How to Write a Speech With AI (That Still Sounds Like You)

Written by

The calendar invite says “Q4 kickoff, 5 minutes from you.” Or a colleague is leaving on Friday and somehow you are the one giving the farewell. You open a blank document, type “Thank you all for coming,” and stall.

Here is how to write a speech with AI without it sounding like a greeting card: do not ask ChatGPT, Claude or Gemini to “write me a speech.” First dump your own facts, names and one or two real stories into the chat. Tell it who is listening, how long you have, and the one thing you want people to remember. Ask for an outline before a draft, then a draft written for the ear: short sentences, plain words, room to pause. Read it out loud, cut every line you would never say, and rehearse once with a timer or an AI coach. A five-minute work speech usually takes about 30 to 45 minutes this way, and most of that time is you talking, not typing.

Below: the six steps, copy-paste prompts, a worked example, prompts for the most common work speeches, and the mistakes that make an AI speech fall flat.

No password needed – your reset opens instantly.

No password. Just your email, and you’re in.

Why “write me a speech” gives you a speech nobody believes

Ask any chat assistant for “a speech for my team’s year-end meeting” and you get the same thing everyone gets: “journey,” “incredible milestones,” “none of this would have been possible without each and every one of you.” It is not wrong. It is just nobody’s speech.

The AI has no idea what your team actually did this year, who stayed late before the launch, or what joke only your department will get. A speech works when it is specific and when it sounds like the person saying it. Those two things have to come from you. What the AI is good at is the slow, fiddly part: putting your scattered points in a clear order, finding a strong opening line, trimming a ten-minute ramble down to five, and pointing out where a sentence is hard to say out loud.

So the method below keeps a simple split. You bring the material and the voice. The AI brings structure, speed and a second pair of eyes.

What you need

A chat assistant: ChatGPT, Claude or Gemini. The free plans are enough for a speech. New to them? Our ChatGPT tutorial for beginners, Claude tutorial for beginners and Gemini tutorial for beginners get you set up.
Ten minutes of your own notes: names, numbers, dates, the moments you want to mention. Rough is fine.
Your phone or a microphone for the rehearsal at the end.

Step 1: Dump the facts, not the speech (5 minutes)

Before you open the AI, write down everything the speech could include. Do not try to make it sound good. Just list it:

The occasion and the audience (“all-hands, about 40 people, half of them remote”).
What happened: projects, results, the hard bits, the people involved.
One or two small stories. “Sam fixed the checkout bug at 11pm and then brought croissants the next morning” beats any adjective the AI can come up with.
Anything you must say (a thank-you, an announcement, a next step) and anything you must not say (numbers that are not public yet, a sensitive reorg).

Typing feels slow? Talk instead. Record yourself rambling for three minutes on a walk, then turn the recording into text. Our guide on how to transcribe audio with AI shows the quickest ways. Speech that starts as speech usually sounds more natural at the end.

One rule for work: check your company’s AI policy before you paste in confidential numbers, client names or HR details. If in doubt, use placeholders like “Client A” and fill them in afterwards.

Step 2: Pick the one thing and the length (2 minutes)

Toastmasters, the public speaking organisation, suggests asking yourself: if listeners only remember one thing from my speech, what do I want that one thing to be? Your audience should be able to sum up your speech in one sentence (Toastmasters International, How to Write and Deliver a Speech). Write that sentence down. It goes into every prompt.

Then turn your time slot into a word budget. People speak at very different speeds, but a comfortable pace for slide presentations is often given as roughly 100 to 125 words a minute, and audiobooks are recommended at around 150 to 160 (Wikipedia, Words per minute). Toastmasters gives a similar rule of thumb: about three double-spaced pages take around five minutes to say out loud.

In practice: aim for about 120 words per minute of speaking time. Three minutes is roughly 360 words. Five minutes is roughly 600. If there is no fixed slot, go shorter. Nobody has ever complained that a kickoff speech ended early.

Step 3: Ask for an outline first, not a full draft (5 minutes)

A full draft in one go locks you into the AI’s structure before you have agreed with it. An outline is quick to check and quick to fix. Paste your notes from Step 1 below this prompt:

I need to give a [length]-minute speech at [occasion] to [audience].
The one thing I want them to remember: [your sentence].
Below are my raw notes. Use only facts from my notes. Do not invent stories, names, numbers or quotes. If something is missing, ask me.

Give me 3 different outlines (opening, 2 or 3 main points, close). For each, suggest an opening line that starts with a moment or a question, not with "thank you for coming".

My notes:
[paste notes]

Pick the outline that feels closest to how you would tell it over coffee. Change the order if you like. Then tell the AI which one you chose and what you changed.

Why three openings? Toastmasters points out that you have about 30 seconds to catch the audience’s attention, and recommends starting with a story or a small conflict instead of a formal greeting (Toastmasters). Seeing three options side by side makes the strongest one obvious.

Step 4: Get a draft written for the ear (5 minutes)

A speech is heard once, at the speed you talk. Nobody can scroll back. That changes how it should be written, and you have to tell the AI, because by default it writes for reading.

Write the speech from outline [B], about [number] words.
Write it to be spoken, not read:
- short sentences, most under 15 words
- plain words I would use in a meeting, no "journey", "testament", "synergy" or "each and every one of you"
- use "we" and "you" more than "I"
- repeat the main message once in the middle and once at the end
- mark a pause with [pause] where a line needs a moment to land
- end with one clear thing I want people to do or feel
Use only the facts and stories from my notes.

Two of those lines come straight from speaker advice. Toastmasters recommends using “we” and “you” more than “I” to pull the audience in, and using repetition to reinforce your main point (Toastmasters). The word ban list is ours: those are the words that make listeners tune out, because they have heard them in every corporate video.

Step 5: Make it sound like you (10 minutes)

This is the step most people skip, and it is the one that matters most. Read the draft out loud once, at full volume, as if the room were already there. You will notice three kinds of lines:

Lines you would never say. Too formal, too clever, a word you would not use. Mark them.
Lines you trip over. Too long, or two hard sounds in a row. Mark them.
Lines that say nothing. “This has been a truly remarkable year.” Remarkable how? Mark them.

Then give the AI a sample of how you actually talk. An email you wrote quickly, a Slack message, or a transcript of you explaining something works better than a description like “casual but professional.”

Here is a sample of how I normally talk: [paste a message or transcript].
Rewrite the lines I marked with ** so they sound like the sample. Keep everything else as it is.
Then list any remaining line that sounds like a greeting card or a press release, and suggest a plainer version.

Finally, check every fact. Names spelled right, numbers correct, the project called what your team calls it. AI assistants can get details wrong, and a wrong name in a farewell speech is the one thing people will remember.

Step 6: Rehearse with AI and a timer (10 minutes)

Practising out loud is the foundation of every other speaking tip, according to Toastmasters, which also advises timing yourself and recording yourself to catch filler words and speed (Toastmasters). Two AI tools make this easier than talking to a mirror:

Speaker Coach in PowerPoint. Put your speech title on a single slide, choose Rehearse with Coach on the Slide Show tab, and give the speech. Microsoft says it looks at your pacing, pitch, filler words and wordiness, and gives you a report afterwards. It works in English, and Microsoft describes it as available in free preview on the web and mobile (Microsoft Support, Rehearse your slide show with Speaker Coach).
ChatGPT Voice. Start a voice conversation and say “Wait until I ask you to respond.” OpenAI’s help page lists that as a way to keep Live from answering while you talk. Give the speech, then ask: “What was my main point? Where did it drag? Which line should I cut?” OpenAI also notes that voice transcripts may not exactly match what was said, so treat its feedback as a rough check (OpenAI Help Center, ChatGPT Voice).

If you run over time, do not speed up. Ask the AI: “Cut this to [number] words without losing the two stories.” Then rehearse once more. After two run-throughs, switch from the full text to five or six bullet points on a card, so you look at people instead of paper.

An example: a 3-minute thank-you at the team meeting

Made-up example to show the flow. Lena leads a support team of twelve. After a rough quarter with a new ticketing system, she has three minutes at the monthly meeting to say thank you without it sounding like an HR email.

Her notes (Step 1): new system went live in July, first two weeks chaos, queue doubled, Priya built a cheat sheet everyone used, Marco answered tickets on his day off (she wants to thank him but not encourage that), by September the queue was back to normal, next goal: shorter first replies.

Her one thing (Step 2): “We got through the hard part because people helped each other, and that is how we will hit the next goal too.” Three minutes, so about 360 words.

What the AI did: gave three outlines; she picked the one that opens with “Week two of the new system, the queue counter hit a number I had never seen before.” The ear-friendly draft came back at 410 words. Reading it out loud, she cut “a testament to our resilience,” replaced “leverage” with “use,” and asked the AI to rewrite the Marco line so it thanked him and also said clearly that days off are days off.

Result: a 3 minute 10 second rehearsal in Speaker Coach, which flagged a lot of “so” at the start of sentences. She took out four of them. Total time spent: about 35 minutes, of which ten were the walk where she recorded her notes.

Prompts for the most common work speeches

Use these as the first line of the Step 3 prompt, then add your notes.

KICKOFF / ALL-HANDS
A [5]-minute kickoff for [team] at the start of [quarter/project]. Goal: everyone leaves knowing the top 3 priorities and why they matter. Open with one concrete moment from last quarter.

FAREWELL FOR A COLLEAGUE
A [3]-minute farewell for [name], who is leaving after [years]. Warm, a little funny, never embarrassing. Use only the stories in my notes. End with a toast or a clear send-off line.

ACCEPTING AN AWARD OR THANKS
A [90-second] thank-you after receiving [award]. Short, humble, names the people who helped, ends quickly.

WELCOME AT A CLIENT OR COMPANY EVENT
A [2]-minute welcome for [event]. Say who we are, why we are all here, and what happens next. One light opener, no jokes about the audience.

OPENING OF A CONFERENCE TALK
The first 60 seconds of a [20]-minute talk on [topic] for [audience]. Start with a problem they recognise, then say what they will be able to do after the talk.

Giving a talk with slides? Write the speech first with this method, then build the deck around it. Our guide on how to use AI to make a presentation covers the slides and speaker notes side.

Which AI is best for writing a speech?

For a work speech, the honest answer is: the one you already use. ChatGPT, Claude and Gemini can all outline, draft and tighten a speech well, and the difference between them matters much less than the quality of your notes and the sample of your own voice. A few practical differences:

ChatGPT has the voice mode described above, which is handy for rehearsing on your phone.
Claude handles long pasted notes and transcripts comfortably and tends to follow “do not invent anything” instructions closely, in our experience.
Gemini is convenient if your notes already live in Google Docs or Gmail.

Dedicated “AI speech writer” websites exist too. Many of them work much like the prompts above, with a form in front. If you use one, the same rule applies: feed it your real material, or you get a generic speech.

Mistakes that make an AI speech fall flat

Letting it invent stories. An AI will happily write “I remember the day Sarah walked into my office…” about a Sarah who does not exist. Always include “use only facts from my notes” and check the draft.
Going over time. Drafts often come back longer than you asked. Count the words and cut before you rehearse, not on the day.
Keeping the written style. Long sentences with three commas read fine and sound terrible. If you run out of breath mid-sentence, split it.
Reading it word for word. Use the full text to rehearse, bullet points to deliver.
Pasting in confidential details. Follow your company’s AI rules and use placeholders when unsure.
Skipping the read-aloud. Ten minutes of reading out loud fixes more than an hour of prompting.

Is it OK to use AI to write a speech?

For most work occasions, yes, as long as the speech is honest. The stories are yours, the thanks are real, and you are the one standing up and saying it. Think of the AI as the colleague who helps you tidy your notes, not the person who talks for you. Where it gets uncomfortable is when the content is not yours: invented anecdotes, borrowed jokes passed off as your own experience, or a eulogy for someone the AI never knew. Keep the material real and the help is no different from asking a friend to read your draft.

Quick answers

Can AI write a speech for free?

Yes. The free plans of ChatGPT, Claude and Gemini are enough for outlining, drafting and editing a speech of a few minutes. Speaker Coach is described by Microsoft as available in free preview on the web, even without an Office subscription.

How many words is a 5-minute speech?

Roughly 500 to 750 words, depending on how fast you speak. About 600 words is a safe starting point. The only reliable check is reading it out loud with a timer, and remember you will probably read faster alone than in front of an audience.

How do I make ChatGPT write a speech in my own voice?

Give it a sample of how you really talk (a quick email, a Slack message, a transcript of you explaining something) and ask it to match that. Then read the result aloud and rewrite any line you would not say. The sample does most of the work.

Can AI help me with nerves before a speech?

It can help with the part you control, which is preparation: a clear structure, a speech that fits the time, and a few rehearsals with feedback. It cannot stand up for you. If public speaking is a big hurdle, practising in front of real people, for example at a Toastmasters club, gives you something an AI cannot.

Can I use the same method for a wedding toast?

Yes. The steps are the same: your stories, one main message, a word budget, a draft for the ear, and a read-aloud. Just be extra strict about “use only my notes,” because invented memories at a wedding are worse than at a team meeting.

Your next step

Got a speech coming up? Do Step 1 today: five minutes of notes, or a three-minute voice memo on your next walk. That is the part only you can do, and with it in hand the rest of this guide takes about half an hour.

Want more short, practical ways to use AI at work, one real task at a time? Join the free Launchpad.