

You just finished a 45-minute client interview. The recording is on your phone, the good quotes are somewhere in the middle, and typing it out yourself would eat the rest of your afternoon.

Here is how to transcribe audio with AI: first check whether the tool you recorded in already made a transcript (Google Meet, Word and the iPhone Voice Memos app often do). If not, upload the file to Word for the web, ChatGPT or Gemini, tell it what you want (word for word, speaker labels, timestamps), and give it the names and jargon it will hear. Then spend ten minutes checking names, numbers and every quote you plan to use against the audio. The AI does the typing. You check what matters.

Below: which route fits your recording, a prompt you can copy, how to get a cleaner transcript, the check that keeps a wrong quote out of your report, and what to do with the text once you have it.

Pick the right route for your recording

There is no single best transcription tool. The right one depends on where the audio lives and what you already pay for. Start at the top of this list and stop at the first one that fits.

1. The meeting already happened in Google Meet

Look for an existing transcript before you upload anything. Google Meet can transcribe a meeting while it runs (Meeting tools, then Transcribe). The transcript is saved in the organizer’s Google Drive and attached to the Calendar event, and the host gets an email with a link after the call (Google Meet Help, Use Transcripts with Google Meet). Many other meeting tools have a similar transcript setting, so check yours first.

2. It is a voice memo on your iPhone

On a recent iPhone, the Voice Memos app can show a transcript of a recording (the feature is not available in every region or language). Tap the recording, tap the More button, then choose Copy Transcript and paste it wherever you need it (Apple Support, View a Voice Memos transcription). The Notes app can record and transcribe too on iPhone 12 and later; Apple warns that deleting the recording there also deletes its transcript, so add the text to the note first (Apple Support, Record and transcribe audio in Notes).

3. You have Microsoft 365

Word has a Transcribe feature: Home, Dictate dropdown, Transcribe, Upload audio. It takes .wav, .mp4, .m4a and .mp3 files, separates the speakers, and gives you timestamps you can play back while you correct the text. A Microsoft 365 subscription covers up to 300 minutes of uploaded audio per month, and in a browser it works in Edge and Chrome (Microsoft Support, Transcribe your recordings). For an interview, this is often the most comfortable editor of the lot, because you can play and fix each line in the same pane.

4. You use ChatGPT on a paid plan

You can attach an audio file (MP3, M4A, WAV, FLAC and more, up to 512 MB) to a ChatGPT chat and ask for a transcript. Audio uploads are not on the Free plan. OpenAI also says plainly that transcripts may contain errors and speaker identification may be unreliable, so you review important details against the recording (OpenAI Help, Uploading files and audio to ChatGPT). The big plus: in the same chat you can ask follow-up questions about what was said.

5. You use Gemini

The Gemini app accepts audio files too (Add files, then Upload files). On a free account the total audio length per upload is 10 minutes; Google AI Pro or Ultra raises that to 3 hours, and each audio file can be up to 100 MB (Gemini Apps Help, Upload and analyze files). That makes the free version fine for a short voice note, and too small for a full interview.

6. You want to transcribe lots of files automatically

If you get a steady stream of recordings (sales calls, podcast episodes, support voicemails), the OpenAI speech-to-text API can take files up to 25 MB in formats like mp3, m4a and wav. You can pass it a short description of the recording and a list of expected terms to help with names and jargon (OpenAI API docs, File transcription). This needs a developer, or an automation tool such as n8n or Zapier in front of it. Our n8n tutorial for beginners shows how such a flow is put together.

How to transcribe audio with AI in ChatGPT or Gemini, step by step

1. Check the file. MP3, M4A or WAV work almost everywhere. If your recording is a video, export the audio track first: ChatGPT refuses files it identifies as video.
2. Cut what you do not need. Ten minutes of waiting for people to join is ten minutes of noise for the AI and for you.
3. Attach the file and paste a clear instruction. Tell it the format you want, who is speaking, and the words it will stumble on.
4. Read the first two minutes against the audio. If the speaker labels are wrong from the start, correct them in a follow-up message before you go further.
5. Save the transcript in a document of your own, with the date and the name of the recording at the top.

Here is a prompt you can copy and adjust:

Transcribe this recording word for word.

Context: a 40-minute interview between me (Sam, the interviewer) and Lena Hart, operations lead at a logistics company, about how her team plans weekly routes.

Rules:
- Label each speaker: SAM or LENA.
- Start a new paragraph at every change of speaker and add [mm:ss] at the start.
- Keep filler words out, but do not shorten or rephrase anything else.
- Names and terms you will hear: Lena Hart, RouteMaster, TMS, Q3, Rotterdam depot.
- If a passage is unclear, write [unclear mm:ss] instead of guessing.

That last rule matters more than it looks. You would rather see a gap you can fill from the audio than a smooth sentence nobody said.

Get a cleaner transcript before you even upload

Most transcript errors start in the recording, not in the AI. A few habits help a lot:

• Put the microphone close to the people talking. A phone in the middle of a big table hears the air conditioning better than the quiet person at the end.
• One voice at a time. When people talk over each other, every tool struggles to split the speakers. OpenAI’s own help page warns that speaker identification may be unreliable.
• Say names at the start. “This is Sam, talking with Lena from operations” gives the AI and your future self a clear anchor.
• Give it your vocabulary. Product names, client names and abbreviations are where transcripts go wrong. OpenAI’s documentation suggests giving the model context and the terms you expect for exactly this reason (OpenAI API docs). In a chat, a short list in your prompt does the same job.
• Mind the language. All the tools above list a set of supported languages, and OpenAI notes that performance varies by language. If your meeting switches between English and Dutch, say so in the prompt.

Check it before you trust it

AI transcripts read beautifully, which is exactly why mistakes slip through. A wrong word looks just as confident as a right one.

There is research on this. A Cornell-led study of OpenAI’s Whisper model found that while most transcriptions were highly accurate, roughly 1% contained whole phrases or sentences that were not in the audio at all, and these happened more often around longer pauses and silences (Koenecke et al., Careless Whisper, FAccT 2024). That study tested a 2023 model, and tools have changed since, but the lesson holds: the transcript is a draft, the recording is the source.

A ten-minute check that catches the errors that hurt:

1. Every quote you will use. Play the exact moment and compare word for word. A quote in a case study or a report has to be what the person said.
2. Every number and date. “Fifteen” and “fifty” sound alike. So do many prices.
3. Every name. People, companies, products. Search the transcript for each one.
4. Any sentence that surprises you. If it does not sound like the speaker, listen. Pay extra attention to the bits after a long silence.
5. The speaker labels on decisions. “Who said we would ship on Friday” matters later.

Word makes this easy because each paragraph is linked to its spot in the audio. In ChatGPT or Gemini, the [mm:ss] timestamps from the prompt above do the same job: jump to the time and listen.

Turn the transcript into something useful

A transcript on its own is a wall of text. The real time saving comes when you ask the AI to turn it into the thing you actually need. In the same chat, try:

Using only the transcript above:
1. List every decision, with who made it and the timestamp.
2. List every action item: what, who, by when (write "no date given" if none was said).
3. Give me the 5 strongest quotes from LENA about route planning, word for word, each with its timestamp.
Do not add anything that is not in the transcript.

Then go where the work goes next:

• For a meeting, follow our guide to summarize meeting notes with ChatGPT, or turn meeting notes into a report your manager will read.
• For a customer interview, the quotes feed straight into a case study written with AI or into your market research.
• For a recording in another language, transcribe it first in its own language, then translate the document with AI. You keep the original to check against.
• For a quick follow-up after a call, ask for a draft and polish it with our tips to draft reply emails faster.

A worked example

This is an illustration, not a customer story. Say you are a consultant with a 38-minute interview recorded on your phone as an M4A file of about 35 MB.

• Gemini free is out: the limit is 10 minutes of audio.
• The OpenAI API is out unless you split the file: it is over 25 MB.
• ChatGPT on a paid plan takes it in one go (well under 512 MB), and so does Word for the web if your 300 monthly minutes are not used up.

You pick Word because you want to fix lines while listening. You start the upload and answer email while it works. Then you paste the corrected transcript into ChatGPT with the action-item prompt above, check the five quotes against the audio, and your interview notes are done before the end of the day.

Privacy and consent: settle this before you press record

• Ask first. Tell people you are recording and why, at the start of the call. It is basic courtesy, and depending on where you and the other person are, the law may require it. If you are unsure, check with your company’s legal or privacy contact.
• Know where the audio goes. Word stores uploaded recordings in a Transcribed Files folder on your OneDrive, where you can delete them (Microsoft Support). Google Meet saves transcripts in the organizer’s Drive.
• Check your data settings. Every AI tool has its own settings for how your uploads are stored and used. Read them once for the tool you pick, and for confidential client calls use the tool your company has approved.
• Do not upload what you would not email. Health details, salaries and personal data about third parties deserve extra thought, whatever the tool.

Questions people ask about AI transcription

Can ChatGPT transcribe audio for free?

Not by file upload. OpenAI lists audio uploads for paid ChatGPT plans and workspaces only. Free routes that do exist: Voice Memos or Notes on a recent iPhone, Gemini for recordings up to 10 minutes, and a transcript from your meeting tool if your plan includes it.

How long a recording can AI transcribe?

It depends on the tool. ChatGPT takes audio files up to 512 MB, though OpenAI says very long recordings are handled on a best-effort basis and can time out. Gemini allows 10 minutes on a free account and 3 hours on Google AI Pro or Ultra. Word covers 300 uploaded minutes a month on a Microsoft 365 subscription. For longer material, split the file at natural breaks, never in the middle of a sentence.

Can AI tell the speakers apart?

Usually, but not perfectly. Word separates speakers automatically and lets you rename them. In a chat tool, tell it who is in the recording and how to label them, then check the labels on anything that matters, like who agreed to what.

Is an AI transcript accurate enough to quote?

Only after you have checked the quote against the audio. Use the transcript to find the moment fast, then listen to that moment yourself.

What is the difference between transcribing and summarizing?

A transcript is everything that was said, in order. A summary is the AI’s selection. Get the transcript first and keep it: when someone asks “did she really say that?”, the summary cannot answer, the transcript and the recording can.

Start with your next recording

Pick the route that fits what you already have, use the prompt above, and do the ten-minute check. After two or three recordings you will know which tool suits your work, and typing out interviews by hand will feel like something from another decade.

If you need notes during live Zoom or Teams calls rather than a transcript after the fact, compare the best AI note taker for meetings by fit for your stack.

Want more practical how-tos like this one, step by step? Join Launchpad for free on our home page, or right here:

The survey closed on Friday. You have a spreadsheet with 140 rows, eight rating questions and two open text boxes full of honest, rambling answers. Your manager wants “the key takeaways” by Wednesday, and you have no idea where to start reading.

Here is how to use AI to analyze survey data without fooling yourself: export the responses, strip out names and emails, give every row an ID, and upload the file to ChatGPT, Claude or Gemini. Let it count the rating questions and break them down by group. For the open answers, have it propose a short list of themes, fix that list yourself, then let it tag every answer by ID. Check a sample against your own reading, pull the quotes from your original file, and write the summary. The AI does the sorting and counting. You decide what it means.

That is about an hour for a survey of a few hundred responses. Below are the six steps, prompts you can copy, a worked example and the checks that keep a wrong number out of your report.

What AI is good at here, and where it slips

Survey data comes in two kinds, and AI handles them differently.

• Closed questions (ratings, multiple choice, yes/no) are numbers. ChatGPT and Claude can run real calculations on an uploaded spreadsheet, so counts, averages and breakdowns by team or region come back fast. OpenAI’s help page explains that ChatGPT writes and runs Python for this kind of work, and tells you to review the method and the assumptions before you rely on the result (OpenAI Help, Data analysis with ChatGPT).
• Open answers are the slow part for a human and the strong part for AI. Sorting hundreds of comments into themes can take a whole afternoon with a highlighter. Researchers now use language models for exactly this, with people checking the work. Pew Research Center says AI tools assist its researchers with sorting open-ended survey responses into categories, while the researchers still design the analysis and interpret the data (Pew Research Center, How we are and are not using AI).

Where it slips: a chatbot that reads your answers as plain text and “estimates” a count can be off, it can merge two rows or skip one, and a quote in a summary can be quietly reworded. Every step below is built around those three failure points.

Before you start: clean the export (10 minutes)

Get the raw responses out of your survey tool as a spreadsheet. In Google Forms, open Responses and either use Link to Sheets or More > Download responses (.csv) (Google Forms Help, View and manage form responses). Microsoft Forms, SurveyMonkey and Typeform all have a similar export.

Then, before anything goes near a chatbot:

1. Delete names, email addresses and phone numbers. Also delete anything that points to one person, like “the only designer in the Leeds office”. Pew’s own rule is that it never exposes respondents’ personal identifying information to an open, public AI tool. That is a good rule to borrow.
2. Add a response ID column (R001, R002 and so on). Every check later in this guide depends on it.
3. Keep one row per respondent and one column per question, with plain column names in the first row. OpenAI’s guidance asks for exactly that: descriptive headers, one row per record, no empty rows splitting the data, no second table on the same sheet.
4. Check your company’s rules on AI tools if this is customer or employee data. A work account with data controls is a different thing from a personal free account.

Save it as a new file. Keep the original untouched, because you will pull quotes from it at the end.

Step 1: Have the AI describe the file back to you (2 minutes)

Upload the clean file. In Claude, use the + button and Add files or photos; CSV is a supported type (Claude Help, Upload files to Claude). In ChatGPT, attach it to your message the same way. Do not ask for insights yet. Ask it to tell you what it sees.

This file holds the responses to a survey of [who: e.g. our 140 employees / customers who bought in Q3].
The goal of the survey was [one sentence].

Before any analysis:
1. Tell me how many rows and columns you see.
2. List every column with its type (rating, multiple choice, open text, ID, date).
3. Flag empty rows, duplicate response IDs, and answers like "n/a" or "-".
4. Tell me which columns look like they still contain personal data.
Do not summarize the answers yet.

If the row count does not match what your survey tool says, stop and find out why before you go further. Usually it is a header row, a blank line or a test response you forgot about.

Step 2: Get the numbers on the closed questions (10 minutes)

“Analyze this survey” gets you a generic essay. Ask for the tables your report actually needs.

Calculate from the file, do not estimate.

1. For every rating question: number of answers, average, and the share of each score (1 to 5).
2. Break those down by [team / region / customer type]. Show the number of respondents (n) in every group.
3. Mark any group with fewer than 10 respondents as "small group, read with care".
4. List the 3 biggest differences between groups.

Show the results as tables. Then tell me exactly how you calculated each one.

Now check three numbers yourself. Pick one average, one percentage and one group count, and work them out in your spreadsheet with a pivot table or a COUNTIF. If the formula part scares you, our guide on using AI to write Excel formulas gets you there in a couple of minutes. If all three match, you can trust the method a lot more. If one is off, ask it to show how it got there.

Watch the small groups. If the sales team is four people and two of them gave a 1, the chatbot will happily report “50% of sales is very unhappy”. True, and almost meaningless. That is why the prompt asks it to show n everywhere.

Step 3: Turn the open answers into a theme list (10 minutes)

This is the part people get wrong. They paste 300 comments and ask “what are the main themes?” The AI produces a neat list, with counts, and nobody can tell whether those counts are real.

Split it into two jobs instead. First, only a theme list, built from a sample:

Here are 60 answers to the question "[exact question text]". Each starts with its response ID.

Propose a list of 6 to 10 themes. For each theme give:
- a short name
- one sentence that says what belongs in it
- what does NOT belong in it
- two response IDs that are clear examples

Add "Other" and "No real answer" as the last two themes.
Do not count anything yet.

[paste 60 answers with IDs]

Read that list like an editor. Merge two themes that mean the same thing, split a theme that is really two complaints in one, and rename anything vague. “Communication” is too broad to act on. “Hear about changes too late” is something a manager can fix. The list you end up with is yours, not the AI’s.

Step 4: Tag every answer, in batches, by ID (15 minutes)

Now give the final theme list back and let the AI tag every response. For a few hundred short answers, an uploaded file in ChatGPT or Claude can handle it in one go. For long answers or a few thousand rows, work in batches of 100 to 200.

Use ONLY the theme list below. Tag every answer in the column "[column name]".
An answer can get more than one theme. Use "Other" if nothing fits.

Return a table with exactly one row per response ID and two columns: Response ID, Themes.
Do not skip, merge or invent response IDs. At the end, tell me how many rows you returned.

Theme list:
[paste your edited list with the one-sentence definitions]

Check the row count against your file. Then paste the table back into your spreadsheet next to the original answers, matched on the response ID, and do the counting there with a pivot table. Counting in the sheet is exact, and you can show anyone how you got the number. In ChatGPT or Claude you can also ask it to add the theme column to the file and give you the spreadsheet back; Claude’s file tools can create an Excel file directly (Claude Help, Create and edit files with Claude).

Step 5: Check the AI against your own reading (10 minutes)

Pick 20 random responses. Tag them yourself first, then compare with the AI’s tags. If you agree on most of them, carry on. If you keep disagreeing on the same theme, the definition is the problem: tighten that one sentence and re-run the batch.

That is a small version of what professional survey teams do. For one 2026 report, Pew built a set of 1,200 answers that people had sorted by hand, ran its prompt on them, looked at every case where the model and the humans disagreed, and refined the prompt several times until the agreement was good enough (Pew Research Center, methodology for the bogus respondents report). A July 2026 report for the General Social Survey found that language models can effectively support this work, that human review stays important, and that both people and models were more accurate on factual answers than on opinions (GSS Methodological Report 150, NORC).

The practical lesson for a work survey: a comment like “the new rota is posted on Thursdays, too late to plan” is easy to tag. A comment like “fine I guess, same as always” is not, and you should expect to disagree with the AI on those. Sarcasm in particular trips it up, so read the negative and the “Other” pile yourself.

Step 6: Write the summary people will actually read (10 minutes)

Now let the AI help with the writing, using only the checked numbers.

Here are the checked results of our survey: [paste the tables from Steps 2 and 4].

Write a one-page summary for [audience: e.g. the leadership team].
- Start with the 3 findings that matter most, each with its number and n.
- For each open-answer theme, give the count and list 2 response IDs that illustrate it best. Do not write quotes yourself.
- Separate what the data shows from what we might do about it.
- Flag any finding based on a group smaller than 10 people.
Plain English, no jargon.

Notice the line about quotes. Ask for response IDs, then copy the real words from your original file. A summary that “cleans up” a respondent’s sentence is no longer a quote, and if someone checks, your whole report loses trust.

If the summary needs to become slides or a longer write-up, our guides on making a presentation with AI and writing a work report with ChatGPT pick up from here.

Already using Google Forms or Microsoft Forms? Try the built-in AI first

For a quick read, you may not need to export anything.

• Google Forms with Gemini: on an eligible Google Workspace or Google AI plan, the Responses tab summarizes open text questions once there are 3 to 200 responses. It can also group answers into themes and open a Google Sheet with the theme counts and a pie chart (works best with 8 to 200 responses), and an Actionable Insights option needs at least 12 responses and 3 questions (Google Docs Editors Help, Summarize and analyze responses with Gemini in Google Forms).
• Microsoft Forms with Copilot: with a Microsoft 365 account and a Copilot license, go to View responses and open Copilot to get a summary of key findings and ask follow-up questions about the data (Microsoft Support, Welcome to Copilot in Forms).

These are great for a first look. For anything that goes into a decision, you still want the checks from Steps 4 and 5: themes you have edited, counts you can trace, and quotes from the source.

A worked example (fictional, to show the flow)

Sam runs operations at a 140-person logistics company. The quarterly pulse survey got 112 responses: six ratings from 1 to 5 and one open question, “What one thing would make your work week easier?”

• Step 1: Claude reports 113 rows. Row 113 is Sam’s own test response from before launch. Deleted.
• Step 2: The average for “I have the information I need to do my job” is clearly lower in the warehouse than in the office. Sam rebuilds that one number in a pivot table. It matches.
• Step 3: From 60 sample answers, the AI proposes nine themes. Sam merges “Scheduling” and “Rota” into one theme, “Shift rota published too late”, and splits “Tools” into “Scanners keep failing” and “Too many logins”.
• Step 4: All 112 answers are tagged. The table has 112 rows. Sam counts the themes with a pivot table.
• Step 5: Sam tags 20 answers by hand and disagrees with the AI on three, all of them short sarcastic replies. The themes stay; those three get fixed by hand.
• Step 6: The summary leads with the rota finding, links it to the low information score in the warehouse, and quotes two real answers pulled from the original file by ID.

Sam’s report is not longer than last quarter’s. It is just traceable. When the warehouse manager asks “where does that come from?”, Sam can show the rows.

What not to paste into a chatbot

• Names, emails, employee numbers, or anything else that identifies a respondent.
• Small-group answers that make someone recognisable, even without a name.
• Health, salary or disciplinary details, unless your company has approved a tool for that.
• Answers from a survey that promised anonymity, unless your AI setup matches that promise. If you told people “only HR sees this”, a personal chatbot account is the wrong place for it.

Common mistakes

• Asking for themes and counts in the same prompt. You get tidy numbers you cannot check. Theme list first, tagging second, counting in the sheet.
• Trusting the first theme list. It is a draft. Your edits are what make it fit your organisation.
• Reporting percentages for tiny groups without saying how many people are behind them.
• Letting the AI write the quotes. Ask for IDs and copy the original words.
• Skipping the “Other” pile. The surprising answer that nobody planned a theme for often lands there.
• Jumping to recommendations. “The rota is published too late” is a finding. “Publish it two weeks ahead” is a decision someone with context should make.

FAQ

Can ChatGPT analyze survey results?

Yes. Upload the responses as a CSV or Excel file and it can calculate counts, averages and breakdowns, make charts, and sort open answers into themes. OpenAI asks you to review the method and the assumptions it used, which is why this guide has you check three numbers and a sample of tags yourself.

Which AI is best for survey analysis?

For most work surveys, the one you already have access to through work. ChatGPT and Claude both take uploaded spreadsheets and can run real calculations. Gemini is built into Google Forms and Copilot into Microsoft Forms, which is handy for a first look. The checks in Steps 4 and 5 matter more than the tool. If you are choosing between them, our ChatGPT tutorial for beginners and Claude tutorial for beginners show how each one works.

How do I analyze open-ended survey responses in Excel with AI?

Export the answers with a response ID column, let the AI propose and then tag themes (Steps 3 and 4), paste the tag table back next to your data in Excel, and count the themes with a pivot table or COUNTIF. Excel does the counting, the AI does the reading.

How many responses can AI handle?

An uploaded file holds far more than you could paste into a chat, and the exact upload limits are on each tool’s help pages. The practical limit is checking: with long answers or thousands of rows, work in batches and check the row count of every batch. Gemini in Google Forms generates its theme view for up to 200 responses.

Is an AI analysis accurate enough to present to leadership?

The numbers are, once you have checked them in the sheet. The themes are, once you have edited the list and tested a sample against your own reading. Say in your report that AI helped sort the open answers and that you checked it. Pew discloses its AI use in the methodology section of its reports, which is a fair standard to follow.

Can AI write the survey too?

Yes, and a well-built survey is much easier to analyze. See how to use AI to create a survey for the brief, the prompt and the test run.

Sources

• OpenAI Help Center: Data analysis with ChatGPT
• Claude Help Center: Upload files to Claude
• Claude Help Center: Create and edit files with Claude
• Google Docs Editors Help: View and manage form responses
• Google Docs Editors Help: Summarize and analyze responses with Gemini in Google Forms
• Microsoft Support: Welcome to Copilot in Forms
• Pew Research Center: How Pew Research Center is and is not using AI in our work (Sept 2026)
• Pew Research Center: Methodology, bogus respondents report (Aug 2026)
• NORC, General Social Survey Methodological Report 150: Leveraging Large Language Models for open-ended responses (July 2026)

Still designing the questions you will analyse later? If you need a scored knowledge check rather than open opinions, start with how to make a quiz with AI before you collect the answers.

Your next 20 minutes

Open your last survey export, delete the names and emails, add a response ID column and run the Step 1 prompt. Then take one open question and build the theme list from Step 3. In 20 minutes you will know what your people are really saying in that comment box, and you will be able to show where every number comes from.

Want to see what else is in the spreadsheet you have been avoiding? Our guide to analyzing a spreadsheet with ChatGPT uses the same checks, and using AI for market research shows how to put survey findings next to other sources. For more practical AI workflows you can use at work this week, start with the free Launchpad on our home page.

Your new hire starts on Monday. The laptop is ordered, maybe. Someone said they would set up the accounts. And the “plan” for their first weeks is a vague idea in your head plus a folder of documents nobody has opened since last year.

Here is how to use AI for employee onboarding without making it feel robotic: spend ten minutes writing down what the new person must be able to do at day 30, 60 and 90, then let ChatGPT, Claude or Gemini turn those facts into a week-by-week plan, a first-week schedule, a welcome email, a new-hire FAQ from your own documents and the questions for each check-in. You check every line against reality. The AI does the drafting. The welcome, the buddy and the conversations stay human.

Done this way, a full onboarding pack takes about an hour instead of a lost afternoon. Below are the six steps, the prompts to copy, a worked example and the things you should never paste into a chatbot.

Why most onboarding falls flat (and where AI actually helps)

If your onboarding feels improvised, you are in good company. Gallup finds that only 12% of employees strongly agree their organization does a great job onboarding new people (Gallup, Why the Onboarding Experience Is Key for Retention).

The reason is rarely bad intentions. It is time. The manager is busy, HR is busy, and the useful knowledge (“who do I ask about invoices”, “what does good look like in month one”) lives in people’s heads. So the new hire gets a login, a handbook and a lot of “just shout if you need anything”.

The SHRM Foundation’s guide on onboarding, written by Talya Bauer, lists what works: do the basics before day one, make the first day special, write an onboarding plan down, involve the people around the new hire, and check in at milestones such as 30, 60 and 90 days. It also notes that new hires get roughly 90 days to prove themselves (SHRM Foundation, Onboarding New Employees: Maximizing Success).

Look at that list again. Most of it is writing: a plan, a schedule, a welcome message, an FAQ, check-in questions. That is exactly the part AI is good at. What it cannot do is the part that matters most to the new person: a manager who makes time and a colleague who answers the “stupid” questions.

• The AI’s job: turn your facts into a clear plan, schedules, emails, an FAQ and check-in questions, and spot gaps in what you wrote.
• Your job: decide what success looks like, check every detail, and show up for the conversations.

Step 1: Write the manager’s brief first (10 minutes)

This is the step everyone wants to skip, and it is the one that makes the rest work. A chatbot that only knows “marketing coordinator” will write a plan for every marketing coordinator on earth. A chatbot that knows what your manager expects writes a plan for this person.

Answer these questions in a plain note, or send them to the hiring manager if that is not you:

1. What must this person be able to do on their own after 30 days? After 60? After 90?
2. Which three or four people (roles are enough) do they need a good working relationship with first, and why?
3. Which tools and accounts do they need, and which ones take longest to learn?
4. What do new people in this role usually get wrong in the first months?
5. Who is their onboarding buddy?
6. Remote, hybrid or on site? Which days are they in the office?

Write the 30, 60 and 90 day answers as things someone could observe. “Understands our processes” is not checkable. “Sends the weekly client report on Friday without help” is.

Step 2: Turn the brief into a 30-60-90 day plan (10 minutes)

Open a new chat in ChatGPT, Claude or Gemini, paste your brief and use this prompt:

You are helping me onboard a new employee. Below is the manager's brief.

Role: [job title]
Team and company: [one line on what the team and company do]
Work set-up: [remote / hybrid / on site]
Brief: [paste your answers from Step 1]

Write a 90-day onboarding plan, week by week, as a table with these columns:
week | focus | tasks | people to meet | how we know this week went well

Rules:
- Use only the facts in the brief. If something is missing, put [ASK] in the cell instead of guessing.
- Every "how we know" cell must be something a manager can observe, not "understands X".
- Weeks 1 to 4 focus on learning and relationships, weeks 5 to 8 on doing real work with support, weeks 9 to 13 on working independently.
- Plain language, no jargon. The new hire will read this on day one.

The [ASK] rule is the important one. AI fills gaps with confident, plausible details: a system you do not use, a meeting that does not exist. Making it mark gaps instead of filling them saves you from handing a new person a plan full of fiction.

Read the table week by week. Fix anything that is wrong, answer the [ASK] cells, and ask a follow-up if a week looks overloaded: “Week 2 has nine tasks. Cut it to the five that matter most and move the rest.”

Step 3: Build the first-week schedule and day-one checklist (5 minutes)

In the same chat, ask for the practical part. This is where onboarding usually breaks: no account on day one, no one free to say hello, a calendar that is empty or packed wall to wall.

Using the plan above, write:

1. A checklist of everything that must be ready BEFORE day one (accounts, hardware, access, calendar invites), with a column for who owns each item.
2. A day-by-day schedule for week one. Max 3 meetings per day, each with a length and a one-line purpose. Leave real time for setting up and reading.
3. A short list of things the new hire can read or try on their own when they have a quiet hour.

Mark anything you are not sure about with [ASK].

Two things are worth putting in that week on purpose. First, a one-on-one with the manager early in week one. Microsoft’s research found that new hires were 3.5 times more likely to be satisfied with their onboarding when their manager played an active role (Microsoft WorkLab, Strategies for Onboarding in a Hybrid World). Second, a first meeting with the buddy, which leads to the next point.

Step 4: Write the welcome email and intro messages (5 minutes)

A good welcome email before day one answers the questions people lie awake with: where do I go, what time, who meets me, what do I bring, is there lunch. AI writes these well, as long as you give it the real answers.

Write a welcome email to [first name], who starts as [role] on [date].

Facts (use only these, do not add anything):
- Start time and place: [address or "online, link follows"]
- Who meets them: [name, role]
- Their buddy: [name, role]
- What to bring: [list]
- First day in short: [2-3 lines from the schedule]

Tone: warm and human, like a colleague, not HR. Max 150 words.
Then write a 3-line message I can post in our team channel to introduce [first name], and a short note to the buddy explaining what I am asking of them in the first month.

Read the welcome email line by line before you send it. If the AI added “free parking” or “access to our learning portal” and neither exists, that is the first impression you make. The “use only these” instruction cuts this down a lot, but it does not replace your read.

About that buddy: it is worth the effort. Microsoft found that 56% of new hires who met their onboarding buddy at least once in the first 90 days said it helped them get productive quickly, rising to 97% for those who met more than eight times (same Microsoft WorkLab article). So put buddy meetings in the plan, every week or two, not just once.

Step 5: Turn your documents into a new-hire FAQ (15 minutes)

Every team has the handbook, the process docs and that one long email thread that explains how things really work. New people do not read them. They ask the same twenty questions instead.

Upload or paste the documents (with the personal data removed, see below) and ask:

These are our internal documents for new [role]s. Read them as if you started here today.

1. Write the 20 questions a new person is most likely to ask in their first two weeks, and answer each one using only these documents. Quote the document name for each answer.
2. List questions the documents do NOT answer, or where two documents say different things.
3. List anything that looks outdated (old tool names, people, dates).

Parts 2 and 3 are the hidden win. You find out where your documentation has holes before the new person falls into them. If a process is missing completely, write it down properly first. Our guide on using AI to write SOPs shows how to turn “ask Sanne, she knows” into a document anyone can follow.

Claude and ChatGPT both handle long documents well. If you work mostly in Claude, our Claude tutorial for beginners covers uploading files and Projects, so you can keep all onboarding documents in one place and reuse them for the next hire.

Step 6: Prepare the 30, 60 and 90 day check-ins (5 minutes)

The check-ins are where you catch problems while they are still small. SHRM recommends structured conversations at 30, 60 and 90 days, with questions like “who do you talk to when you have questions about your work?” and “what could we have done differently in setting your expectations?” (SHRM, Effective Onboarding Should Last for Months).

Based on the 90-day plan above, write a 30-minute agenda for the [30 / 60 / 90] day check-in.

Include:
- 2 questions about the milestones for this period (from the plan)
- 3 open questions about how they are settling in: people, workload, clarity
- 1 question about what we should change in onboarding for the next person
- a short space for agreeing the goals for the next 30 days

Keep the questions open and friendly. This is a conversation, not an exam.

After the conversation, you can paste your own notes back in and ask for a short summary with agreed actions. For the longer formal review later, see how to use AI to write a performance review. For any meeting in the plan, AI meeting agendas work the same way.

A worked example (fictional, to show the flow)

Say you run a 12-person agency and Lina starts as account coordinator. Your brief says: after 30 days she runs the Monday client status email for two accounts with a review; after 60 days she runs it for all five accounts alone; after 90 days she prepares the monthly client report.

The first draft plan from ChatGPT looked tidy, but it scheduled “CRM training” in week one. You do not use a CRM. You use a shared spreadsheet. Because of the [ASK] rule, the tool also flagged that it did not know who approves her client emails in the first month. You filled that in (you, until week four), replaced the CRM line and cut week two from eight tasks to five.

The FAQ step was the eye-opener. Out of 20 questions, the AI could not answer four from your documents, including “what do I do if a client emails me directly about billing?” Nobody had ever written that down. Ten minutes later it was a short paragraph in the handbook, and Lina never had to ask.

Total time: about an hour, spread over two days. Lina got a plan on day one that said what was expected of her, in her words, with names next to the tasks.

Can you automate employee onboarding with AI?

Partly, and the boring parts are the ones worth automating. Once your plan and emails exist, a tool like Zapier, Make or n8n can send the welcome email on a set date, create the checklist tasks in your project tool, and remind the buddy and manager before each check-in. Our guide to Zapier AI automation shows how to set up flows like that, and automating client onboarding with Make.com uses the same pattern for clients that you can copy for staff.

Start manual for the first hire. Automate on the second or third, once you know which steps you repeat every time. Automating a messy process just gives you a fast messy process.

What not to paste into a chatbot

Onboarding touches personal data, so be careful here:

• Leave out: salary, contract details, home address, ID or bank numbers, health information, and anything from the hiring process about the candidate. The AI does not need any of it to write a good plan. A first name and a role are enough.
• Check your account type: on ChatGPT Free, Plus and Pro, your conversations can be used to train models unless you switch off “Improve the model for everyone” under Settings, Data controls. On ChatGPT Business and Enterprise, OpenAI says it does not train on your inputs by default (OpenAI Help, How your data is used).
• Follow your company’s AI policy. If there is one, it wins. If there is none, internal process documents are usually fine to use once personal details are removed, but ask before uploading anything marked confidential.

Common mistakes

• Skipping the brief. Without it you get a generic plan that could be for any company, and the new hire notices.
• Sending AI text unread. One invented benefit or wrong address in a welcome email costs trust before day one.
• Filling every hour. A week one with eight meetings a day leaves no time to set up, read and think. Three is plenty.
• Letting the plan replace the people. A perfect document does not answer the question “is it OK to ask this?” A buddy and a manager do.
• Making a new pack for every hire. Save the brief, prompts and FAQ. Next time you only change what is different for the role.

Questions people ask about AI for employee onboarding

Which AI tool is best for onboarding?

For writing plans, emails and FAQs, ChatGPT, Claude and Gemini all do a good job. Use the one your company already pays for or allows, because the data rules matter more here than small quality differences. If you are new to it, start with our ChatGPT tutorial for beginners.

Will new hires feel it is impersonal?

Only if you let the AI do the human parts. A clear plan written with AI help, plus a manager who meets them in week one and a buddy who checks in, feels more personal than an improvised first week. Do not use AI to write the things that should come from you, like a personal note on their first day.

Can I use this for remote employees?

Yes, and it matters even more there, because nobody overhears answers at the next desk. Put the buddy meetings and the manager’s one-on-ones in the calendar from day one, and make the FAQ easy to find.

How long should onboarding last?

Longer than a first day or week. The SHRM Foundation guide suggests check-in milestones at 30, 60, 90 and 120 days and up to a year (SHRM Foundation). The 90-day plan in this guide covers the most intense part.

I am the new hire. Can I use this too?

Yes. Ask your manager the Step 1 questions in your first week, then use the Step 2 prompt to draft your own 30-60-90 day plan and bring it to your first one-on-one. Most managers are glad when someone does that.

Want a quick knowledge check in week one? A short quiz built from your own handbook beats a long FAQ nobody opens. Here is how to make a quiz with AI from the same documents you already gathered for the new-hire FAQ.

Your next 20 minutes

Before your next hire starts, write the Step 1 brief, paste it into a new chat with the Step 2 prompt, and fix whatever the [ASK] cells turn up. That plan alone puts you ahead of most first weeks people experience.

Still hiring? Our guides on using AI to write a job description and creating interview questions with AI cover the steps before day one. And for more practical AI workflows you can use at work this week, start with the free Launchpad on our home page.

The self evaluation form is open in another tab. The box says “Describe your key accomplishments this year”, the deadline is Friday, and your mind is blank. You know you worked hard. You just can’t remember what you did in March, and everything you type sounds either like bragging or like an apology.

Here is how to use AI to write a self evaluation that actually helps you: first collect your evidence (emails, tickets, calendar, thank-you messages), then let ChatGPT, Claude or Gemini interview you about it one question at a time, have it sort your answers into the questions on your form, and only then ask for a draft that uses nothing but your facts. Anything you can’t back up gets marked so you fill it in or cut it. The AI does the remembering prompts, the structure and the wording. The facts and the final say stay yours.

Plan on about an hour: 20 minutes of digging, 20 minutes of being interviewed, 20 minutes of editing. Below are the five steps, the prompts to copy, a worked example, and the mistakes that make a manager stop reading.

Why “write my self evaluation” is the wrong first prompt

Type “write a self evaluation for a marketing manager” and you get a page that could belong to anyone: “consistently exceeded expectations”, “strong team player”, “drove impactful results”. Your manager has read that paragraph forty times. It tells them nothing, and it quietly suggests you had nothing specific to say.

The AI isn’t the problem. The input is. A chatbot only knows what you give it, so a vague request gets a vague answer, and when it has no numbers it may simply make some up. HR teams that coach employees through reviews say the same thing in plainer words. The University of California, Santa Barbara’s guidelines tell staff to be specific: not “supported the student records project”, but what exactly you did and how it supported the goal (UCSB HR, Guidelines for Writing Your Employee Self-Assessment).

So the order matters. Evidence first, questions second, draft last.

Step 1: Dig up your evidence (20 minutes)

You did more than you remember. Almost everyone does. Software engineer Julia Evans describes the review-season feeling of “wait, what did I do in the last 6 months?” and the fix she uses: go through your tickets, emails and documents and write it all down in one list, which she calls a brag document (Julia Evans, Get your work recognized: write a brag document).

You don’t need a perfect list. You need raw material. Open a plain document and spend 20 minutes pasting in short notes from:

• Your calendar. Scroll back month by month. Recurring meetings show what you owned; one-off meetings often mark a project or a problem you solved.
• Sent email and chat. Search for words like “launched”, “done”, “fixed”, “thanks” and “great job”.
• Your task tool. Closed tickets, finished cards, shipped items.
• Last year’s goals. The ones on the form, word for word.
• Feedback you received. Thank-you notes from clients or colleagues, comments from your manager.
• The fuzzy work. Training a new colleague, cleaning up a messy process, being the person everyone asks about the CRM. This is the work most people forget to mention.

UC Davis HR gives employees nearly the same list: accomplishments, letters of thanks, completed goals, training given or received, and written customer feedback (UC Davis HR, Writing a Self-Evaluation & Goals). Don’t polish anything yet. “Q2, took over supplier invoices when Sam left, cleared backlog” is perfect.

Before you paste anything into a chatbot, swap client names, colleague names and confidential numbers for placeholders like [Client A] or [teammate]. More on that further down.

Step 2: Let the AI interview you (20 minutes)

This is where AI earns its place. Instead of staring at the form, you answer short questions, one at a time. It is much easier to talk about your work when someone asks “and what changed after that?” than to write about it cold.

I am writing my self evaluation as a [job title] for [review period]. Do not write anything yet.

Here are my raw notes: [paste your notes from Step 1]
Here are the questions on my form: [paste the form questions]
Here were my goals for this period: [paste goals]

Act as a friendly coach. Ask me up to 8 short questions, one at a time, and wait for my answer before asking the next. Ask about what I did myself, what changed because of it, any number or before-and-after I can give, what went wrong, and what I learned.
If I say I don't know, write UNKNOWN. Never guess a number, a result or a compliment.

Answer like you would talk to a colleague over coffee. Short is fine. If you don’t know a number, say so. You can often find it later in a report or a dashboard.

When the questions are done, ask for a check before any writing:

Summarise what I told you as a list of facts, grouped by the questions on my form. Use my words where you can. Put UNKNOWN next to anything I did not answer. Do not add any result, number or praise I did not give you. Do not draft yet.

Read the list line by line. If it says “reduced processing time by 40%” and you never said 40%, correct it now. This one check is what keeps invented facts out of your review.

Step 3: Match your facts to the form (5 minutes)

Most forms ask a handful of fixed things: accomplishments, goals, competencies, challenges, development. UCSB’s guidelines suggest that for each work priority you say what the task was, how success was measured, what impact it had, and what obstacles you had to overcome (UCSB HR). Ask the AI to sort your facts the same way:

Using only the approved list, map my facts to each form question. For every goal from last period, tell me: met, partly met, or not met, and which fact supports that. Flag any form question that has no fact behind it yet.

A flagged question is useful information. Either you go find the evidence, or you write a short, honest answer instead of padding.

Step 4: Draft it in your own voice (10 minutes)

Now you ask for words. Give the AI a sample of how you actually write, a few sentences from a recent email, so the draft sounds like you and not like a press release.

Draft my answer to each form question using only the approved facts. First person. Lead with the result, then what I did, then how. Maximum [word or character limit from the form] per answer.

Here is how I write: [paste 3 or 4 sentences from a recent email].

Rules: no buzzwords like "spearheaded", "leveraged" or "synergy". Do not turn team results into my personal results; say "I" for what I did and "we" for what the team did. Where a fact is UNKNOWN, write [ADD DETAIL] instead of filling the gap.

Check the character limit before you start. Many review systems limit the number of characters, and UC Davis HR recommends writing your self evaluation in a separate document first and pasting it in afterwards (UC Davis HR). That also protects your work if the form times out.

Step 5: The honesty pass (10 minutes)

Read the draft with one question per sentence: “Could I point to where this came from?” If not, it comes out or gets rewritten until you can. Then ask the AI to argue the other side:

Read my self evaluation as my manager would. Point out: (1) any sentence that sounds vague or generic, (2) any claim that sounds bigger than the evidence, (3) anything I undersold, where the fact is strong but my wording is timid. Quote each sentence and suggest a fix. Do not add new facts.

Point 3 matters more than it looks. Many people write their own reviews too modestly. Research by economists Christine Exley and Judd Kessler, with more than 4,000 participants from online labour markets plus over 10,000 school pupils, found that women described their own performance less favourably than men who had performed equally well, and the gap remained even when there was nothing to gain from self-promotion (Exley and Kessler, The Gender Gap in Self-Promotion, NBER). Whoever you are, if a result is real and you have the proof, say it plainly. Julia Evans puts it well: you don’t have to make your work sound better than it is, just exactly as good as it is.

A worked example: from notes to a finished answer

Fictional example, details made up for illustration. Maria is an office coordinator. Her form asks: “Describe your most important accomplishment this period.”

Her raw note from Step 1:

Spring: moved supplier invoices from email to shared folder + checklist. Fewer late payments? Finance thanked me in May.

During the Step 2 interview the AI asks: “What happened before the change, and how do you know it got better?” Maria checks and finds that late-payment reminders from suppliers went from about six a month to one or two. The AI also asks who else worked on it. She says she built the checklist herself and finance agreed the process.

The first draft from Step 4 said: “I spearheaded a complete overhaul of invoice processing that eliminated late payments.” The honesty pass in Step 5 flagged two problems: “spearheaded” is a buzzword, and “eliminated” isn’t true, because there were still one or two reminders a month.

Final answer, in her own voice:

In spring I moved supplier invoices out of individual inboxes into a shared folder with a simple checklist, agreed with finance. Late-payment reminders from suppliers dropped from about six a month to one or two. Finance can now see every open invoice without asking me, which helped when I was on leave in August.

Three sentences, a real before and after, credit where it belongs. That is the whole goal.

The hard parts: weaknesses, goals and the rating

Writing “areas for improvement” without sounding like a confession

Most people either skip this or bury themselves. UC Davis HR asks employees to discuss challenges, how they overcame them, and their plan for the next period. That three-part shape works well with AI:

Here is something that did not go well this period: [describe it honestly]. Write 2 or 3 calm sentences: what happened, what I learned, and one concrete thing I will do differently next period. Do not turn it into a hidden strength and do not add problems I did not mention.

Turning next year’s ambitions into real goals

Here is what I want to work on next year: [your rough ideas]. Turn each into one specific goal with a clear way to measure it and a date. Keep them realistic for someone in my role. Give me 3, not 10.

Then cut anything you wouldn’t actually be willing to be measured on next year.

Not sure where to aim next year? A personal SWOT can help you see your strengths and gaps on one page before you set goals. Here is how to do a SWOT analysis with AI, including a version for your own career.

Choosing your rating

If your form asks you to rate yourself, don’t let the AI pick the score. That is a judgement call for you and your manager. You can ask it to list the evidence for and against each level so you choose with your eyes open.

Is it OK to use AI for your self evaluation?

For most people, yes, as long as the facts are yours. But check two things first. Does your company have an AI policy that says which tools you may use for work documents? And is your review content considered confidential? If the policy says use the company’s own Copilot or ChatGPT workspace, use that one.

The deeper rule is simple: your self evaluation is a statement about your own work, signed by you. AI can help you remember, organise and phrase. It must not invent. If a sentence would embarrass you when your manager asks “tell me more about that”, it doesn’t belong in the document.

Your self evaluation also feeds straight into the next conversation. If you are preparing to ask for a raise based on this year, our guide on how to use AI to negotiate salary picks up where this page stops.

What not to paste into the chat

• Names of colleagues and clients. Use roles or placeholders instead.
• Confidential figures. Revenue, prices, contract terms. Use rounded or relative numbers (“about a third fewer”) in the chat and put the real figure in yourself afterwards.
• Feedback about other people. If you are writing peer feedback in the same session, keep it out.
• Your full HR file. The tool doesn’t need it.

Also look at your settings. On personal ChatGPT plans (Free, Plus, Pro), “Improve the model for everyone” is on by default; you can switch it off under Settings, then Data controls. Temporary chats are not used to train the models, though OpenAI says they may be kept for up to 30 days for safety purposes. By default, OpenAI does not train on content from Business and Enterprise workspaces (OpenAI Help, Data controls in ChatGPT).

ChatGPT, Claude or Gemini for a self evaluation?

All three handle this well, because the hard part is your input, not the model. A few practical differences:

• ChatGPT is good at the interview step, and you can do it by voice on your phone while walking, which makes the answers more natural.
• Claude works just as well for the text steps. Run Step 4 in two tools if you like and keep the draft that sounds more like you.
• Gemini is handy if your notes and documents already live in Google Docs and Gmail.
• Microsoft Copilot is the safe choice if your employer provides it and your review system runs on Microsoft 365.

Use the one you already have open all day. New to these tools? Start with our ChatGPT tutorial for beginners, the Claude tutorial for beginners or the Gemini tutorial for beginners.

Common mistakes (and the quick fix)

• Starting with the draft. Fix: evidence first, interview second, draft last.
• Accepting numbers you didn’t give. Fix: the summary check in Step 2, every time.
• Claiming the team’s result as yours. Fix: “I” for your part, “we” for the team’s. Managers notice, and colleagues read reviews too.
• Every answer sounds the same. Fix: paste a real sample of your writing and ban the buzzwords by name.
• Leaving it to Thursday night. Fix: next year, keep a running list. UC Davis HR suggests a simple performance journal in Outlook, OneNote or a notebook. Five minutes every two weeks makes this whole page a 20-minute job.

FAQ

Can ChatGPT write my self evaluation?

It can write the words, but it can’t know what you did. Give it your notes and let it interview you first, or you will get generic text that could describe anyone.

Will my manager know I used AI?

They will notice generic, buzzword-heavy text, whether a person or a tool wrote it. A review full of your own specific facts and your own phrasing reads like you, because it is you.

How long should a self evaluation be?

As long as the form allows and no longer than your evidence supports. UCSB’s guidelines note that a self-assessment doesn’t have to be long to be effective. Two to four specific sentences per question usually beat a full page of adjectives.

What if I had a bad year?

Be honest and specific about what happened and what you are changing. The AI can help you find calm wording, but don’t let it spin. A clear plan reads far better than excuses.

Can I use the same method for a manager’s review of me, or for reviewing my team?

The method for reviewing others is different, because you are judging someone else’s work and their privacy matters more. See our guide on how to use AI to write a performance review for that side of the table.

Your next hour

Open your calendar and scroll back to January. Spend 20 minutes pasting short notes into one document. Then open a chat, paste the Step 2 prompt, and let it interview you. By the end of the hour you will have a list of real things you did this year, and a draft that sounds like you on a good day.

The same rule, your facts and AI’s help with the wording, runs through our other career guides: using AI to write a resume and preparing for an interview with AI. And for more practical AI workflows you can use at work this week, start with the free Launchpad on our home page.

The offer email landed an hour ago. The number is fine. Not great, just fine. You want to ask for more, and your stomach already turns at the thought of typing the reply. What if they pull the offer? What if you sound greedy?

Here is how to use AI to negotiate salary without the dread: collect real pay data for the role yourself, then let ChatGPT, Claude or Gemini help you with the three parts that feel hardest. It turns your facts into a clear case, it plays the recruiter so you can rehearse the call out loud, and it helps you write a short, warm counteroffer email. The AI does the drafting and the practice. The number comes from real data, and the final words are yours.

That takes about an hour, usually in the day or two after the offer. Below are the five steps, the prompts to copy, a worked example, and the one thing you should never let a chatbot decide for you.

Why “what salary should I ask for?” is the wrong first prompt

It is the obvious question to type. It is also the one place where AI is weakest. A chatbot does not know what this company pays, what your city pays, or what the last three people in this role signed for. It produces a plausible-looking range, and plausible is not the same as true.

There is a second problem, and it is not a small one. Researchers who asked several large language models for a starting salary to negotiate found that the suggested numbers shifted with who the user said they were. Women were often advised to ask for substantially less than men applying for the same role (Sorokovikova et al., “Surface Fairness, Deep Bias”, 2025). A separate audit sent ChatGPT tens of thousands of salary prompts and concluded it was not consistent enough to be trusted for picking an opening offer: the answers moved with gender, university, major and even the model version (Geiger et al., PLOS One, 2025).

So the rule for this whole guide is simple. You bring the number. AI helps you argue for it and say it well.

Step 1: Get the real pay range yourself (20 minutes)

Negotiation experts at Harvard’s Program on Negotiation put research first: your counteroffer should rest on credible benchmarks, such as salary databases filtered by role, location and experience, industry reports, conversations with peers, and competing offers if you have them (PON, How to Negotiate a Higher Salary after a Job Offer).

Pull two or three of these:

• Official pay data. In the US, the Bureau of Labor Statistics Occupational Outlook Handbook lists median pay per occupation. Most countries have a national statistics office with similar tables.
• Salary sites with entries filtered by job title, city and years of experience. Note the range, not one figure.
• The job ad itself. Many ads now include a salary band. If this one did, that band is your strongest anchor.
• A person. One message to a former colleague or a recruiter you trust is often worth more than any website.

Write down a low, middle and high figure, and where each came from. That list is what you will give the AI. Harvard Business School Online suggests turning it into two numbers of your own: a baseline you will not go below, and a stretch goal informed by the higher end of the market data (HBS Online, 7 salary negotiation tips).

Step 2: Let AI build your case from your facts (10 minutes)

Now open a chat and give it everything it needs, minus anything private. A counteroffer lands better with a reason attached. PON quotes Harvard professor Deepak Malhotra on this: never let your proposal speak for itself, always tell the story that goes with it (PON, How to Counter a Job Offer). AI is good at finding that story in your own material.

I received a job offer and want to negotiate. Help me build my case.

The offer: [base salary, bonus, benefits, start date, title]
The job ad: [paste it]
My background: [paste your CV with contact details removed]
Market data I found myself: [low / middle / high, with the source of each]
My baseline: [number]. My stretch goal: [number].

Rules:
- Do not suggest a salary number of your own. Use only my market data.
- Use only facts from my CV and the job ad. If something would make my case stronger but you don't see it, ask me instead of inventing it.
- Give me the 3 strongest reasons this employer would agree to my stretch goal, each in one sentence.
- Then list 4 things besides base pay I could ask for if salary is fixed.

The second list matters more than people expect. PON’s advice is to negotiate the whole package, not one line: bonus, retirement contributions, remote or hybrid days, training budget, title, travel or relocation support (PON). If the recruiter says the salary band is fixed, you want a second ask ready, not a blank silence.

Step 3: Rehearse the call with AI as the recruiter (20 minutes)

This is where AI earns its place. Most of us know roughly what we want to say. We freeze when the other person says “that’s above our budget.” So practise that moment until it stops scaring you.

Play the recruiter for this role at [company]. You are friendly but under budget pressure.

I will open by asking for [stretch goal]. Respond the way a real recruiter might:
sometimes agree to check, sometimes say the budget is fixed, sometimes ask why I'm worth it, sometimes offer something other than salary.

Rules:
- One reply at a time, then wait for me.
- After I answer, step out of the role and give me one sentence of feedback: was I clear, did I give a reason, did I back down too fast?
- After 4 rounds, tell me which of my answers was weakest.

Do this out loud if you can. In the ChatGPT app you can use voice mode and talk it through like a phone call; Claude and Gemini have voice options in their mobile apps too. Typing a calm answer is easy. Saying “I was hoping for something closer to 72” without laughing nervously takes a few tries. That is exactly what the rehearsal is for.

Three lines worth practising until they feel normal:

• When the offer arrives: “Thank you, I’m really excited about this. Can I take a day to look at the full offer?” PON notes that accepting on the spot gives away your leverage, and that employers expect you to take a little time (PON).
• The ask: your number, then your reason, then stop talking.
• When the salary is fixed: “I understand. Is there flexibility on [your second ask from Step 2]?”

Step 4: Write the counteroffer email with AI (10 minutes)

Many negotiations now happen by email, and a clear written reply also gives the recruiter something to forward to the person who approves the budget. Let AI draft it, then make it sound like you.

Write a short counteroffer email to [recruiter's first name].

- Open with genuine thanks and say I want this role.
- Ask for [stretch goal] base salary.
- Give my 2 strongest reasons from our earlier chat, one sentence each.
- Mention [second ask] as an alternative if base salary is fixed.
- Close warmly and suggest a quick call.

Under 150 words. Friendly and confident, not apologetic, not pushy.
No phrases like "I hope this email finds you well". Do not add any facts I did not give you.

Then read the draft out loud once. Swap any word you would never actually say. If it still sounds stiff, our guide on making an email sound more professional with ChatGPT works in the other direction too: ask it to make the email sound more natural.

Step 5: Plan your answer to their reply

When the recruiter answers, paste their reply into the same chat and ask:

Here is the recruiter's reply: [paste].
1. In plain words, what did they actually offer or refuse?
2. Is it above or below my baseline of [number]?
3. Give me two short ways to respond: one that accepts, one that makes a final, polite ask.
Do not tell me which to choose.

That last line is deliberate. Whether to accept is your decision, and it depends on things the AI does not see: how much you want this job, your other options, your savings. Harvard’s career center lists using data, “and, optionally, AI tools” to support your ask, while also warning never to inflate a competing offer and never to keep interviewing after you accept (Harvard FAS Mignone Center, Evaluating and Negotiating Job Offers).

A worked example (fictional, numbers made up for illustration)

Sam is offered a project coordinator role at 62,000 with a small bonus. Sam’s own research, from a salary site, the national statistics office and a former colleague, says similar roles in the city sit between 60,000 and 70,000. Sam sets a baseline of 64,000 and a stretch goal of 69,000.

In Step 2, the AI points out that Sam’s CV shows two years running vendor schedules in the exact software named in the job ad, which means less ramp-up time. It also asks a question instead of guessing: “Did you lead any project with a budget? If so, how large?” Sam had, and adds the figure. That becomes reason number two.

In the Step 3 rehearsal, the “recruiter” says the band tops out at 65,000. Sam’s first answer is a rushed “oh, okay, that’s fine.” The feedback line: “You accepted before mentioning your second ask.” Round two, Sam asks for 65,000 plus a training budget for a project certification. That is the version that goes into the email.

Notice what the AI never did: it never suggested the 69,000. Sam brought that.

What not to paste into the chat

• The offer letter as a whole file if it contains your address, ID numbers or a confidentiality clause. Copy only the figures and terms you need.
• Anything an employer marked confidential, including internal salary bands someone shared with you privately.
• Your personal details. The AI does not need your name, age or background to help with an argument. Given the bias research above, there is a good reason to leave them out.

ChatGPT also keeps memories from earlier chats if you have memory switched on. For a negotiation, you can start a temporary chat and choose “Unpersonalized”, so it does not draw on memory or custom instructions and the chat stays out of your history (OpenAI may still keep a copy for up to 30 days for safety reasons).

ChatGPT, Claude or Gemini for salary negotiation?

Any of the three handles these prompts well. The steps above matter far more than the tool. A few practical differences:

• ChatGPT has a mature voice mode for the rehearsal and temporary chats for privacy. New to it? Start with our ChatGPT tutorial for beginners.
• Claude, in our experience, tends to follow “do not invent facts” style instructions closely and writes a calm, natural email. See our Claude tutorial for beginners.
• Gemini is handy if your offer and notes already live in Gmail and Google Docs.

Whichever you use, never ask any of them for “the market rate.” Give them yours.

Common mistakes (and the quick fix)

• Asking AI for the number. Fix: research it yourself in Step 1, then tell the AI not to suggest one.
• Sending the AI’s first draft. Fix: read it aloud and cut anything that does not sound like you.
• Asking “is this negotiable?” PON’s advice is to skip that question and make your counteroffer, because asking permission can close the door before it opens (PON).
• Only rehearsing the yes. Fix: in Step 3, tell the AI to refuse at least twice. The no is the moment you need practice for.
• Bluffing. Never invent a competing offer or let AI write one into your email. It is easy to check and hard to undo.

FAQ

Can ChatGPT help me negotiate my salary?

Yes, for the preparation: building your argument from your own facts, rehearsing the conversation and drafting the email. It should not pick your target number. Bring real pay data and tell it to work only from that.

What is a good ChatGPT prompt for salary negotiation?

The Step 2 prompt above. The parts that make it work are the rules: no salary number of its own, only facts you provide, and ask you rather than invent. Without those, you get a generic script with a made-up range.

Can AI write my salary negotiation email?

It can write a solid first draft in seconds. Keep it under 150 words, check every fact, and change the wording until it sounds like you on a good day.

Should I tell the employer I used AI?

There is no need to. Using a tool to prepare is like asking a friend to read your email. What matters is that every fact in your message is true and that you can say it yourself on a call.

What if they say the salary is not negotiable?

Thank them and ask about the rest of the package. PON suggests probing for flexibility elsewhere, such as your start date or whether your experience fits a more senior position (PON, When a Job Offer is “Nonnegotiable”). And some roles, such as certain government or consulting jobs, really do have fixed pay scales (Harvard FAS).

Your next hour

Write down your three market figures and where they came from. Paste them into the Step 2 prompt with your offer and CV. Then run four rounds of the recruiter rehearsal, out loud, until the ask comes out steady. You will go into that call knowing your number, your reason and your backup ask. That is most of the battle.

Still in the middle of interviews? Our guide on using AI to prepare for an interview uses the same “your facts, AI’s practice” approach, and so do the guides on writing a resume with AI and writing a cover letter with AI. For more practical AI workflows you can use at work this week, start with the free Launchpad on our home page.

Asking for a raise at your yearly review rather than a new job? Start with a strong self-assessment. Our guide on how to use AI to write a self evaluation helps you turn a year of work into the proof that backs your number.

The interview is on Thursday. You have read the job ad twice, you know you should “prepare”, and you still have no idea what they will actually ask or how you will answer without rambling. Most people end up skimming a list of “top 50 interview questions” the night before and hoping for the best.

Here is how to use AI to prepare for an interview properly: give ChatGPT, Claude or Gemini the job ad and your CV, let it predict what the interviewer will test, build short stories from your own real work, then have it play the interviewer, one question at a time, while you answer out loud. After each answer it gives you one piece of feedback, and you redo the weakest answer. The AI does the questions and the coaching. The stories and the words stay yours.

That takes about an hour spread over two evenings. Below are the five steps, the prompts to copy, a worked example, and where the line is between smart preparation and something an employer would not accept.

Why “give me interview questions” is not preparation

Ask any chatbot for “interview questions for a marketing manager” and you get the same generic list everyone else gets. Tell me about yourself. What is your biggest weakness. Where do you see yourself in five years. Useful to glance at, but it does not tell you what this employer cares about, and it does nothing for the part that actually goes wrong in interviews: saying a clear, specific answer out loud, under a bit of pressure.

Career services that coach candidates every day land on the same approach. MIT’s career office suggests feeding the AI your resume and the job description so it can match your experience to what the employer is looking for, and then practising answers with voice so you get feedback on structure (MIT CAPD, Using AI for interview prep). Harvard’s career center adds the warning that matters most: do not memorise answers, keep them in your own voice, and double-check anything factual the AI tells you (Harvard FAS Mignone Center, AI for Interviews and Offers).

So the job splits in two:

• The AI’s job: read the job ad like a hiring manager, predict the questions, play the interviewer, and point out what is missing from your answers.
• Your job: bring real examples from your work, say the answers out loud, and decide what sounds like you.

Step 1: Let AI decode the job ad (5 minutes)

Paste the full job ad and your CV into a new chat. Take your phone number, home address and anything confidential out of the CV first. The AI does not need them.

I have an interview for the job below. Act as the hiring manager for this role.

1. List the 5 to 7 skills or qualities you would most want to test in the interview, based only on the job ad. For each, quote the line in the ad that tells you it matters.
2. Compare them with my CV. For each skill, tell me which part of my experience is my best evidence, and where my evidence is thin.
3. Do not write answers for me. Give me talking points only.

JOB AD:
[paste]

MY CV:
[paste]

The “quote the line” part keeps the AI honest. If it claims the role needs stakeholder management, you can see exactly where the ad says so. The “where my evidence is thin” part is gold: those are the questions most likely to catch you out, so you prepare them first.

Step 2: Turn your real work into short stories (20 minutes)

Many of the interview questions that matter are some version of “tell me about a time when…”. The standard way to answer them is STAR: the Situation, your Task, the Action you took, and the Result. Harvard’s career center recommends typing your answers in STAR format and asking the AI for feedback on them.

The trap is letting the AI invent the story. It will happily write a perfect STAR answer about a project you never ran, and the first follow-up question will expose it. So flip it around: you talk, the AI asks.

I want to prepare a story for the skill "[skill from Step 1]". I will describe something that really happened at my work. Before you do anything else, ask me up to 5 short questions to get the details you need: what the situation was, what exactly I did, and what changed afterwards.

Then put it in STAR order as bullet points (not a script). Use only facts I gave you. If a part is missing, such as the result, say so and ask me for it. Do not make up numbers.

Speaking instead of typing helps here. MIT notes that speech to text can feel more natural and lets you be more descriptive, which gives the AI better material to work with. Most chat apps have a microphone button in the message box.

Aim for five or six stories that each cover two or three skills from Step 1. A story about fixing a messy handover, for example, can answer questions about problem solving, working with other teams and handling pressure.

Step 3: Cut the list down to a realistic interview (5 minutes)

Ask for questions and you will get thirty. Real interviews are much shorter. MIT makes the same point: a half-hour interview covers far fewer questions than you might expect, so ask the AI to narrow the list down to a realistic set.

This is a [length, e.g. 45-minute] [type, e.g. first-round video] interview with [who, e.g. the team lead]. Based on the job ad, give me the 8 questions I am most likely to get, in the order an interviewer would probably ask them. Mark which of my stories fits each question, and flag any question I have no story for yet.

Any question flagged “no story yet” goes back to Step 2. If a question looks off for the role, say so and ask the AI to refocus. You know the job better than it does.

Step 4: Run a mock interview out loud (20 minutes)

This is the step that makes the difference, and the one people skip. Reading a good answer and saying a good answer are different skills. Use a voice conversation if you can:

• ChatGPT Voice works in the app and on chatgpt.com: select the Voice icon in the message box and start talking. OpenAI’s help page says you can ask it at the start to wait until you ask it to respond, which is handy when you need a moment to think, and the transcript is added to the chat afterwards so you can read your answers back (OpenAI Help, ChatGPT Voice).
• Gemini Live gives you a free-flowing spoken conversation in the Gemini app on your phone. Google notes it is not available in the Gemini web app (Gemini Apps Help, Gemini Live).
• No voice option? Type the prompt, then say each answer out loud before you type it in. It is slower, but you still practise the speaking part.

Start the session with this prompt (paste or read it out):

You are the interviewer for the [job title] role at [company], using the 8 questions from earlier. Ask me one question at a time and wait for my full answer.

After each answer, give me short feedback in three lines: what worked, the one change that would improve it most, and whether any part of STAR is missing. Do not rewrite my answer for me. Then ask one realistic follow-up question, the kind a curious interviewer would ask, before moving on.

After all 8 questions, tell me the patterns you noticed across my answers and which answer was weakest.

The follow-up question is the point. Anyone can rehearse a polished opening. Follow-ups are where you show the story is real.

Step 5: Redo the weakest answer, then stop

Take the weakest answer from the summary and do it again, out loud, using the one change the AI suggested. Then leave it. Coming back the next day for a second short round works better than drilling the same answer twenty times in one evening.

Watch for one thing: if your answers start sounding like the AI’s phrasing, you have gone too far. MIT calls memorising answers word for word the biggest trap, because you sound robotic and cannot adapt when the question comes in slightly differently. You want to know your stories and your key points, not a script.

Research the company with AI (and check every fact)

AI is a fast way to get oriented: what the company does, who its customers are, what is going on in its industry. It is also a fast way to walk into an interview with a confident, wrong fact. Harvard’s career center says it plainly: generative AI can produce incorrect information, so double-check your research with trusted sources.

I have an interview at [company] for [role]. Give me a one-page briefing: what they sell and to whom, how they describe themselves, and 3 things that are probably on the team's mind right now. For every fact, tell me where I should check it (their website, annual report, recent press release). Mark anything you are unsure about.

Then open the company’s own site and recent news yourself and check each point. The University of Manchester’s careers service goes further: the organisation’s own advice about its interview process should be your primary source, with AI as a helper (University of Manchester Careers, AI interview prep). If you want a deeper method for this kind of research, our guide on using AI for market research shows how to keep the sources attached to every claim.

Prepare your own questions for them

“Do you have any questions for us?” comes at the end of almost every interview, and “no, I think you covered everything” wastes it. Ask the AI for help, but feed it your actual concerns:

Here is what I want to know before I would accept this job: [e.g. how success is measured in the first 6 months, how the team works with sales, why the role is open]. Turn these into 4 short, friendly questions I can ask the interviewer. No flattering questions, nothing I could find on their website in two minutes.

A worked example: from job ad to mock round

Here is how it played out for a made-up candidate, Lena, applying for an operations coordinator role at a mid-sized logistics company. The details are fictional; the process is the one above.

Step 1 output (shortened): the AI picked out “coordinating between warehouse and customer service”, “keeping shipment data accurate” and “handling urgent escalations”, each with the line from the ad. It flagged her evidence for data accuracy as thin, because her CV only said “maintained spreadsheets”.

Step 2: Lena described, by voice, the week a supplier renumbered its products without warning. The AI asked what she noticed first, what she changed, and what happened after. Her bullets:

S: Supplier switched its product numbers overnight; our orders started showing the wrong items.
T: Find the mismatches before the next morning's shipments went out.
A: Pulled the order list, matched old and new numbers in a lookup sheet, asked the supplier to confirm the 12 I wasn't sure about, then briefed customer service on what to tell affected customers.
R: Next morning's shipments went out correct. We added a check for supplier numbering changes to the weekly routine.
Missing: how many orders were affected? (Lena to check her notes.)

Notice the last line. The AI did not invent a number; it told her to find one.

Step 4: in the mock round, her first answer to “tell me about a time you caught a mistake” took nearly three minutes and buried the result at the end. The feedback: lead with one sentence of situation, get to the action faster, and say the result clearly. The follow-up question was “what would you do differently?”, which she had not thought about. That became the answer she redid in Step 5.

Is it OK to use AI to prepare for an interview?

For preparation, yes. Even AI companies say so. Anthropic’s guidance for its own job candidates encourages using Claude to research the company, practise answers and prepare questions for the interviewers. The same page draws two clear lines: no AI help during live interviews unless they say otherwise, and no using AI to create experiences you have not had (Anthropic, Guidance on Candidates’ AI Usage).

That is a sensible rule for any employer:

• Fine: predicting questions, shaping your real stories, mock interviews, company research, preparing your questions.
• Not fine: stories that did not happen, numbers you cannot back up, or an AI feeding you answers during the real interview or a test, unless the employer explicitly allows it.
• When unsure: check the employer’s careers page or ask the recruiter. Some companies publish their own rules.

What not to paste into the chat

Interview prep pulls in a lot of personal and work detail, so a quick check before you paste:

• Your contact details and ID numbers: leave them out of the CV you paste. The AI only needs your experience.
• Your current employer’s confidential information: client names, internal figures, unreleased plans. Describe it in general terms (“a large retail client”, “about a third of our orders”) instead.
• Voice recordings: know what is kept. OpenAI’s help page says audio clips from its real-time voice conversations are stored with the chat transcript and kept for 30 days, and that deleting the chat also deletes the clips. Check the settings of whichever tool you use.

ChatGPT, Claude or Gemini for interview prep?

All three handle Steps 1 to 3 and the company research well, as long as you give them the job ad and your real material. The difference is mostly in the mock interview:

• ChatGPT: a voice option on your phone and in the browser, with a transcript you can review afterwards.
• Gemini: Gemini Live is a good spoken sparring partner on your phone, and it suits you if you already live in Google’s apps.
• Claude: very good at the typed parts: decoding the job ad, shaping stories and giving honest, detailed feedback on a written answer.

Use the one you already pay for or have open all day. If you are new to these tools, start with our ChatGPT tutorial for beginners or the Claude tutorial for beginners.

Common mistakes (and the quick fix)

• Asking for “common interview questions”: you get the generic list. Fix: always start from the job ad (Step 1).
• Letting the AI write your stories: they fall apart at the first follow-up. Fix: you talk, it asks questions, it only orders your facts.
• Only typing: you practise writing, not speaking. Fix: answer out loud, even if the AI only reads text.
• Asking for praise: “was that good?” gets you “great answer!”. Fix: ask for the one change that would improve it most.
• Memorising the AI’s wording: you sound rehearsed and get thrown by a reworded question. Fix: learn your bullet points, not sentences.
• Trusting the company briefing: one wrong fact in the interview costs more than no fact. Fix: check every point on the company’s own pages.

FAQ

Can ChatGPT do a mock interview?

Yes. Give it the job ad, tell it to act as the interviewer, ask one question at a time and wait for your answer, and give short feedback after each one. The prompt in Step 4 does exactly that. Using ChatGPT Voice makes it feel much closer to a real interview than typing.

How long before the interview should I start?

Two or three days is comfortable: Steps 1 to 3 on the first evening, a mock round on the second, and a short redo of your weakest answer the day before. If the interview is tomorrow, do Step 1, prepare your three strongest stories and run one mock round.

Does this work for technical interviews?

For the conversational parts, yes: explaining past projects, walking through a decision, talking about trade-offs. For a coding or practical test, check the employer’s rules first. Anthropic, for example, asks candidates to do take-home assessments without AI unless it says otherwise.

Can AI replace a mock interview with a real person?

Not fully. MIT points out that AI can help with the structure and content of your answers but not with fine-tuning your delivery, and suggests pairing AI practice with a real-life mock interview. Use AI for the repetitions, and a friend, mentor or your career service for one final run-through.

What if the AI’s feedback seems wrong?

Push back. Tell it what the role really involves or why the answer works for this employer, and ask it to reassess. The AI is a sparring partner, not a judge.

Need a longer talk after the interview circuit, a meetup pitch or an all-hands update? The same “your facts, AI’s structure” rule applies in how to write a speech with AI.

Your next 30 minutes

Open a new chat, paste the job ad and your CV (contact details removed), and run the Step 1 prompt. Pick the skill where your evidence is thinnest and build that story with Step 2. You will walk out of those 30 minutes knowing your weakest question, and already having a real answer for it.

Still polishing the application itself? Our guides on using AI to write a resume, writing a cover letter with AI and writing your LinkedIn profile with AI use the same rule: your facts, AI’s help with the wording. Sitting on the other side of the table? See how to use AI to create interview questions. And for more practical AI workflows you can use at work this week, start with the free Launchpad on our home page.

Did the interview go well and an offer is on the table? Our guide on how to use AI to negotiate salary helps you research a fair range, plan your counter and rehearse the conversation before you reply.

You found a job that fits. Then you open your resume and it reads like it was written for the last job you applied to, three years ago. You paste it into ChatGPT, type “make this better”, and get back a page full of “results-driven professional” and “spearheaded cross-functional synergies”. It sounds impressive. It also sounds like everybody else, and half of it isn’t even true.

Here is how to use AI to write a resume that works: you supply the facts, the AI does the matching and the wording. Start with a rough list of what you actually did, paste in the job ad and let the AI pull out what the employer wants, then have it rewrite your bullets one section at a time with a strict rule: use only the facts I gave you, and ask me when something is missing. Finish with a recruiter-style review and a line-by-line fact check. That is roughly 30 minutes for a first version and 10 minutes for each tailored copy after that.

Below you get the prompts I would copy, a worked example from vague bullet to sharp bullet, how to keep the file readable for applicant tracking systems, and what to never paste into a chatbot.

Why “make my resume better” gives you a worse resume

The AI only knows what you put in the chat. If you give it a job title and a few lines, it fills the gaps with what people in that role usually do. That is where the trouble starts.

Northern Kentucky University’s career team describes exactly this: when you ask ChatGPT to improve your bullets, it makes guesses about your experience based on what it expects from someone in that role, and it may suggest numbers that are hypothetical, so you should never accept those changes without checking them for accuracy (NKU Career Services, PDF). Harvard’s career center goes one step further: generative AI should not be the primary author, “not least because its output will likely be very generic” (Harvard FAS Mignone Center).

So split the work in two:

• Your job: the raw facts, the real numbers you can defend, and the final say on every line.
• The AI’s job: reading the job ad closely, spotting which of your facts matter for it, tightening the wording, and pointing out gaps.

Get that split right and the rest of this is easy.

Step 1: Dump your raw facts first (10 minutes)

Don’t start in the chat. Start in a plain note. For each job you’ve had in the last ten years or so, write down what you were responsible for, what you actually did, and what changed because of it. Messy is fine. Fragments are fine.

• “Ran weekly supplier calls, 6 suppliers.”
• “Rebuilt the onboarding checklist, new hires stopped asking the same 10 questions.”
• “Took over the shared inbox when Sam left, cleared the backlog in about two weeks.”

Only write numbers you know are true. “About two weeks” is honest. “Improved efficiency by 47%” when you never measured it is not, and you will be asked about it in the interview. Harvard’s advice is the same: begin with your existing resume or as many details as you can pull together, including the impact you had and relevant numbers.

Already have a resume? Great, that’s your raw material. Add anything it leaves out, because the best stories often never made it onto the old version.

If you’re stuck, let the AI interview you instead of guessing:

I'm updating my resume. My role was [job title] at a [type of company] from [year] to [year].

Ask me 6 short questions, one at a time, to find concrete things I did and what changed because of them. Wait for my answer before asking the next one. Don't invent anything about me, and don't write any resume text yet.

Step 2: Let AI decode the job ad (3 minutes)

A resume isn’t one document. It’s one document per job. To tailor it, you first need to know what this employer is really asking for. Paste the full ad into ChatGPT, Claude or Gemini with this:

Here is a job ad. Read it like the recruiter who has to shortlist for it.

1. List the 5 skills or experiences this role needs most, in order of importance.
2. For each one, quote the words in the ad that show it.
3. Separate the must-haves from the nice-to-haves.
4. List the exact keywords and tool names I should use on my resume, but only where they are true for me.

Do not rewrite my resume yet.

[paste the full job ad]

The keyword list matters for software, not just for people. Harvard notes that many employers, especially those with lots of applicants, use applicant tracking systems that scan resumes for certain keywords or phrases. Their advice: use basic formatting, list your skills explicitly, and include keywords from the job description in context. “In context” is the key part. A keyword dumped in a list at the bottom does less for you than the same word inside a bullet that proves you used it.

Step 3: Rewrite your bullets, one section at a time

Now bring the two together. Do it per job, not the whole resume at once. Long inputs get you long, flat outputs. Short inputs get you sharp ones.

Below are my raw notes for one job, and the 5 needs you found in the job ad.

Rewrite my notes into 4 to 6 resume bullets.
Rules:
- Use ONLY the facts in my notes. Do not add numbers, tools, or results I didn't give you.
- If a bullet would be stronger with a number or detail I didn't give, ask me for it instead of guessing.
- Start each bullet with a plain, strong verb. No "spearheaded", no "synergy", no "results-driven".
- Put the bullets that match the job ad's top needs first.
- Max 2 lines per bullet.

My notes:
[paste notes for this job]

Top needs from the ad:
[paste the list from Step 2]

The line “ask me instead of guessing” does most of the work. The AI will come back with questions like “roughly how many new hires used the checklist?” That’s the moment to dig up a real number, or to leave the bullet without one. Both are fine. A made-up number is not.

A worked example: from vague to specific

This example is made up to show the method. Use your own facts.

The old bullet:

Responsible for onboarding and team communication.

The raw notes: “Rebuilt the onboarding checklist in Notion. New hires kept asking the same questions in Slack. After the new checklist most of those questions stopped. 9 people onboarded with it last year. Also started a Friday update email for the team.”

The job ad’s top need (from Step 2): “Improve internal processes and documentation.”

What the AI gave back, after asking how many people used the checklist:

Rebuilt the team's onboarding checklist in Notion, used to onboard 9 new hires last year, which cut repeat setup questions in Slack.

Started a weekly Friday update email so the whole team saw priorities and blockers in one place.

Notice what it did not do. No percentage, no “drove a culture of excellence”. Every word can be backed up in an interview. Harvard puts it simply: be able to speak to every line of your resume if asked in an interview. That’s your test for every bullet.

Step 4: Write a short summary that matches this job

The summary at the top is where most people paste the same generic paragraph into every application. It’s also the first thing a recruiter reads. Write it last, once your bullets are done:

Using only the bullets below, write a 2-sentence resume summary aimed at the job ad from Step 2.
Mention my years of experience in [field], my 2 strongest proof points for this role, and nothing I haven't listed.
Plain language. No adjectives like "passionate", "dynamic" or "results-driven".

[paste your finished bullets]

Then read it out loud. If it doesn’t sound like something you’d say to a person, change the words until it does. The University of New Hampshire’s career team puts the rule well: AI is your assistant, not your author, and how you write should reflect how you speak (UNH Career and Professional Success).

Step 5: Ask for a recruiter’s review

Once the draft is together, flip the AI’s role. Instead of writer, make it the tough reader:

Act as the recruiter for the job ad above. Review my resume against it.

1. In 3 lines: would you shortlist me, and why or why not?
2. Which must-haves from the ad are missing or weak on my resume?
3. Which bullets are vague or sound generic? Quote them.
4. List every number or claim on the resume so I can check each one.

Don't rewrite anything. Just give me the review.

[paste the resume without your name and contact details]

Point 2 is gold. If the ad needs something you really have but forgot to mention, add it. If it needs something you don’t have, don’t fake it. That is a gap to address in your cover letter or in the interview, not a line to invent.

UC San Diego’s career center suggests almost the same prompt: review the resume against the job description, identify the skills the employer is seeking, suggest where to highlight experiences you actually have, and do not invent experience or qualifications (UC San Diego Career Center).

Keep the file readable for applicant tracking systems

The best bullets won’t help if the system that receives your file can’t read it. NKU’s guide lists the basics: avoid photos, images, icons and graphics, avoid two-column layouts, mirror the language of the job description where you can, and include keywords from the posting. They also suggest putting the most relevant information in the top third of the first page.

In practice:

• One column, standard headings (Summary, Experience, Skills, Education).
• A plain font and a normal file type. Check the application instructions for which one they want.
• Tool names written the way the ad writes them (“Microsoft Excel” if that’s what it says).
• No skill bars, no star ratings, no headshot unless the employer or country norm asks for one.

You can also paste your final text back and ask: “Is there anything in this layout or wording that an applicant tracking system might read wrongly?” It’s a quick sanity check, not a guarantee.

What not to paste into the chat

Your resume is full of personal data. Every career center I checked says to strip it before pasting. NKU suggests removing your name and contact details before sharing your resume. UC San Diego warns against entering sensitive personal information or confidential employer information unless you understand how the tool stores it.

• Remove your name, phone, email, home address and date of birth. Add them back in your own document at the end.
• Leave out anything confidential from your current employer: client names, internal revenue, unreleased projects.
• If you use a work account, check your company’s AI policy first.

ChatGPT, Claude or Gemini for a resume?

Any of the three handles this method well. From my own use: Claude tends to keep to “only my facts” rules a bit more strictly and writes plainer English, ChatGPT is quick for back-and-forth questions in Step 1, and Gemini is handy if your notes already live in Google Docs. That’s opinion, not a benchmark. Pick the one you already use and spend the time on your facts instead. If you want the longer comparison, see ChatGPT vs Claude for writing.

Common mistakes (and the quick fix)

• Pasting the whole resume and saying “improve it”. Fix: one job section at a time, with the job ad’s needs next to it.
• Accepting numbers the AI suggests. Fix: the “ask me instead of guessing” rule, and the claim list from Step 5.
• One resume for every application. Fix: keep a master file with all your bullets, then rerun Steps 2 and 3 for each ad and pick the best 4 to 6 bullets per job. About 10 minutes per application.
• Keyword stuffing. Fix: every keyword lives inside a bullet that proves it. If you can’t prove it, leave it out.
• Ignoring the employer’s rules. Fix: read the application instructions. Harvard notes some organizations have rules about whether AI can be used in preparing application materials.

FAQ

Is it okay to use AI to write my resume?

Using AI to edit and tailor a resume built from your own facts is what university career centers themselves recommend, as long as every line is accurate and you can explain it. Using it to invent experience is not okay, and it tends to fall apart in the first interview question. Always check the employer’s own instructions too.

Can employers tell my resume was written with AI?

They can tell when it’s generic. The same buzzwords on every line, no specifics, nothing a person could ask a follow-up about. A resume built from your real facts, with numbers you can explain, reads like you, because the content is yours.

Should I write a new resume for every job?

Not a new one, a tailored one. Keep one long master version with everything, then let the AI pick and reorder the bullets for each ad using Steps 2 and 3. The facts stay the same. The order and emphasis change.

What about the cover letter and LinkedIn?

Keep all three consistent. The same “your facts, AI’s wording” method works for writing a cover letter with AI and for your LinkedIn profile. And if you hire people yourself, here is how to write a job description with AI.

Your next 30 minutes

Open a blank note and dump your raw facts for your last two jobs. Paste the job ad you care about into the Step 2 prompt. Rewrite one job section with the Step 3 prompt, answer its questions honestly, then run the recruiter review. By the end you’ll have one tailored resume and a master file you can reuse for every application after it.

This is how we teach at RobotsHelp: one real work task at a time, with prompts you can use the same day. If you want more of these, our free Launchpad on the RobotsHelp home page is the place to start. For everyday writing beyond job applications, see the AI writing assistant guide for busy professionals.

Got the interview? Our guide on how to use AI to prepare for an interview walks you through decoding the job ad, shaping your real stories, and running a spoken mock interview.

When that tailored resume turns into an offer, read our guide on how to use AI to negotiate salary to check the number and prepare your counter.

The wins you collect for your resume are just as useful once you have the job. When review season comes, our guide on how to use AI to write a self evaluation shows how to turn them into a clear self-review.

You found a job you actually want. The CV is ready. Then the form asks for a cover letter, and you sit there with a blank page and the same three sentences you wrote for the last application. You ask ChatGPT, and it hands you “I am writing to express my keen interest in this exciting opportunity.” Every hiring manager has read that line a thousand times.

Here is how to use AI to write a cover letter that still sounds like you: paste the job ad in first and let the AI tell you what the employer really wants, then give it two or three real stories from your own work that prove you can do that, ask for a short draft that uses only those facts, and finish by rewriting the opening and the closing line yourself and checking every claim. The AI does the matching and the structure. You bring the proof and the final say.

That is the whole method. Below you get the prompts I would copy, a worked example from job ad to finished letter, and the honest answers to “is this allowed?” and “can they tell?”

Why most AI cover letters get skipped

When you type “write me a cover letter for a project manager job”, the AI has nothing real to work with. No company, no stories, no reason why you want this role. So it fills the gap with the most average letter it can produce: big adjectives, a list of soft skills, and a promise to be “a valuable asset to your team”.

The fix is not a better tool. It is better input. Career services at universities that have looked hard at this say the same thing in different words. Harvard’s career center tells job seekers to paste in the full job description, ask for the top skills it needs, then add their own examples and “make it your own” (Harvard FAS Mignone Center). Michigan State puts it even more plainly: don’t ask the AI to write a letter, ask it to coach your story (MSU Career Services Network).

So split the job in two:

• Your job: the real stories, the real reason you want this role, the final edit.
• The AI’s job: reading the job ad closely, matching your stories to what they ask for, tightening the wording, and catching what you missed.

Step 1: Let AI decode the job ad (3 minutes)

Most people skip this, and it is the step that makes the rest work. A job ad is a wish list with a few must-haves buried in it. AI is good at finding them.

Open ChatGPT, Claude or Gemini and paste this, with the full ad underneath:

Here is a job ad. Read it like the hiring manager who wrote it.

1. List the 3 things this person most needs to be good at, in order of importance.
2. For each one, tell me which words in the ad gave it away.
3. What problem is this team probably trying to solve by hiring?
4. List any must-have requirements separately from the nice-to-haves.

Do not write a cover letter yet.

[paste the full job ad]

That last line matters. If you let it, the AI will jump straight to a letter. You want its reading of the ad first, because now you know what your letter has to prove.

Step 2: Bring your proof, not your CV

Your CV lists what you did. A cover letter shows one or two moments where you did it well. For each of the three needs from Step 1, write down one real story in plain words. Four lines each is plenty:

• The situation: what was going on.
• What you did: your part, not the team’s.
• The result: what changed. Use a real number only if you know it is true.
• Why it matters here: how it links to this job.

Stuck? Ask the AI to interview you instead of guessing. This works far better than handing it your whole CV:

I'm applying for the job above. Ask me 5 short questions, one at a time, to find real examples from my work that prove I can do the 3 things you listed. Wait for my answer before asking the next question. Don't invent anything about me.

A quick privacy note before you paste anything. You do not need your home address, phone number or date of birth in the chat. Leave them out and add them yourself at the end. MSU’s guide gives the same advice: share the job description and the experiences you want to highlight, not your full contact details (MSU). If you use a work account, check your company’s rules on what you can put into an AI tool.

Step 3: The draft prompt (copy this)

Now you have the employer’s needs and your proof. Put them together with rules that stop the AI from padding or making things up:

Write a cover letter for the [job title] role at [company], using ONLY the facts below. If something is missing, ask me instead of making it up.

What they need (from the ad): [the 3 needs from step 1]
My proof: [your 2 or 3 stories]
Why I want this role: [one honest sentence in your own words]

Rules:
- Max 250 words, 3 or 4 short paragraphs.
- Open with why this role, not with "I am writing to apply".
- One story per paragraph, each one tied to a need from the ad.
- Plain, confident language. No "passionate", "dynamic", "synergy", "valuable asset".
- Write like I talk: [paste 2 or 3 sentences from an email you wrote yourself].
- End with a simple next step, not a grovel.

The “write like I talk” line is the quiet hero here. Give the AI a few sentences you actually wrote, a good email to a client works, and the draft stops sounding like a template. Boston University’s career center makes the same point: name the tone you want and feed in your real background so the letter reflects your voice (BU Center for Career Development).

Step 4: Rewrite the first and last line yourself

This takes two minutes and changes how the whole letter lands. The opening is where a reader decides whether to keep going. The closing is what stays with them. Both should sound like a person, so write them yourself.

Good openings name something specific: the problem you noticed in the ad, the product you already use, the person who told you about the role. “Your ad mentions that the team is moving all client reporting into one system. I did exactly that at my last job, and I’d love to do it again with you.” That beats any line the AI suggested, because only you could have written it.

Step 5: Ask for a hiring manager’s critique

Before you send, flip the AI’s role. Ask it to judge instead of write:

You are the hiring manager for this job. Here is the ad and a cover letter.

1. Does the letter prove the 3 most important needs? Which one is weakest?
2. Point out any sentence that sounds generic or could be in anyone's letter.
3. List every fact, number, name or claim in the letter so I can check each one.
Don't rewrite the letter. Just give feedback.

Carnegie Mellon’s career center recommends exactly this switch: prompts like “give me targeted feedback on…” or “suggest areas of improvement” keep you in charge, while “build me” and “create a” lead to generic, homogenized text (CMU CPDC guidelines, PDF).

Step 6: Check every fact (the step that saves you in the interview)

AI tools can state wrong things with full confidence. OpenAI’s own help center says ChatGPT can produce incorrect facts and even made-up quotes or sources, and tells you to verify important information (OpenAI Help Center). In a cover letter that usually shows up in three places:

• Company facts. A “recent project” or “company value” the AI mentions may be old or invented. Check it on the company’s own site.
• Your numbers. If a percentage or team size crept in that you didn’t give it, delete it.
• Skills you can’t back up. If the letter says you “led” something you only helped with, fix it. You will be asked about it.

Then read the letter out loud once. If a sentence feels strange in your mouth, it will feel strange to the reader too.

A worked example: from job ad to letter

Here is how the method plays out for a made-up person, so you can see every step. Sam is an office manager applying for an operations coordinator role at a small logistics company.

Step 1 output (what the ad really wants): someone who can keep schedules and suppliers organized, who is comfortable with spreadsheets and booking tools, and who stays calm when plans change at the last minute.

Sam’s proof, in rough notes:

1. Took over supplier orders at the dental practice, moved them from paper to a shared sheet, no more double orders.
2. Set up the online booking tool for 4 dentists, trained the front desk.
3. Power cut on a full Monday, rebooked the whole day by phone before lunch.
Why this job: I like the puzzle of making a busy day run, and I want to do it at a bigger scale.

The finished letter after Sam’s own edits:

Dear Ms Jansen,

Your ad says the team needs someone who keeps a busy schedule from falling apart. That is the part of my job I enjoy most, and I'd like to do it at a bigger scale with you.

At the dental practice where I manage the office, I took over supplier orders that were tracked on paper. I moved them into one shared spreadsheet the whole team could see, and the double orders stopped.

When we switched to an online booking tool for our four dentists, I set it up and trained the front desk, so patients could book themselves and the phone got quieter.

And when a power cut hit on a fully booked Monday, I rebooked the entire day by phone before lunch. Plans change. I like being the person who sorts them out.

I'd be glad to talk about how I could help your team keep its days running smoothly.

Kind regards,
Sam de Vries

Notice what is not in there: no “passionate”, no invented percentages, no claims Sam can’t explain in an interview. The opening line is Sam’s own. The AI’s work was the structure, the matching and the trimming.

Which AI tool is best for cover letters?

Honestly, for this job the tool matters less than the input. ChatGPT, Claude and Gemini all handle the prompts above well. A few practical differences from my own use:

• Claude tends to keep a natural tone and follows “only these facts” rules well, which helps when you want it to sound like you.
• ChatGPT is strong at the critique step and at producing several opening options to choose from.
• Gemini is handy if your CV and notes already live in Google Docs.

Use whichever one you already have open. If you want the longer comparison, I wrote up how ChatGPT and Claude compare for writing.

Is it OK to use AI to write a cover letter?

In most cases, yes, as long as the letter is true and it’s yours. Career centers at Harvard, Michigan State, Boston University and Carnegie Mellon all publish guides on using AI for application materials. They agree on the line: use it to analyze, draft and give feedback, then edit until it says what you would say. Harvard adds one important caveat: read the application instructions, because some organizations have their own rules about AI in applications (Harvard FAS). If an employer says no AI, respect that.

Can employers tell you used AI?

There is no reliable test. OpenAI’s help center even notes that ChatGPT can’t reliably tell whether a text was AI-generated (OpenAI Help Center). But that is the wrong worry. What a hiring manager can spot is a generic letter: big words, no specifics, nothing that could only come from you. Follow the steps above and that problem disappears, whoever typed the first draft.

Common mistakes (and the quick fix)

• Pasting only “write a cover letter for X”. Fix: decode the ad first and bring real stories.
• Sending the first draft. Fix: rewrite the opening and closing yourself, then run the critique prompt.
• Leaving placeholders. “[Hiring Manager Name]” in a sent letter is the fastest way to the reject pile. Search for brackets before you send.
• One letter for every job. Fix: keep your stories in a note and rerun Step 1 for each new ad. Ten minutes per application, not an hour.
• Letting it inflate you. Fix: the “only these facts” rule, and a final read with one question: could I explain this sentence in an interview?

FAQ

How long should an AI-written cover letter be?

Keep it to one page and short paragraphs. The prompt above caps it at 250 words, which gives the reader the point without scrolling. If the employer gives a word limit, use theirs.

Can I use the same prompt for every application?

Use the same prompt, never the same letter. The prompt is a template. The job ad, the three needs and the stories you choose change every time.

Should I give the AI my whole CV?

You can, but you don’t have to. Your two or three best stories for this job give a sharper letter, and you share less personal information. Remove contact details either way.

What about my CV and LinkedIn?

Keep them consistent with the letter. If you are updating those too, the same “your facts, AI’s wording” method works for writing your LinkedIn profile with AI. And if you are on the other side of the desk, here is how to write a job description with AI.

Your next 15 minutes

Open the job ad you have been putting off. Run the Step 1 prompt. Write your three stories. Draft, rewrite two lines, check the facts, send. That is one application done properly, and you now have stories you can reuse for the next one.

This is the kind of thing we teach at RobotsHelp: one real task at a time, with the prompts you can use the same day. If you’d like more of these, our free Launchpad on the RobotsHelp home page is the place to start. For everyday writing beyond job applications, see the AI writing assistant guide for busy professionals.

Your cover letter and your resume should tell the same story. Our guide on how to use AI to write a resume shows how to tailor your CV to each job ad with the same fact-first approach.

Once the letter lands you a conversation, the next step is getting ready for it. Our guide on how to use AI to prepare for an interview shows how to practise answers out loud and get honest feedback.

And when the process ends in an offer, our guide on how to use AI to negotiate salary shows how to prepare a calm, well-researched counter.

You open your LinkedIn profile, see a headline you wrote three jobs ago and an About section that starts with “Experienced professional with a passion for…”, and you close the tab. You know it should be better. You just don’t want to spend a Sunday writing about yourself.

Here is how to use AI to write a LinkedIn profile that actually sounds like you: first write a one-page fact sheet about your work (real roles, real results, who you help, what you want next), then ask ChatGPT, Claude or Gemini to draft each section from those facts only: headline, About, experience, skills. Rewrite the first two lines of your About section yourself, check every claim, and save. The AI does the wording. You bring the truth.

Below you get the fact sheet template, a copy-ready prompt for each section, a worked example, and the checks that stop your profile from reading like everyone else’s.

Why AI-written LinkedIn profiles all sound the same

Ask any chatbot “write me a LinkedIn summary” and you get the same thing everybody else gets: “results-driven”, “passionate about innovation”, “proven track record”. The model has nothing real to work with, so it fills the gap with the most average words it knows.

The fix is not a cleverer prompt. It is better input. When the AI gets your actual projects, numbers you can stand behind and the people you work for, the draft becomes specific, and specific is what makes a stranger keep reading.

The career team at UMass Isenberg puts it the same way in their guide to using AI to build your LinkedIn presence: draft first, give the AI context, own your narrative, and avoid vague, generic or inflated AI language.

Step 1: Decide who your profile is for (2 minutes)

A profile for recruiters and a profile for clients are different documents. Before you open any AI tool, answer one question: who do I most want to read this?

• Recruiters and hiring managers want your target role, the skills they search for, and proof you have done the job.
• Clients or buyers (freelancers, consultants, founders) want to know what problem you solve, for whom, and how to start working with you.
• Peers and your industry want to know what you focus on and what you talk about.

Pick one main reader. You can still be useful to the others, but the AI writes much sharper copy when it knows who it is talking to.

Step 2: Build your fact sheet (15 minutes, the part that matters most)

This is the raw material. Write it fast, in bullet points, ugly is fine. Nobody sees it but you and the AI.

MY FACT SHEET
Main reader: [recruiters for X role / clients who need Y / peers in Z]
What I do now, in plain words: [one or two sentences]
Who I do it for: [type of company, team or client]
3 to 5 results I can prove: [what changed, for whom, with a number only if I can back it up]
Tools and skills I use every week: [list]
What I am known for among colleagues: [one line]
What I want next: [role, type of client, kind of conversation]
How I talk: [formal / plain / a bit dry / warm]
Words I never want on my profile: [e.g. guru, ninja, passionate, synergy]

Two rules for the fact sheet. First, only results you could explain in an interview. If a number is a guess, leave it out or describe the change in words (“cut the monthly close from a week to two days” beats a made-up percentage). Second, take out anything private: client names you are not allowed to share, internal figures, phone numbers. The Isenberg guide specifically tells you to remove personally identifiable information before you paste text into an AI tool.

If you have a CV, paste it under the fact sheet. It saves typing and gives the AI your job history in order.

Step 3: Let the AI write your LinkedIn headline

Your headline is the line under your name. LinkedIn’s help page on editing your headline explains that it shows up in search results and can be different from your job title. That is the point: you can use it to say what you are good at, not just what your business card says.

Use ONLY the facts in my fact sheet below. Write 8 LinkedIn headline options for my main reader.
Rules:
- Say what I do and who I do it for in plain words.
- Include two or three terms my main reader would actually type into LinkedIn search.
- Mix styles: 3 options as "Role | skill | skill", 3 as "I help [who] [do what]", 2 that show a bit of personality.
- No buzzwords from my "never" list. No emojis.
- After each option, say in five words what it emphasises.
If a fact you need is missing, ask me before you write.

[paste fact sheet]

Pick the two you like best and read them aloud. The one you could say at a networking event without cringing is the one. Then shorten it by hand: headlines get cut off on small screens and in search results, so the first few words carry the weight.

Step 4: Write your About section with AI (and fix the first two lines yourself)

The About section is where people decide if you are worth a message. LinkedIn’s own advice in Create a good LinkedIn profile is that it should express your mission, motivation and skills, ideally in one or two paragraphs, and bullet points are fine if you prefer them. Use that as your brief: short beats long.

Use ONLY the facts in my fact sheet. Write my LinkedIn About section in first person, in the voice I described.
Structure:
1. One or two opening lines that make sense on their own and tell my main reader what I help with.
2. What I focus on now and who I work with.
3. Two or three concrete results from the fact sheet, as short bullet points.
4. One line on what I want next and how to reach me.
Keep it under 250 words. Do not invent clients, numbers, awards, certifications or stories. Do not use any word on my "never" list.
Give me 2 versions: one more formal, one more conversational.

[paste fact sheet]

Now do the one thing the AI is worst at. Take the opening of your favourite version and rewrite it in your own words. LinkedIn shows only the start of your About section until a reader taps “see more”, so this is where a stranger decides. A line like “I help small accounting firms stop chasing clients for receipts” will always beat “I am a dedicated finance professional”.

Still writing from a blank page? The AI writing assistant guide for busy professionals covers the general habit of giving AI your notes instead of a topic.

Step 5: Turn job duties into results in your Experience section

Most Experience sections read like old job adverts: “Responsible for reporting. Managed stakeholders.” AI is genuinely good at turning that into something a reader cares about, as long as you give it the story behind each role.

Here is one job from my CV and my notes on what actually happened there.
Rewrite it as a LinkedIn Experience entry:
- One sentence on what the role was and for whom.
- 3 to 5 bullet points, each starting with a strong verb, each describing a change I made and what it led to.
- Use only facts from my notes. If a bullet has no result, write the action and mark it [ADD RESULT?] so I can fill it in.
- Plain language, no buzzwords.

Job: [title, company, years]
My notes: [what you did, what changed, who noticed]

That [ADD RESULT?] marker is the useful part. It shows you exactly where your profile is vague, without the AI quietly making something up to fill the gap. Do one role at a time. Your last two or three jobs deserve the most detail; older roles can be one line.

Step 6: Pick your skills and keep the whole profile consistent

LinkedIn lets you add up to 100 skills, according to its page on adding and removing skills, and it says a relevant list helps match you with the right opportunities. You do not need 100. You need the ones that match your headline and your main reader.

Here are my new headline, About section and Experience entries.
1. List the 15 skills that best match what these sections say, in order of importance for my main reader.
2. Point out anything that does not match between sections (a skill I claim in the headline that never shows up in my experience, or the other way round).
3. Point out any sentence that sounds generic or that a hundred other people could have on their profile.

This consistency check is the step most people skip. A headline that says “data analyst” above an About section about project management confuses both recruiters and LinkedIn search.

A worked example: from fact sheet to profile

Here is a short, made-up example so you can see the difference. Meet Sam, a fictional operations manager at a 40-person logistics company who wants a role at a bigger firm.

Sam’s fact sheet (shortened):

Main reader: recruiters hiring operations managers in logistics
Now: run day-to-day operations for a regional delivery company, 3 team leads report to me
Results: moved route planning from spreadsheets to a planning tool; set up a weekly KPI review the directors now use; trained 12 new dispatchers in one year
Tools: Excel, Power BI, route planning software
Want next: operations manager role at a larger logistics or e-commerce company
Never: guru, rockstar, passionate

Old headline: Operations Manager at [Company]

AI option Sam kept (lightly edited): Operations Manager | Logistics, route planning and KPI reporting | Building teams that run on clear numbers

Old About opening: “Experienced operations professional with a passion for excellence.”

New About opening, rewritten by Sam: “I make delivery operations calmer. For the last few years I have run daily operations at a regional delivery company, where I moved route planning out of spreadsheets and set up the weekly KPI review our directors now use.”

Notice what is missing: no invented percentages, no “award-winning”. Every line comes from the fact sheet, so Sam can talk about all of it in an interview.

Should you use LinkedIn’s own AI writer?

LinkedIn has a feature called Profile Writing Assistance that suggests text for your headline, About and Experience sections. Its help page says it is only available to a select group of Premium subscribers, and for Experience you need to write at least 20 words yourself before it can help. You can edit a suggestion before saving and revert to your original text.

If you have it, it is handy for quick rewrites inside the profile. The approach in this article works with any chatbot and gives you more control, because the AI only sees the fact sheet you wrote. You can also use both: draft in ChatGPT or Claude, then polish inside LinkedIn. Not sure which chatbot writes better for you? See ChatGPT vs Claude for writing.

Small things that finish the job

• Custom profile URL. LinkedIn’s page on managing your public profile URL says your custom URL can be 3 to 100 characters, with no spaces, symbols or the word LinkedIn, and recommends a variation of your name. Five minutes, and your link looks tidy on your CV and email signature.
• Edit in the right place. LinkedIn’s guide to editing the About section notes that the About section is separate from your headline, and if changes don’t show, check you are editing the right language profile.
• Proof beats claims. If you have a case study, a talk or a report you are proud of, add it to your profile. Writing a case study first? Here is how to write a case study with AI.
• Then post. A good profile gets visits when you are active. When you are ready, use the same fact-first method for writing LinkedIn posts with AI.

The final check before you hit save

Read your new profile once, slowly, and ask:

1. Could I explain every claim here in a job interview or a client call?
2. Would a stranger know in five seconds what I do and who for?
3. Is there any sentence a hundred other people could have? Rewrite it or delete it.
4. Does it sound like me talking, or like a press release?
5. Did I remove anything confidential from my notes and my chat history?

The Isenberg guide ends on the same point: verify AI-generated information before you use it. A profile is a promise. Keep it one you can deliver.

Common mistakes (and the quick fix)

• Asking for a profile with no input. You get generic filler. Fix: always paste the fact sheet.
• Letting the AI add numbers. Fix: tell it “only facts from my notes” and check every figure against your own records.
• Five sections, five different people. Fix: run the consistency prompt from Step 6.
• Keeping the AI’s opening line. Fix: always rewrite the first lines of your About section yourself.
• Copying a “perfect” profile you found online. Fix: use it for structure, never for wording. Your experience is the only thing nobody else can copy.

Applying for a role you found while fixing your profile? The same fact sheet does most of the work for your application too. Here is how to use AI to write a cover letter that still sounds like you.

FAQ

Can I use ChatGPT to write my LinkedIn profile?

Yes. ChatGPT, Claude and Gemini all write good headlines, About sections and experience bullets when you give them real facts. Without facts they write generic text, so start with the fact sheet above and edit the result yourself.

What is the best ChatGPT prompt for a LinkedIn About section?

The prompt in Step 4: first person, your main reader named, a fixed structure (opening, focus, two or three results, what you want next), under 250 words, and the rule “use only the facts in my fact sheet”. The rule matters more than the wording.

Can recruiters tell if my profile was written by AI?

What stands out to a reader is a generic profile, whoever wrote it. Specific results, plain words and an opening in your own voice are what make a profile read as yours.

Is it OK to use an AI LinkedIn headline generator?

For ideas, sure. Most generators ask for a job title and give you buzzwords back. A chatbot with your fact sheet gives you options grounded in what you actually do, and you can ask for more until one fits.

How often should I update my profile with AI?

Keep your fact sheet as a document and add to it when something real happens: a finished project, a new skill, a new goal. Every few months, rerun the prompts with the updated sheet. It takes minutes once the sheet exists.

Your next 20 minutes

Open a blank document, fill in the fact sheet, and run the headline prompt. That alone fixes the line most people see first. Do the About section tomorrow.

If you want more practical AI workflows like this one for your daily work, start with the free Launchpad on our home page.

Updating your profile because you are applying for jobs? Our guide on how to use AI to write a resume uses the same fact sheet to tailor your CV to each job ad.

When a recruiter reaches out and books a call, our guide on how to use AI to prepare for an interview helps you turn the same facts into strong spoken answers.