Author: Pete

You need a small piece of code for work. A little calculator page your team can open in a browser, a script that tidies up last month’s export, a formula that finally does what you meant. You are not a developer. You have ChatGPT open in one tab and Gemini in another, and you just want to know which one to ask.

Here is the short answer. In the ChatGPT vs Gemini for coding question, both write good everyday code for free, and both let you see the result before you copy it anywhere. ChatGPT fits better when your work is files on your laptop and Excel-style data, because it runs Python on what you upload and its coding agent, Codex, is included on every ChatGPT plan, Free too. Gemini fits better when your work lives in Google Sheets, Docs and Gmail, because Canvas previews your code with a console for errors and Apps Script runs right inside your sheet. Do not trust a ranking. Run the 20-minute test further down this page on one real task and keep the one that gets it right.

Want one practical AI lesson like this every week?

ChatGPT vs Gemini for coding at a glance

Names and plans change every few months, so this is what the official pages say in October 2026.

• Write and see code in the chat, free: ChatGPT puts code in code blocks. Supported Python blocks get a Run button with a console, and HTML can be previewed (OpenAI Help Center). The free Gemini app includes Canvas, where you create, edit and preview code and small apps (Gemini plans).
• Coding agent on your own computer: Codex on the ChatGPT side, in the desktop app, a terminal, an editor extension and on the web. OpenAI says it is included across ChatGPT plans, Free and Go too, with usage limits that vary by plan (Using Codex with your ChatGPT plan). On Google’s side it is Antigravity, which has a $0 plan with basic weekly rate limits (Antigravity pricing).
• Paid plan most people start with: ChatGPT Plus at $20 a month (ChatGPT pricing). Google AI Pro at $19.99 a month in the US (Google AI plans).
• Where each one feels at home: ChatGPT with uploaded files, spreadsheets and folders on your laptop. Gemini inside Google Workspace.
• Instruction file the agent reads every time: AGENTS.md for Codex (AGENTS.md guide). GEMINI.md for Google’s tools.

No winner in that list, on purpose. Both write the kind of code a busy professional needs. What differs is where the code ends up running, and which account you already pay for.

Where will your code run? That decides it

Before you compare models, answer one question: where does this code have to live when it is done?

In a Google Sheet, Doc or Gmail

Then it is Apps Script. In Sheets you open Extensions, then Apps Script, paste the code and save. A script bound to a sheet can run when the sheet opens, when someone edits it, when a form comes in, or on a timer (Extend Google Sheets with Apps Script). Both tools write Apps Script well. Gemini is the natural pick if you already pay for Google AI Pro, because it also sits inside Gmail, Docs and Sheets on that plan. We tested exactly this kind of task in our Claude vs Gemini for coding guide.

On your laptop, working on files

A script that merges twelve CSV exports, renames a folder of invoices, or cleans up a spreadsheet before you send it. ChatGPT has the easier start here: upload a .csv or .xlsx and it writes and runs Python on it in a sandbox, then hands you the result (Data analysis with ChatGPT). When the job is a whole folder, Codex can work on it directly.

In a browser, as a small tool for your team

A quote calculator, a meeting cost counter, a simple checklist page. One HTML file, no server. Both tools can build this and show you a live preview, so it is the fairest head-to-head. That is the test we use below.

ChatGPT for coding: what you actually get

Code blocks you can run and preview. Ask for code and ChatGPT places it in a code block. Depending on your plan and the block, you can edit it directly, ask for changes, preview HTML and charts, or press Run on Python and read the output or the error in a console. It all runs in a sandbox (OpenAI Help Center).

Python on your own data. Upload a spreadsheet and ChatGPT can clean, merge or chart it in a notebook-style environment. Two limits to know: that environment cannot make web requests or API calls, and tables inside scanned PDFs or images may not come through with exact values (Data analysis with ChatGPT). Free accounts get limited data analysis, paid plans get more.

Codex, the agent. Sign in with your ChatGPT account in the desktop app (Codex mode), the Codex CLI, the editor extension or Codex web. In a terminal you open your project folder, type codex, sign in and describe the task. Inside the desktop app, /init creates an AGENTS.md file where you write your house rules, such as “never touch the originals” (OpenAI Help Center). In a plain folder Codex starts carefully: it recommends read-only mode for folders that are not under version control, and it asks before it edits outside your working folder or uses the network (Codex approvals and security).

The catch: Free and Go get Codex with tighter limits, and Codex Cloud, which runs tasks on OpenAI’s computers while yours sleeps, is not included on those plans. Codex also shares one allowance with ChatGPT Work on plans that use shared allowances, so a long agent task eats into the same pool.

Gemini for coding: what you actually get

Canvas, with a console. Turn on Canvas below the text box and ask for the code. Canvas has a Code view where you can edit it yourself, a preview, a console that shows errors and logs from that preview, and a “Show recent changes” view so you can see what Gemini changed between versions (Gemini Apps Help: Canvas). For small web tools, that write, preview, read the error, fix loop is exactly what a non-developer needs.

Google AI Pro ($19.99 a month in the US). More access to Google’s Pro models, a 1 million token context window, more agent requests in Antigravity, and Gemini inside Gmail, Docs and Sheets (Google AI plans). Google also says you can upload a code repository of up to 30,000 lines and have Gemini reason across it (Gemini plans). Handy when a former colleague left you a long script and nobody knows what it does.

Antigravity, the agent. Google’s coding agent comes as a desktop app and a CLI, and the $0 plan gives you access to several agent models with basic weekly rate limits (Antigravity pricing). Paid Google plans raise those limits.

The 2026 change most ChatGPT vs Gemini guides miss

Two things moved this year, one on each side, and older comparison articles get both wrong.

• Gemini CLI is no longer the free option for personal accounts. Google is moving its terminal tool to Antigravity CLI. Since June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions no longer serve free users or Google AI Pro and Ultra subscribers; company licences and paid API keys keep working (Google Developers Blog). If a tutorial tells you to install Gemini CLI on your personal account, use Antigravity instead.
• ChatGPT’s canvas side panel is gone with the newer models. OpenAI moved writing and coding into editable blocks directly in the chat. Paid users can still reach canvas through legacy models until those are retired (ChatGPT release notes). If you cannot find canvas, nothing is wrong with your account. Use the Run, Edit and Preview buttons on the code block.

So when you read “Gemini has the bigger free coding limit” or “open canvas in ChatGPT”, check the date first.

Test ChatGPT vs Gemini on one small web tool

Benchmarks measure puzzles you will never write. One real task from your own week tells you more in 20 minutes. This test uses a tiny browser tool because both chats can preview it for free, so neither gets a head start.

Step 1. Pick something your team would actually use. The example below is a meeting cost calculator. Swap it for a quote calculator, a shift planner, whatever you keep doing on paper. Small, real, easy to check.

Step 2. Paste the same prompt into ChatGPT and into Gemini with Canvas turned on:

I am not a developer. Build a single HTML file I can open in any browser, no installs, no internet connection needed.

It is a "meeting cost calculator":
- I enter the number of people, the average hourly rate and the meeting length in minutes.
- It shows the total cost of the meeting, rounded to whole euros.
- It has a Start button that counts the cost up live, second by second, and a Stop button.
- It works on a phone screen too.
- If I leave a field empty or type letters, it shows a friendly message instead of breaking.

Then:
1. Show me a preview I can try right here.
2. Tell me how to save it and share it with my team.
3. List anything you assumed.

Step 3. Try to break both versions. Leave a field empty. Type “abc” as the rate. Press Start twice. Shrink the window to phone width. If one breaks, paste the error or describe what happened once, and see if the fix works.

Step 4. Score them on four questions:

1. Did it work in the preview on the first try, or after one fix?
2. Did it handle your silly inputs with a clear message?
3. Could you follow the save and share steps without searching anything?
4. Did it list its assumptions honestly, such as the currency or how it rounds?

The one that wins on your own task is your answer for this kind of job. If both pass, keep the one you already have open all day. And if what you really wanted was answers from a spreadsheet, not a tool, you may not need code at all: see how to analyze a spreadsheet with ChatGPT.

Which one for which job

• A formula, a regex or a quick snippet: either free chat. Use the one already open.
• Apps Script for Sheets, Docs or Gmail: both write it. Gemini wins on convenience if you pay for Google AI Pro anyway.
• Cleaning or merging files you can upload: ChatGPT, because it runs Python on your upload and gives you the finished file back.
• A script that works through a folder on your laptop: an agent. Codex comes with the ChatGPT account you may already have; Antigravity has a free plan too. Start either one in its most careful mode on a copied folder.
• Understanding a long script someone else wrote: Gemini Pro’s large context and repository upload help here. In Codex, open the folder and ask “explain what this does, file by file, before you change anything”.
• A small web tool for your team: run the test above. Check the sharing rules before you send a link: Google notes that anyone with the public link to a shared Canvas app can view and edit data associated with it.

Weighing Claude as well? Our ChatGPT vs Claude for coding guide compares Codex with Claude Code. And if you want one everyday assistant rather than a coding tool, ChatGPT vs Gemini vs Copilot and ChatGPT vs Gemini vs Grok cover writing, files and fresh news.

Four habits before you run AI-written code at work

1. Work on a copy. A copied sheet, a copied folder, trimmed sample data. Every time.
2. Ask for the plain-English version first. “Explain this line by line, and tell me what it deletes or overwrites” takes ten seconds and catches most surprises.
3. Keep secrets out of the chat. Never paste passwords or API keys. Use a placeholder like YOUR_KEY_HERE and fill it in yourself.
4. Check your data settings and your company’s rules. On ChatGPT Plus and Pro, conversations may be used to improve models unless you turn training off in data controls, and that applies to Codex too (OpenAI Help Center). Business accounts on both sides work differently, so ask IT which tool may see work files.

Free or paid: what to do this month

Start free on both. The free chats write and preview the code most professionals need, and both agents have a free tier to try on one small folder.

Pay when you hit a real wall, not before. Usually that means you run out of usage in the middle of real work, or you keep copying code between the chat and your computer every day. Then pick one plan, ChatGPT Plus or Google AI Pro, about $20 a month each, and use it for a month on real tasks. If your days are in Gmail and Sheets, Google AI Pro also brings Gemini into those apps. If your days are in files and folders, Plus gives Codex more room.

If you would rather have something that runs on its own every morning than a script you start by hand, a no-code automation may beat both. Our guide on building an AI agent without code shows that route.

Quick answers

Is ChatGPT or Gemini better for coding?

Neither wins for everyone. ChatGPT is the easier start for code that works on your own files, and Codex comes with every ChatGPT plan. Gemini is the easier start for code that lives in Google Workspace, and Canvas shows errors in a console as you go. Your own one-task test is the honest tie-breaker.

Can I use ChatGPT and Gemini for coding for free?

Yes. Both free chats write code and let you preview or run it. ChatGPT Free includes Codex with limited usage, and Antigravity has a $0 plan with basic weekly limits. Paid plans mostly buy you more room.

Is Gemini CLI still free?

Not for personal accounts. Since June 18, 2026, Gemini CLI no longer serves free or Google AI Pro and Ultra individual users, and Google points them to Antigravity CLI. Paid API keys and company Code Assist licences still work.

Do I need to know how to code?

No. You need to know what result you want and how to check it. Describe the input and the output in plain words, ask for setup steps and assumptions, and test on a copy. You pick up the basics as you go, and our free Launchpad sends you one practical lesson like this at a time.

Pick the one small tool you keep doing on paper. Paste the test prompt into both chats tonight, try to break each version, and keep the one that held up.

Deciding which paid plan to keep for the rest of your work, not only coding? Read ChatGPT vs Gemini Pro: which $20 plan to pay for at work.

Want the next one in your inbox?

Sources

• Using Codex with your ChatGPT plan (OpenAI Help Center)
• Working with writing blocks and code blocks in ChatGPT (OpenAI Help Center)
• Data analysis with ChatGPT (OpenAI Help Center)
• ChatGPT release notes (OpenAI Help Center)
• Codex approvals and security (OpenAI Developers)
• Custom instructions with AGENTS.md (OpenAI Developers)
• ChatGPT plans and pricing (OpenAI)
• Gemini plans (Google)
• Google AI plans and prices (Google One)
• Create docs, apps and more with Canvas (Gemini Apps Help)
• Google Antigravity pricing (Google)
• Transitioning Gemini CLI to Antigravity CLI (Google Developers Blog)
• Extend Google Sheets with Apps Script (Google for Developers)

You have a free Claude tab open and a free Gemini tab next to it. Both keep nudging you to upgrade, both cost about the same, and you only want to pay for one. Which one actually earns its $20 at your desk?

The short answer: pick Claude Pro if your hardest work is reading and writing: long reports, contracts, proposals, client emails that have to sound exactly right. Pick Gemini Pro (Google’s AI Pro plan) if your work already lives in Gmail, Docs, Sheets and Drive, or if you also want images, video and a big pile of cloud storage in the same bill. The two plans cost almost the same, so the deciding question is not “which AI is smartest”. It is “where does my work live, and what kind of work eats my week?” Further down there is a 30-minute test that settles it with your own files.

Want one practical AI lesson like this every week? Join Launchpad free, right above.

First: what “Gemini Pro” actually means

This trips up a lot of people. “Gemini Pro” is used for two different things:

• The plan. Google’s paid plan is called Google AI Pro. It used to be sold as Gemini Advanced. It costs $19.99 a month in the US and bundles Gemini with 5 TB of storage and extras across Google apps (Google AI plans).
• The model. Google also names its models “Pro” and “Flash”. Right now the AI Pro plan gives you Gemini 4 Argon, Google’s newest frontier model, plus expanded access to Gemini 3.1 Pro (Gemini plans).

On the other side, Claude Pro is Anthropic’s personal plan: $20 a month, or $17 a month if you pay for a year up front. It unlocks Opus, Anthropic’s strongest everyday model, plus Projects, Research and Claude Code (Claude pricing).

So when people search “Claude vs Gemini Pro”, they are really comparing two $20 plans. That is what this page does, for normal office work. If your question is about scripts and code, we wrote a separate guide: Claude vs Gemini for coding.

Claude Pro vs Gemini Pro at a glance

Both companies rename features often. This is what the official plan pages say on 22 October 2026.

Question	Claude Pro	Google AI Pro (Gemini)
Price	$20 a month, or $17 a month billed yearly	$19.99 a month (US)
Main models	Opus, Sonnet, Haiku	Gemini 4 Argon, 3.1 Pro, Flash
Long files	Context window up to 1M tokens, varies by model	1M token context, uploads up to 1,500 pages
Lives inside	Its own app, Chrome, Microsoft 365 and Outlook, Google Docs, Sheets and Slides (beta add-on)	The Gemini app, Gmail, Docs, Sheets, Slides, Meet, Chrome
Images and video	Reads images, does not generate them natively	Generates images and video
Extras in the bill	Claude Code, Projects, Research	5 TB storage, Deep Research, Gemini Notebook, YouTube Premium Lite where available
Usage limits	Resets every 5 hours, plus weekly limits	Refreshes every 5 hours, until a weekly limit

Notice what is not in that table: a clear winner. Each one is better at different jobs.

Where Claude Pro wins: careful reading and writing

The work where Claude shines is the work where every word counts. Think of a 40-page supplier contract you need to understand before a call, or a proposal that has to sound like you, not like a robot.

• Writing you barely have to edit. Independent reviewers keep pointing to the same thing: Claude’s writing tends to read more naturally and needs less editing (The AI Rankings: Claude vs Gemini). That matches my own experience. For a client email or a LinkedIn post, Claude’s first draft is usually closer to what I would have written myself.
• Close reading of long documents. Ask Claude to quote the exact clause behind every answer, and you can check each quote in seconds. A side-by-side work guide reached the same conclusion: Claude for close reading of contracts, policies and reports (Ynteractive: Claude vs Gemini for work). We show the exact method in Claude AI: analyze long PDF files.
• Projects. On Pro you can keep a project per client with its own files and instructions, so you stop re-explaining the same background every morning (Claude pricing).
• Microsoft 365 and Outlook. If your company runs on Word, Excel and Outlook, Claude Pro now includes Claude for Microsoft 365 and Outlook. Gemini does not live there.
• Claude Code included. Even if you never write code, Claude Code can tidy a folder of files or rename 300 exports for you. It comes with Pro, not with the free plan.

New to Claude? Our Claude tutorial for beginners gets you productive in your first 30 minutes.

Where Gemini Pro wins: your Google apps and the extras

Gemini’s big advantage is not a benchmark. It is location. If your day is Gmail, Docs, Sheets and Drive, Gemini is already sitting in those apps.

• It knows your Workspace. Gemini in Docs can draft using information from Drive, Gmail, Chat and the web, and it can match the style and structure of an existing document (TechRepublic: Claude vs Gemini in Google Workspace). No copy and paste between tabs.
• Very big uploads. Google lists a 1 million token context window on AI Pro and file uploads of up to 1,500 pages (Gemini plans). Think a year of meeting minutes, or every customer survey answer you have.
• Deep Research. Gemini browses hundreds of sources and writes a research report in minutes, which is handy for a competitor overview or market scan (Google AI plans). Always open the sources it lists before you use a number.
• Images and video in the same plan. Need a quick visual for a slide or a short clip for social? Gemini generates both. Claude does not generate images natively (The AI Rankings).
• The bundle. AI Pro includes 5 TB of storage across Gmail, Drive and Photos, and you can share the plan’s AI benefits with family members (Google AI plans). If you already pay Google for storage, part of the $19.99 is money you were spending anyway.

Want a ready-made Gemini job to try? Our Gemini for automated status updates guide turns scattered notes into a weekly update in a few minutes.

Claude vs Gemini inside Google Docs and Sheets

This used to be an easy call. Since 6 October 2026, Claude has a public beta add-on that opens in a sidebar inside Google Docs, Sheets and Slides for paid Claude plans (TechRepublic). So what is the difference now?

• Gemini sees more. It pulls context from across your Google apps. Claude works with the open file plus the connectors you switch on.
• Claude asks first. Its default “Ask before edits” mode lets you approve each change before it touches your document. If you have ever had an AI rewrite a paragraph you liked, you will appreciate that.
• Sheets is a toss-up. Claude can build models, fix formulas and clean data with Python behind the scenes. Gemini handles a wide range of native Sheets actions like pivots, charts, filters and dropdowns. TechRepublic’s verdict: choose by task.
• The beta has gaps. Claude cannot read Docs comments yet, and the sidebar has to stay open while it works.

If your company is on Google Workspace, check with whoever runs it first. Workspace customers get Gemini through a business add-on, not the personal AI Pro plan (Google AI plans FAQ).

Usage limits: will you run out?

Both plans work the same way now, which makes comparing easier. Claude’s limits reset on a rolling 5-hour window, with weekly limits on top, and Pro gives you at least 5 times the usage of Free per session (Claude pricing FAQ). Gemini’s limits also refresh every 5 hours until you hit a weekly limit, and AI Pro gives 4 times the usage of the free plan (Gemini plans).

Neither company publishes a fixed message count, because long chats, big files and the heavier models use more. In my experience, a normal office day of emails, summaries and a document or two fits comfortably in either plan. Where people hit the wall is long chats with huge files. The fix on both sides: start a fresh chat per task, and use a Project or Notebook for files you keep coming back to.

Privacy: what happens to what you paste

This matters more than any feature if you handle client data. Check these settings before you paste anything sensitive.

• Claude (Free, Pro, Max): the plan page lists model training as “Opt-out”, so it is a setting you control. Team and Enterprise plans are not used for training by default (Claude pricing).
• Gemini (personal account): with the Keep Activity setting on, Google uses your chats to improve its services, including training AI models, and a subset is read by human reviewers. Google itself says not to enter confidential information you would not want a reviewer to see. Turn Keep Activity off, or use temporary chats, and future chats are not used for training (Gemini Apps Privacy Hub).

The simplest rule: if your employer has approved one of them for work files, use that one for work files. Use your personal plan for everything else.

The 30-minute test: one real task in both

Reviews help, but your own work decides. Here is the test I run whenever someone asks me which plan to buy. Both have a free version, so you can do this before paying anything.

1. Pick one real task from this week. Something you actually have to deliver: a reply to a tricky client email, a summary of a long report, or a first draft of a proposal.
2. Remove anything confidential. Swap client names for “Client A” and drop the numbers you would not share.
3. Paste the exact same prompt into both. Use this one, or adapt it:

You are helping me with a real work task.

Context: I am a [your job] and this is for [who will read it].
Task: [for example: read the attached report and write a one-page summary my manager can act on].

Rules:
- Start with the 3 points that matter most.
- Quote the exact sentence from the document behind every key point.
- List anything you are unsure about at the end.
- Write in plain, friendly English. No jargon.

4. Score both answers on three things, each from 1 to 5: how much you had to edit, whether the quotes were real when you checked them, and whether you would send it as is.
5. Repeat with a second, different task. One writing job, one reading job. That covers most office work.

The plan with the higher total is your plan. It usually takes half an hour, and it beats any ranking because it uses your files, your tone and your standards.

So which one should you pay for?

• You write for clients all day (proposals, emails, reports): Claude Pro.
• You review long documents (contracts, policies, tenders): Claude Pro, with the quote-every-answer habit.
• Your work lives in Gmail, Docs and Drive: Gemini Pro.
• You also need images, short videos or more storage: Gemini Pro, because the bundle covers all of it.
• Your company runs on Microsoft 365: Claude Pro fits better than Gemini there. Also look at Copilot, compared in ChatGPT vs Gemini vs Copilot.
• You are torn between Claude and ChatGPT instead? Read ChatGPT vs Claude AI for work.

Plenty of people end up using both: Gemini for the everyday Gmail and Docs work, Claude for the careful writing and the long reviews. That is a fine setup. Just start with one paid plan, use it hard for a month, and only add the second when you keep hitting a job it does badly.

Quick answers

Is Claude better than Gemini Pro?

For careful writing and close reading of long documents, Claude is usually the stronger pick. For work inside Google apps, research reports, images and video, Gemini Pro usually wins. Neither is better at everything, which is why the 30-minute test above is worth doing.

Is Gemini Pro the same as Gemini Advanced?

Gemini Advanced was the old name. Today Google sells the paid plan as Google AI Pro, and inside it you get Gemini 4 Argon plus expanded access to Gemini 3.1 Pro (Gemini plans).

Can I try both for free first?

Yes. Claude Free includes chat, web search, file creation and connectors. The free Gemini app includes Canvas, Deep Research and image generation. The paid plans add stronger models and higher limits (Claude pricing, Gemini plans).

Does Claude work in Gmail?

Claude can connect to Google services through its connectors and now has a beta add-on for Docs, Sheets and Slides, but you mostly work in Claude’s own app. Gemini sits directly inside Gmail, which is the smoother option if email is where your day happens.

Which one handles bigger files?

Both list context windows of up to 1 million tokens on paid plans, which is far more than most office documents. Google also states uploads of up to 1,500 pages. For a normal report or contract, size will not be the deciding factor. Accuracy will, so always check the quotes.

Still weighing ChatGPT instead of Claude? Here is the same head-to-head for the other pair: ChatGPT vs Gemini Pro.

Make your pick this week

Take one real task, run it through both free versions, and pay for the one that saved you the most editing. Then use it every day for a month. That is how you stop comparing AI tools and start getting hours back. If you want a short, practical lesson like this one every week, it is free:

Sources

• Claude pricing and plan FAQ (checked 22 October 2026)
• Gemini plans (checked 22 October 2026)
• Google One: Google AI plans (checked 22 October 2026)
• Gemini Apps Privacy Hub
• TechRepublic: Claude vs Gemini in Google Workspace (7 October 2026)
• Ynteractive: Claude vs Gemini for work
• The AI Rankings: Claude vs Gemini

You have heard n8n can take whole chores off your plate. Then you open it, see an empty canvas full of plus signs, and close the tab. That is normal. Nobody learns n8n by staring at it.

The short answer: start on n8n Cloud’s free trial, build n8n’s own 15-minute “first workflow” to learn the three ideas that matter (trigger, nodes, data passing from one node to the next), then build one small workflow for your own work with an AI step in the middle. This guide walks you through exactly that, in about an hour, without writing code.

Want short, practical AI lessons like this one? Join Launchpad free, right above.

What n8n actually is, in plain words

n8n is a tool that connects your apps and runs steps for you. You draw the steps on a canvas, like boxes on a whiteboard, and n8n runs them in order. Each box is called a node, and each node does one job: read a form, ask an AI model a question, write a row in a sheet, send an email.

Every workflow starts with a trigger. That is the answer to one question: “When should this run?” Every morning at 8? When someone fills in a form? When a new email arrives? n8n’s own quickstart puts it the same way: the first thing you decide is how the workflow starts (n8n docs: Build your first workflow).

The third idea is the one that makes it click. The output of one node becomes the input of the next. You see that data on screen and drag fields from it into the next step. That drag-and-drop is how you use live data without code.

That is really it. Trigger, nodes, data flowing between them. Everything else is detail you pick up as you go.

Cloud or self-hosted: what a beginner should pick

You can use n8n two ways. n8n hosts it for you (n8n Cloud), or you run it on your own computer or server (self-hosted). n8n’s own guide is clear about who should do what: if you want to start right away and you do not have technical expertise, pick n8n Cloud, because there is nothing to install. If you want to run n8n for free, self-host the Community edition (n8n docs: Choose how to use n8n).

My honest advice: start on Cloud. Self-hosting is great later, but on day one it adds Docker, updates and backups to a learning curve you do not need yet.

Here is what I checked on n8n’s pricing page today (n8n pricing):

• Free trial: no credit card needed. It comes with 1,000 executions, 5 at the same time, a 180-second run time limit and 100 AI credits. n8n’s docs describe it as a 14-day trial with Pro features.
• Starter: 20 euro a month, billed annually. Unlimited workflows and users, AI nodes included, 5 executions at the same time.
• Pro: 50 euro a month, billed annually, for people running automations every day.
• Community edition: free, but you host it yourself (n8n docs: Host n8n).

One thing that makes n8n easy to budget: it charges per execution, and an execution is one full run of a workflow, no matter how many steps it has. A workflow that runs once a day uses about 30 executions a month. That is how n8n explains it on the same pricing page. If you want the full comparison with a per-task tool, read n8n vs Zapier for agency workflows.

Step 1: Build n8n’s own first workflow (15 minutes)

Do not skip this. n8n’s official quickstart is short, it needs no accounts or API keys, and it teaches exactly the moves you need for everything else (n8n docs: Build your first workflow). You build a small “Weather Log”:

1. Schedule Trigger: run every day at 8am.
2. HTTP Request: fetch the weather for Berlin from n8n’s sample-data service.
3. Edit Fields (Set): turn the raw data into one readable sentence by dragging fields in.
4. Data table: save each report as a new row, so you build a log over time.

When you drag a field into a box, n8n writes something like {{ $json.city }} for you. That is an expression: a placeholder n8n swaps for the real value when the workflow runs. You never have to type these by hand. Drag, look at the preview under the field, done.

At the end you press Publish. That is the switch between “I am testing this” and “run it for me from now on.” Remember that word, because it trips up almost every beginner later.

Step 2: Your first useful workflow with an AI step

Now the part that actually saves you time. We build a workflow many freelancers and small teams would love: someone fills in your contact form, AI reads the message, sorts it and writes a draft reply, and the draft waits in your Gmail for you to check.

The problem it solves: enquiries sit unanswered because writing that first reply feels like a task. The result: you open Gmail, find a ready draft, tweak one line and press send. You stay in charge of every word that leaves your inbox.

Four nodes. That is the whole build.

Node 1: n8n Form Trigger (the “when”)

Click Add first step and pick the n8n Form trigger. n8n builds the form page for you (n8n docs: n8n Form Trigger). Give it a title like “Ask me anything” and add three fields:

• name (Text, required)
• email (Email, required)
• message (Textarea, required)

Press Execute step. n8n opens the test form. Fill it in as if you were a client, something like “Hi, do you do bookkeeping for small webshops? What would a monthly package cost?” Submit, and you will see your answers appear as data in the node.

Node 2: OpenAI (the AI step)

Click the plus after the form and add the OpenAI node, then choose a text action that sends a message to a model. On n8n Cloud you can pick Use Gateway credits in the credential field, so you can run this step without your own OpenAI account (n8n docs: OpenAI node). If you prefer Claude or Gemini, n8n has an Anthropic node and a Google Gemini node too. The idea stays the same.

In the prompt field, paste this and drag the form fields into the spots in brackets:

You help me answer enquiries for my small business.

Here is a new message from [name] ([email]):
[message]

Do two things:
1. On the first line, write ONE label: QUOTE, QUESTION, SUPPORT or OTHER.
2. Below that, write a short, friendly reply in my voice (max 120 words).
   Thank them by first name, answer what you can, and suggest a 15-minute call
   if they asked about prices. Never invent prices or promises.
   Sign off with "Best, [your first name]".

Change the sign-off and the labels to fit your work. Press Execute step and read what comes back. If the tone is off, fix the prompt, not the output. Two or three rounds is normal.

Tip from experience: the line “Never invent prices or promises” is not decoration. Without it, a model will sometimes make up a number that sounds right. If you want to get better at prompts like this, my Claude tutorial for beginners covers the same three-part prompt idea.

Node 3: Edit Fields (tidy up)

Add an Edit Fields (Set) node, the same one you used in the weather workflow. Create two fields:

• subject: type “Re: your question” and drag the label from the AI output after it, so you can see at a glance what kind of message it is.
• reply: drag in the AI’s text.

This step is optional, but it keeps the next node clean and teaches you the single most useful habit in n8n: shape the data before you send it anywhere.

Node 4: Gmail, create a draft (the “result”)

Add the Gmail node and choose Create a draft (n8n docs: Gmail node). Connect your Google account when n8n asks. Then drag in the fields: the visitor’s email as recipient, your subject, and your reply as the message.

Why a draft and not “Send a message”? Because your first AI workflow should never email a client without you seeing it. Run it for a few weeks. Once you trust the replies for one label, say QUESTION, you can switch that branch to sending. Not before.

Test it, then publish it

Press Execute workflow, fill in the test form again and check Gmail. You should find a draft with the client’s address, your subject and a reply that sounds roughly like you.

Then press Publish and switch the form to its Production URL. This is the step beginners miss. n8n’s docs explain that the Test URL only works while you are testing in the editor, and the Production URL only works once the workflow is published (n8n docs: n8n Form Trigger). Put the production link on your website or in your email signature.

After it runs for real, open the Executions tab of the workflow to see every run and what each node did (n8n docs: Executions). That tab is where you find out why something did not happen.

What the finished workflow looks like

Read it left to right, like a sentence:

Form submitted  ->  OpenAI sorts it and drafts a reply  ->  Edit Fields tidies it  ->  Gmail saves a draft

Picture a one-person design studio. Before: enquiries pile up until Sunday night. After: the first draft is already waiting, and answering takes a coffee, not an evening. That is the whole point. Not magic, just one boring step gone.

Five beginner mistakes that cost the most time

1. Forgetting to publish. The workflow works when you test it, then “does nothing” all week. Check that it is published, and that you use the Production URL.
2. Building the dream workflow first. Start with four nodes that solve one small problem. Add branches later, when the simple version has run for real.
3. Typing data instead of dragging it. Dragging fields in builds the expression for you and shows a preview. Typing by hand is where typos break things.
4. Letting AI send on its own too early. Draft first, send later. One wrong automatic email to a client costs more than all the time you saved.
5. Never looking at Executions. When something goes wrong, the Executions tab shows the exact node and the exact data. Look there before you rebuild anything.

Where to go after your first two workflows

Once these two run, you know enough to copy almost any n8n template and understand it. Good next steps:

• Keep your CRM up to date on its own: n8n CRM automation for small business has four workflows that build on exactly the skills above.
• Stop typing invoices into a spreadsheet: n8n workflow for automatic invoice processing takes Gmail attachments into Google Sheets.
• Turn the AI step into a real agent: how to build an AI agent without code shows when you need an agent and when a simple AI step like ours is enough.
• Chase replies automatically: automate follow-up emails without code.

Want structured lessons from n8n itself? The free n8n Academy has beginner courses, starting with “Essentials: Your First Workflows”. And the n8n template library is a great place to steal ideas once you know how to read a workflow.

Got your first two workflows running? The next step is an agent that decides which tool to use. Our n8n AI agent workflow guide walks you through building a meeting-prep agent that reads your calendar and email and drafts a short prep note for you to check.

FAQ

Do I need to know how to code to use n8n?

No. n8n’s own first-workflow tutorial says you need no coding experience, and everything in this guide is drag and drop. n8n does let you add code steps later if you ever want to, but you can get real work done without them.

Is n8n free?

The self-hosted Community edition is free. n8n Cloud has a free trial without a credit card, then paid plans starting at 20 euro a month billed annually, as shown on the n8n pricing page today. Prices change, so check that page before you decide.

How long does it take to learn n8n?

The official first workflow takes about 15 minutes, according to n8n. Plan about an hour for the AI workflow in this guide, including connecting Gmail. In my experience, feeling at home with branches and error handling takes a few weeks of building real things, not a weekend of videos.

Is n8n better than Zapier for a beginner?

In my experience, Zapier is a little quicker for a simple two-step automation. n8n shines once your workflow has more steps or an AI part, because you pay per full run instead of per step. Our Zapier free plan examples show what the other side looks like.

Can n8n use Claude or Gemini instead of OpenAI?

Yes. n8n has its own nodes for Anthropic (Claude) and Google Gemini, next to OpenAI. Swap the OpenAI node for the one you prefer and keep the same prompt.

Build yours this week

Pick one message you answer over and over. Build the four-node workflow tonight, run it as a draft-only helper for two weeks, and see how much of your inbox quietly gets lighter. If you want more small, practical wins like this one, every week, it is free:

Sources

• n8n docs: Build your first workflow
• n8n docs: Choose how to use n8n
• n8n pricing (checked 22 October 2026)
• n8n docs: Host n8n
• n8n docs: n8n Form Trigger
• n8n docs: OpenAI node
• n8n docs: Anthropic node
• n8n docs: Gmail node
• n8n docs: Executions
• n8n Academy

You want an AI agent that answers you at 7am on a Saturday, runs your Friday report while your laptop is closed, and does not die every time you shut the lid. That is why so many people put OpenClaw on a Mac mini.

The short answer: give the Mac mini its own user account, stop it from sleeping, turn on automatic login and Remote Login, install OpenClaw as a background service, then reach it from your laptop or phone over Tailscale. Restart it once to prove it comes back on its own. Plan on about an hour.

Want short, practical AI lessons like this one? Join Launchpad free, right above.

Why put OpenClaw on a Mac mini at all?

OpenClaw has one part that has to stay awake: the Gateway. It holds your settings, your chats and your connections to Telegram, WhatsApp or Slack. If it runs on your laptop, your agent sleeps when your laptop sleeps.

OpenClaw’s own docs give three ways to keep it on all the time (OpenClaw docs: macOS VMs):

• A small Linux server (VPS). Cheap and always on.
• Dedicated hardware, like a Mac mini. Full control, and a home internet address. The docs point out that many sites block data center addresses, so browser jobs often work better from home.
• A mix. The Gateway on a server, your Mac connected as a helper when a job needs a real screen.

The Mac mini wins when you want three things: macOS-only features such as iMessage, browser jobs that do not get blocked, and a box you own instead of rent. If none of those matter to you, a VPS is the simpler choice. That is an honest trade, not a sales pitch.

Still deciding which model to buy? Our guide to OpenClaw Mac mini specs covers how much memory, chip and storage you really need, so you do not overspend.

Before you start: give the agent its own account

This is the step most people skip, and it is the one that keeps your personal life out of the agent’s reach.

OpenClaw is built as a personal agent for one trusted person. Its security guide works from the idea of one trust boundary per Gateway (OpenClaw docs: Security). On a Mac mini that means:

• Create a separate macOS user for OpenClaw, for example “agent”. Do not run it in the account where your photos, passwords and work files live.
• If you plan to use iMessage, sign that user in with its own Apple Account. The docs describe exactly this setup, a dedicated macOS user with a dedicated Apple ID, so bot messages stay apart from your personal Messages (OpenClaw docs: iMessage deployment patterns).
• Only give it the files it needs. Put those in its workspace folder, ~/.openclaw/workspace, instead of pointing it at your whole drive.

Step 1: Set up macOS so the Mac mini never goes dark

A Mac mini that sleeps is just a quiet laptop. Do these four things while you still have a screen and keyboard plugged in.

Stop it from sleeping

Open System Settings > Energy. Turn on Prevent automatic sleeping when the display is off. If you see Wake for network access, turn that on too. Apple notes that keeping a Mac awake can use more power, which on a Mac mini is the whole point (Apple: Set sleep and wake settings).

Log in automatically after a restart

The OpenClaw Gateway runs as a LaunchAgent, a macOS background service that starts when a user logs in. No login after a restart means no agent. So go to System Settings > Users & Groups and set Automatically log in as to the agent account. OpenClaw’s docs give the same advice for headless Macs (OpenClaw docs: macOS VMs).

One catch: if FileVault is on, macOS does not allow automatic login (Apple: Log in automatically to a Mac user account). That is a real choice. FileVault encrypts the disk, which matters if the Mac mini could be stolen. Without it, you get a box that recovers from a restart on its own. Pick based on where the machine sits. A locked home office is a different risk from an open shop floor.

Turn on Remote Login (SSH)

Go to System Settings > General > Sharing and turn on Remote Login. Now you can type commands on the Mac mini from your laptop (Apple: Allow a remote computer to access your Mac). Under the small info button, allow only the agent account and your own.

Turn on Screen Sharing

Same Sharing screen, turn on Screen Sharing (Apple: Turn screen sharing on or off). You will need it. Some macOS permission prompts only appear on screen, and the docs warn that the first run of some features asks for approvals in that user’s desktop session. With Screen Sharing on, you click “Allow” from your laptop instead of digging out a monitor.

Step 2: Install OpenClaw as a background service

Log in to the agent account. The install itself is the same as on any Mac, and we cover it step by step in how to install OpenClaw on Mac. Here is what is different on a Mac mini that has to run on its own.

Terminal route (best for a Mac mini). Paste the installer:

curl -fsSL https://openclaw.ai/install.sh | bash

It installs Node if needed, installs OpenClaw and starts the setup. Quick start keeps the Gateway running inside that Terminal window. On a machine nobody sits at, that is not good enough. Press Control-C, then run:

openclaw gateway install
openclaw gateway status

That creates the LaunchAgent. The docs say launchd then starts the Gateway at login, restarts it after a crash and keeps one predictable log (OpenClaw docs: Gateway on macOS).

App route. If the Mac mini runs macOS 26.2 or later and you prefer the menu bar app, that works too. But change one setting: in the dashboard’s device settings, turn on Keep OpenClaw running when the app is closed, next to Launch at login. It is off by default, so without it the agent stops when the app quits.

Connect your AI model during setup. OpenClaw itself is free; the model behind it is what you pay for, through a ChatGPT or Claude login or an API key.

Step 3: Reach your Mac mini from your laptop and phone

The Gateway listens on port 18789, and by default only on the Mac mini itself (OpenClaw docs: Remote access). Leave it that way. Do not open that port on your router.

The calm way to reach it from anywhere is Tailscale, a free private network between your own devices. Install it on the Mac mini and your laptop, then tell OpenClaw to use Tailscale Serve. The Gateway stays private on the Mac mini, and Tailscale gives you an HTTPS address that only your own devices can open (OpenClaw docs: Tailscale):

openclaw gateway --tailscale serve

Skip Funnel mode unless you know why you need it. Funnel puts the dashboard on the public internet behind a password.

On your laptop, the OpenClaw Mac app can act as a remote control for the Mac mini. Choose Connect to an existing Gateway on first launch, or add it later under Connection. It connects directly over Tailscale or opens an SSH tunnel for you (OpenClaw docs: Remote control).

On your phone, the quickest channel is Telegram. New people who message your bot get a pairing code, and nothing they send runs until you approve them (OpenClaw docs: Pairing). That is what you want on an agent that is online all day.

Step 4: Prove it survives a restart

Most “always-on” setups fail the first time the power blips. Test it now, while you are paying attention.

1. Restart the Mac mini from the Apple menu.
2. Do not touch it. From your laptop, wait a couple of minutes, then SSH in and run openclaw gateway status. You want to see the Gateway listening on port 18789.
3. Send your agent a message from your phone. If it answers, the chain works: power on, auto-login, LaunchAgent, Gateway, model.
4. Run openclaw security audit. It flags settings that drifted away from the safe defaults (OpenClaw docs: Security). Fix what it finds before you add more channels.

If step 2 fails, it is almost always auto-login. Check that FileVault is off and that the agent account, not yours, is the one set to log in.

What to give your Mac mini agent first

A box that runs all night is only worth it if it does real work. Start with one job you repeat every week and that does not need you in the room:

• A Monday morning brief. “Every Monday at 7, read the notes in my workspace folder from last week and send me five bullet points on Telegram.”
• An inbox sorter. Have it sort incoming requests into “needs me”, “can wait” and “info only”, and send you only the first group.
• A browser check from home. Because the Mac mini uses your home connection, it can open sites that often block server addresses, like checking a supplier’s page for a price change.

Keep the first job small. Watch it run for a week. Then give it the second one. That is how you end up with an agent you trust, instead of one you babysit.

Quick answers

Do I need a monitor on the Mac mini?

Only for the first setup. After that, Remote Login and Screen Sharing let you do everything from your laptop, including clicking “Allow” on permission prompts.

Mac mini or VPS for OpenClaw?

A VPS is cheaper to start and has nothing to plug in. A Mac mini gives you iMessage, a home internet address for browser jobs, and hardware you own. OpenClaw’s docs list both as good defaults, so pick by the jobs you want it to do.

Can the Mac mini run the AI model itself?

It can, through tools like Ollama or LM Studio, and OpenClaw can pick up a model that is already loaded there. A cloud model is still the easier start. See the local model part of our OpenClaw Mac install guide before you decide.

Should I use my work Mac mini?

Ask first. A machine your company manages may block automatic login or SSH, and an agent with access to work files is something your IT team will want to know about.

Is OpenClaw the right tool for this?

If you want an agent you message from your phone that runs on your own hardware, yes. If you mostly want help writing code at your desk, read OpenClaw vs Claude Code first. On Windows instead? Here is how to install OpenClaw on Windows.

Sources

Every step above comes from OpenClaw’s own documentation and Apple’s support pages:

• OpenClaw docs: macOS VMs (always-on options, auto-login, running 24/7)
• OpenClaw docs: Gateway on macOS (LaunchAgent)
• OpenClaw docs: Remote access
• OpenClaw docs: Tailscale
• OpenClaw docs: Remote control from the Mac app
• OpenClaw docs: iMessage deployment patterns
• OpenClaw docs: Security
• Apple: Set sleep and wake settings for your Mac
• Apple: How to log in automatically to a Mac user account
• Apple: Allow a remote computer to access your Mac
• Apple: Turn Mac screen sharing on or off

You asked one simple work question in three tabs. ChatGPT gave you a confident answer, Claude gave you a careful one, Perplexity gave you five links. Now you are not sure which one to trust, or which one deserves your money.

The short answer. In the ChatGPT vs Claude vs Perplexity choice, you are really choosing between two kinds of tool. Perplexity is for finding out: it searches the web and puts a source on every answer, so you can check it. ChatGPT and Claude are for making things: the email, the report, the plan. Both of them can search the web too now, but writing is where they shine. For most people the best setup is one writer you like, plus Perplexity when the facts have to be right. Below: the side by side (prices checked on 22 October 2026), where each one wins, a 20-minute demo on a real work job, and what to pay for.

Want one practical AI lesson like this every week?

ChatGPT vs Claude vs Perplexity at a glance

Three companies, three products. ChatGPT is made by OpenAI, Claude by Anthropic, Perplexity by Perplexity. This is what their own plan and help pages say right now (ChatGPT pricing, Claude pricing, Perplexity pricing).

	ChatGPT	Claude	Perplexity
Built for	An all-round assistant: writing, files, everyday tasks	Writing, thinking things through, documents	Answers from the web, with citations in every answer
Free plan	Yes, $0, web search included	Yes, $0, with web search, file creation and code	Yes, $0, limited daily use, citations in every answer
Cheapest paid plan	Go, $8 a month	Pro, $17 a month billed yearly ($200 up front), or $20 monthly	Pro, $17 a month billed yearly (the page shows only the yearly price)
Biggest personal plan	Pro, $100, $200 or $500 a month	Max, from $100 a month	Max, $167 a month billed yearly
Web search with sources	On every plan, answers may include citations	On every plan, search answers include citations	The whole product is built around it
Pick the AI model	OpenAI models	Claude models	Pro and Max can choose a preferred model, including OpenAI, Google and Anthropic ones

That last row surprises people. Perplexity says it picks the best model per question from a mix that includes ChatGPT, Gemini and Claude, and Pro users can choose one themselves (Perplexity pricing FAQ). So it is less “Perplexity instead of ChatGPT” and more “a research desk that can use the same brains”.

Where Perplexity is the better pick

Use Perplexity when the job starts with “find out”. What does that tool cost now? What changed in this rule this year? Which three vendors do this, and what do users complain about?

• Every answer shows its sources. Perplexity searches the web in real time and adds citations and links to the original pages, so you can check a claim before you repeat it in a meeting (What is Perplexity?).
• Pro Search digs deeper. It runs several searches across the web, compiles what it finds, and shows you how it broke your question down, so you can spot where it went off track (What is Pro Search?).
• Projects keep research together. A Project holds your searches, files and instructions in one place, and it is private by default (What are Projects?).

The catch. Perplexity is not built to read a long document word for word. Its own help page says that for a long file it “will extract the most important parts”, files are capped at 40 MB, and uploads come from a weekly allowance on Free, Pro and Max (File Uploads). For a contract where every clause matters, use a writer tool and the method in our guide to analyzing long PDF files with Claude.

Where ChatGPT or Claude is the better pick

Use ChatGPT or Claude when the job starts with “make”. A reply to a tricky client email. Meeting notes turned into a report. A one-page plan from a messy brain dump. This is where you want a tool that holds your tone, takes feedback (“shorter, warmer, no jargon”) and rewrites without fuss.

Both can search the web now. ChatGPT web search works on every plan, including Free (Searching the web with ChatGPT). Claude can search too, and its search answers come with citations (Enable and use web search). So you do not need Perplexity for a quick lookup in the middle of writing.

One line from OpenAI’s own help page is worth taping to your screen: search results and citations “can be incomplete, outdated, or incorrect”, so open a cited source and check it supports the answer (OpenAI Help Center). That goes for all three tools, Perplexity included. A citation is a place to check, not a guarantee.

Choosing between ChatGPT and Claude themselves? That is a separate decision with its own trade-offs, and our ChatGPT vs Claude AI comparison for work walks through it plan by plan. The short version: ChatGPT has the cheaper $8 step, Claude Pro is $17 a month on a yearly plan, and Claude’s usage resets on a rolling five-hour window with weekly limits on paid plans (Claude pricing).

The 20-minute demo: research in Perplexity, write in ChatGPT or Claude

Here is the split in action, on a job most of us know. Tomorrow you have a call about switching your team to a new tool (say, an invoicing app). You want one page: what it costs, what users complain about, and three sharp questions to ask.

Step 1. Research in Perplexity (8 minutes). Paste this, with your tool’s name:

I have a call tomorrow about switching my team of 8 to [tool name] for invoicing.
Find, with a source for each point:
1. The current price for a team of 8, from the official pricing page
2. The 3 most common complaints from real users in the last 12 months
3. Anything that changed in the product or pricing this year
If you cannot find a source for a point, say so. Do not guess.

Step 2. Check the sources (4 minutes). Open at least three of the links. Does the page really say what the answer says? Is the price on the vendor’s own page, and is the complaint recent? Cross out anything you cannot confirm. This step is the whole reason you used Perplexity, so do not skip it.

Step 3. Write in ChatGPT or Claude (6 minutes). Paste only the checked facts, with their links:

Write a one-page brief for my call tomorrow, using ONLY the facts below.
Sections: what it costs us, the 3 risks, 3 questions to ask them.
Plain language, short sentences, under 300 words.
Keep the source link next to each fact.
If something important is missing, list it at the end under "Still to find out".

Facts I checked:
[paste your checked points and links]

Step 4. Read it once and cut (2 minutes). Delete anything that sounds smart but says nothing. Add one line in your own words about what you actually want from the call.

The result. A brief where every number has a link you opened yourself, written in a tone you can send to your manager. That is the split we use when we write the guides on this site: sources first, checked by hand, then the writing. It is slower than asking one chatbot, and much safer.

Which one for which job

The job	Start in	Why
Check a price or a fact before a meeting	Perplexity	A source on every answer, so you can check it fast
Shortlist tools or vendors	Perplexity, then a writer	Find and check in Perplexity, write the comparison in ChatGPT or Claude
Reply to a tricky email	ChatGPT or Claude	Tone and rewrites matter more than search. Our prompts to make an email sound more professional work in both
Read a long report or contract	Claude or ChatGPT	You want the whole document read, not the highlights
Turn research into a report	ChatGPT or Claude	See how to get ChatGPT to help write a report
Today’s news in your field	Perplexity, or Grok	Fresh sources matter. Our ChatGPT vs Gemini vs Grok guide covers the news side

Privacy: what to check before you paste work in

Before any client name or company number goes into any of these tools, check your company’s AI policy. Then flip the right switch.

• Perplexity. In Settings, under Preferences, there is an “AI data retention” switch that controls whether your data is retained. Enterprise customers are excluded from model-provider training automatically (Account Settings). Perplexity also says its agreements with OpenAI and Anthropic forbid those companies from training on Perplexity data (Are third-party model providers training on my data?).
• ChatGPT and Claude. Both have their own training switches. We show where they are, step by step, in the privacy section of our ChatGPT vs Claude AI guide.

So which one should you pay for?

• Most professionals: pay for one writer (ChatGPT or Claude), and use Perplexity Free for the moments when facts must be right. That covers a normal week.
• Research-heavy work (consultants, analysts, marketers who check sources all day): Perplexity Pro, plus the free plan of the writer you like best.
• Everyone: do not buy all three on day one. Use the free plans for a real week, and pay only where you hit a limit twice.

If you also live in Gmail or Word, the assistants built into those apps are worth a look. Our ChatGPT vs Gemini vs Copilot guide covers that side.

If most of your research is about markets, customers or competitors, our guide on how to use AI for market research walks through a full workflow with fact checks.

Quick answers

Is Perplexity better than ChatGPT?

For finding facts with sources, it is the more focused tool: citations are on every answer. For writing and rewriting, ChatGPT and Claude are the better fit. And ChatGPT can search the web on every plan too, so the gap is smaller than it was.

Does Perplexity use ChatGPT or Claude?

Yes, in a way. Perplexity says it orchestrates several top models, including ChatGPT, Google Gemini and Anthropic Claude, and picks one per question. On Pro and Max you can choose your preferred model (Perplexity pricing FAQ).

Can I use all three for free?

Yes. Each one has a $0 plan with limits. For a fair test, use the same job in all three, like the demo above.

Which is cheapest if I want to pay a little?

ChatGPT Go at $8 a month is the lowest paid step of the three (ChatGPT pricing). Claude Pro and Perplexity Pro both work out at $17 a month on a yearly plan.

Can Claude search the web like Perplexity?

Yes. Claude can search the web, and its search answers include citations you can click (Claude Help Center). Perplexity is still the one built around research first.

Want the next one in your inbox?

Sources

• Perplexity pricing and plan FAQ (Perplexity), checked 22 Oct 2026
• What is Perplexity? (Perplexity Help Center)
• What is Pro Search? (Perplexity Help Center)
• File Uploads (Perplexity Help Center)
• What are Projects? (Perplexity Help Center)
• Account Settings (Perplexity Help Center)
• Are third-party model providers training on my data? (Perplexity Help Center)
• ChatGPT plans and pricing (OpenAI), checked 22 Oct 2026
• Searching the web with ChatGPT (OpenAI Help Center)
• Claude plans and pricing (Anthropic), checked 22 Oct 2026
• Enable and use web search (Claude Help Center)

You have ChatGPT open in one tab and Claude in the other. The same email is half-written in both, and you still cannot tell which one deserves your $20 a month. Here is the plain answer, checked against both official plan pages on 21 October 2026.

The short answer. In the ChatGPT vs Claude AI choice, both are good enough for everyday office work, so the deciding factor is what fills your day. Pick ChatGPT if you want one tool that also makes images, does voice with video, and comes with a cheaper $8 plan. Pick Claude if your day is long documents, reports and careful writing, and you want a very large reading window on every plan. Both start free. Both main paid plans cost $20 a month. Below you get the side-by-side, the catch with each one, the privacy switch to flip, and a 30-minute test on your own work.

Want one practical AI lesson like this every week?

ChatGPT vs Claude AI at a glance

ChatGPT is made by OpenAI. Claude is made by Anthropic. They are separate products from separate companies, with their own plans. This is what each plan page lists right now (ChatGPT pricing, ChatGPT plan prices, Claude pricing).

	ChatGPT	Claude
Free plan	Yes. Unlimited text chats within abuse guardrails, limited uploads, images, voice and deep research	Yes. Chat, web search, files, code, memory, connectors, up to 5 projects
Cheapest paid plan	Go, $8 a month (may include ads)	Pro, $20 a month, or $17 a month billed yearly
Standard paid plan	Plus, $20 a month	Pro, same as above
Heavy-use plan	Pro, from $100 a month ($100, $200 or $500 tiers)	Max, from $100 a month (5x or 20x Pro usage)
Makes images	Yes, on every plan (limited on Free)	No image generator listed. Claude Design builds mockups, one-pagers and slides instead
Voice	Yes. Voice with video from Go up	Voice mode on every plan
Office apps	Excel, Word, PowerPoint and Google Sheets extensions (limited on Free and Go)	Excel, Word, PowerPoint and Outlook from Pro. Google Docs, Sheets and Slides add-on in beta
How much it reads at once	About 12 pages on Free, 40 on Plus for the default model, about 320 for reasoning models on Go and Plus	Context window up to 1 million tokens, depending on the model, on every plan
Deep research	Limited on Free and Go, included on Plus	Research from Pro
Your chats used for training	You can opt out	You can opt out

Two things this table makes clear. ChatGPT has more plan steps and more extras. Claude has fewer extras but can read a lot more in one go, on every plan.

Where ChatGPT is the better pick

• You need pictures. Social posts, a header image, a quick diagram for a slide. ChatGPT creates images on every plan, with more and better ones on paid plans. Claude’s plan pages list no image generator.
• You want to talk to it. Voice is on every ChatGPT plan, and from Go you can share video while you talk. Handy when you want to point your phone at a whiteboard and ask what to do next.
• You work in spreadsheets. ChatGPT lists data analysis on every plan (limited on Free) and interactive tables and charts on Plus. Our guide on analyzing a spreadsheet with ChatGPT shows the prompts.
• You want a cheap step up. The $8 Go plan gives more messages, uploads, images and memory than Free. Claude has nothing between free and $20.
• You want it to run things on a schedule. Scheduled tasks start on Go, and projects, scheduled tasks and custom GPTs are all on Plus.

Where Claude is the better pick

• Your work is long documents. Contracts, tender documents, a 90-page report from head office. Claude lists a context window of up to 1 million tokens on every plan, while ChatGPT Plus lists 256K for its reasoning models. If that is your daily work, start with our guide to analyzing long PDF files with Claude.
• You want edits you can accept one by one. In Google Docs, Claude shows its changes as suggestions you apply or dismiss, and in Sheets it highlights the cells it touched (Claude for Google Workspace). That add-on is still in beta.
• You live in Outlook. Claude for Microsoft Outlook is included from Pro, next to Word, Excel and PowerPoint.
• You need a clean one-pager, not a picture. Claude Design turns a proposal into a one-pager or mockup you can export to PDF or PowerPoint (Claude Design).
• You want to try coding without being a developer. Claude Code is included in every paid Claude plan. Our ChatGPT vs Claude for coding guide compares the two for people who do not write code for a living.

The catch with each one

Claude’s catch is usage limits. Every Claude plan has limits that reset on a rolling five-hour window, and paid plans add weekly limits on top. There is no fixed message count: long chats, big files and the strongest models use it up faster. Chats, the desktop app, mobile and Claude Code all draw from the same pool. On paid plans you can turn on usage credits to keep going at API rates (Claude pricing FAQ). If you fire off fifty quick questions a day, you will feel this.

Wondering what actually happens when you hit the cap on either one? Our guide to ChatGPT vs Claude usage limits walks through it, plus how to plan your day around it.

ChatGPT’s catch is the small print on the cheap plans. The free plan’s default model takes in about 12 pages of text at once. The newest models, like GPT-6.1 Sol and GPT-6 Astra, start on Plus. And the $8 Go plan may show ads. None of that is a problem for short emails. It is a problem if you paste a 60-page report into the free version and wonder why it missed the second half.

Privacy: keep your work chats out of training

Do this before you paste anything from work, whichever tool you pick.

• ChatGPT: go to Settings, then Data controls, and turn off “Improve the model for everyone”. Your new chats are then not used to train OpenAI’s models, and your history stays where it is. For one-off sensitive questions, use a Temporary chat: it is not saved to your history, does not update memory and is not used for training, though OpenAI may keep it for up to 30 days for safety (OpenAI data controls).
• Claude: Anthropic uses your chats to improve Claude only if you allow it in your Privacy Settings, if a chat is flagged for a safety review, or if you join a testing program. Incognito chats are never used for training. One thing people miss: a thumbs up or down stores the whole conversation for up to 5 years (Anthropic privacy center).
• Company plans: ChatGPT Business and Enterprise, and Claude Team and Enterprise, do not train on your content by default. If your employer pays for one of those, use that account for client work.

And the rule that beats any setting: take client names, salaries and personal details out before you paste, unless your company has approved the tool for that data.

The 30-minute test: four jobs from your own week

Comparison tables only get you so far. The answer that counts is which one does your work better. Open both free plans, then give them the exact same four jobs. Take about seven minutes per job.

Job 1. An email you have been avoiding.

Here is an email I need to answer. Write a reply that is friendly but firm, under 120 words, and ends with one clear next step. Do not promise anything I did not mention in my notes.

Email: [paste the email]
My notes: [two or three lines on what you want to say]

Job 2. A long document. Use the longest PDF you actually have to read this month.

Read the attached document. Give me:
1. The 5 points that matter most for someone in my role: [your role]
2. Every deadline, amount or commitment, with the page number
3. Anything that looks unclear or risky, with the page number
If something is not in the document, say so. Do not guess.

Job 3. A spreadsheet question. Upload a sheet you know well, so you can spot a wrong answer.

This file has [what is in it]. Tell me the three biggest changes between [period A] and [period B], show how you calculated each one, and make one simple chart I could paste into a slide.

Job 4. A one-page plan.

Turn these notes into a one-page plan with: goal, three steps, owner per step, deadline per step, and the first thing to do tomorrow morning. Keep my wording where you can.

Notes: [paste your notes]

Score each answer from 1 to 5 on three questions: Did it invent anything? How much did I have to fix? Would I send or use this today? Check the numbers in Job 3 yourself. The tool with the higher total on your jobs is your answer, whatever any review says. If the email job is where you spend most time, our guide on making an email sound more professional has more prompts to test with.

Which one for which job

Your job	Try first	Why
Quick emails and replies all day	ChatGPT	Unlimited text chats on every plan, within abuse guardrails
Long reports, contracts, tenders	Claude	Very large reading window on every plan
Social posts with images	ChatGPT	Creates the image in the same chat
Proposal one-pagers and slides	Claude	Claude Design exports to PDF and PowerPoint
Spreadsheet number crunching	ChatGPT	Data analysis on every plan, interactive charts on Plus
Editing inside Google Docs	Claude	Suggested edits you accept or dismiss (beta)
Meeting notes to action items	Either	Both do this well. See summarizing meeting notes
Small scripts and automations	Claude	Claude Code is included from Pro

So which one should you pay for?

• Tight budget, mostly short tasks: ChatGPT Go at $8.
• One tool for everything, including images and voice: ChatGPT Plus at $20.
• Mostly long documents and careful writing: Claude Pro at $20, or $17 a month if you pay for the year.
• Not sure yet: run the four-job test on both free plans first. It costs nothing.

Pay for one, for one month, on real work. If you keep opening the free version of the other one for certain jobs, that is your answer too: some people end up paying for both and splitting the work. Our page on what Reddit users say about ChatGPT vs Claude shows how heavy users split it. Weighing Gemini or Copilot as well? See ChatGPT vs Gemini vs Copilot. New to Claude? The Claude tutorial for beginners gets you to your first work prompt in about 30 minutes.

If a lot of your work is finding facts you can check, not just writing, our ChatGPT vs Claude vs Perplexity comparison shows where Perplexity fits next to these two.

If your other option is Gemini because your work runs on Google, our Claude vs Gemini Pro guide shows which paid plan fits which job.

Leaning toward ChatGPT but never really used it? Our ChatGPT tutorial for beginners takes you from your first chat to real work tasks.

Already chose Claude? See how to use Claude for work to connect it to your email, calendar and files and hand it the first real job from your week.

Quick answers

Is Claude better than ChatGPT?

Not across the board. Claude is the stronger pick for long documents and careful writing, ChatGPT for images, voice, spreadsheets and lots of quick questions. Your own four-job test is the tiebreaker.

Is Claude the same as ChatGPT?

No. Claude is made by Anthropic and ChatGPT by OpenAI. They work in a similar way, a chat window where you type or upload files, but they run on different models with different plans, limits and features.

Which is cheaper, ChatGPT or Claude?

Both have free plans and both main plans cost $20 a month. ChatGPT also has Go at $8 a month. Claude Pro drops to $17 a month if you pay $200 for the year up front.

Can Claude make images like ChatGPT?

Claude’s plan pages do not list an image generator. It can build visual layouts like one-pagers, mockups and slides with Claude Design. For pictures and illustrations, ChatGPT is the one to use.

Can I use both for free?

Yes. Both have free plans with tighter limits, which is enough for the four-job test above.

Torn between ChatGPT and Gemini instead? See ChatGPT vs Gemini Pro: which $20 plan to pay for at work.

Want the next one in your inbox?

Sources

• ChatGPT plans and pricing, feature comparison (OpenAI), checked 21 Oct 2026
• ChatGPT Work and Codex pricing, plan prices (OpenAI), checked 21 Oct 2026
• Data controls in ChatGPT (OpenAI Help Center)
• Claude plans, pricing and usage FAQ (Anthropic), checked 21 Oct 2026
• Is my data used for model training? (Anthropic Privacy Center)
• Claude for Google Workspace (Anthropic)
• Claude Design (Anthropic)

You keep seeing the same three names. ChatGPT is the one everybody already uses. Gemini keeps popping up inside Gmail. And Grok is now in your X app, with a new subscription that bundles it with X Premium and Cursor. You have one workday, a pile of real tasks, and no time to test three chatbots properly.

Here is the short answer. In the ChatGPT vs Gemini vs Grok question, the winner depends on what kind of work you hand it. ChatGPT is the strongest all-rounder for writing, thinking through messy problems and working with files. Gemini fits best when your work already lives in Gmail, Docs and Drive, and its paid plan handles very long documents. Grok’s edge is freshness: it searches the web and X live, so it is the one to open when you need to know what happened today in your field. All three have a free version. Start free, run the 15-minute test further down this page, and only pay for the one that wins on your own tasks.

Want one practical AI lesson like this every week?

ChatGPT vs Gemini vs Grok at a glance

Benchmark charts change every month and say little about your desk. These are the facts from each company’s own pages, checked on 21 October 2026.

• Made by: ChatGPT by OpenAI, Gemini by Google, Grok by SpaceXSI (the company behind X).
• Free version: all three. ChatGPT Free gives unlimited everyday text chats, with limits on uploads, deep research and images (ChatGPT pricing). Gemini Free includes Deep Research, Canvas, Gems and Gemini Live (Google AI plans). Grok Free gives limited chat and agent requests (X Subscription pricing).
• Cheapest paid step: ChatGPT Go at $8 a month (ChatGPT Learn pricing), Google AI Plus at $4.99 (Google One), X Subscription Plus at $8.
• The plan most people end up on: ChatGPT Plus at $20 a month, Google AI Pro at $19.99 a month, and on the Grok side X Subscription Premium at $30 a month, which is the tier with the latest Grok models.
• Lives best in: ChatGPT in its own app on web, desktop and phone, with extensions for Excel, Word, PowerPoint and Google Sheets. Gemini in the Gemini app and inside Gmail and Docs on paid plans. Grok on grok.com, in the Grok apps and in the Grok tab on X.
• Stand-out strength: ChatGPT for writing and working through files. Gemini for very long documents and Google apps. Grok for live web and X search on what is happening right now.

Prices are US list prices without tax. In euros or pounds the numbers differ, so check the page for your country before you pay.

ChatGPT: the all-rounder for writing and files

If you only want to learn one tool this month, ChatGPT is the safest choice. It is the one most colleagues and tutorials assume you use, which makes it easier to get help.

• What the free version does well: drafting emails, turning messy notes into a clear summary, explaining a contract clause in plain words. Text chats are unlimited on Free, subject to OpenAI’s usage rules.
• What Plus adds at $20: projects, scheduled tasks, custom GPTs, expanded deep research and uploads, and full access to the Excel, Word, PowerPoint and Google Sheets extensions (ChatGPT pricing).
• Good to know: the Go plan may show ads, according to OpenAI’s pricing page.

Typical jobs to hand it: a reply to a difficult client, a first draft of a report, or a spreadsheet you want explained. We have step-by-step guides for analyzing a spreadsheet with ChatGPT and summarizing long documents.

Gemini: the one already sitting in Gmail and Docs

If your company runs on Google Workspace, or your own inbox is Gmail, Gemini has a home advantage. Less copy and paste means you actually use it.

• Free: the Gemini app with Deep Research, Canvas, Gems and Gemini Live.
• Google AI Plus ($4.99): higher limits and Gemini in Gmail and other Google apps.
• Google AI Pro ($19.99): Gemini in Gmail, Docs and more, a 1 million token context window, and file uploads of up to 1,500 pages, plus 5 TB of storage (Google AI plans, Google One).

That big upload limit is the practical point. If you regularly need to read a 300-page tender or a year of meeting minutes, Gemini Pro is built for that size. If you already pay for Google storage, check your plan first: you may have more Gemini than you think. For a worked example, see how to use Gemini for automated status updates.

Grok: the one that knows what happened this morning

Grok is the newest name on most people’s list, and the one with the most confusing pricing right now. Here is what matters for work.

• Live search: Grok searches the web and X live and shows citations, so answers can include posts and news from today (Grok product page).
• The everyday tools are there too: file and PDF analysis, screenshots, a canvas for long writing, custom instructions, image generation and short videos.
• No X account needed: you can sign in with X or with an email address, on grok.com or in the iOS and Android apps.

That is how I use it myself. When a new AI tool launches, the first real reactions, screenshots and bug reports show up on X within the hour, long before the articles. A Grok search with X included gets you to those posts fast, with links you can open and check.

Where it is weaker for office work: it does not sit inside Word or Gmail the way ChatGPT’s extensions and Gemini do. You bring the work to Grok, not the other way round.

The new X Subscription, in plain words

In October 2026 SpaceXSI combined Grok, Grok Bot, Cursor and X Premium into one plan called The X Subscription (SpaceXSI announcement). The tiers are Free, Plus at $8, Premium at $30, Super at $100 and Ultra at $200 a month (SpaceXSI docs).

• Plus ($8): Grok Chat, Voice, Imagine and Build, plus X perks like the checkmark. It does not include Grok Bot.
• Premium ($30): the latest Grok models, faster responses, and Grok Bot, the AI teammate that works on multi-step tasks for you.
• Already pay for SuperGrok or X Premium? Your old plan keeps running. Switching is optional, and if you switch you get a prorated refund for unused time.

So the honest comparison with ChatGPT Plus and Google AI Pro is the $30 Premium tier, not the $8 one. Whether that is worth it depends on whether you also use X Premium or Cursor. If you are curious about the Bot part, we wrote about how Grok Bot picks a model per task.

Privacy: what happens to what you paste

This part matters more than any feature list. On personal plans, all three can use your chats to improve their models unless you change a setting.

• ChatGPT: turn off “Improve the model for everyone” under Settings, then Data controls. Temporary chats are not used for training. Business and Enterprise workspaces are not used for training by default (OpenAI data controls FAQ).
• Gemini: while Keep Activity is on, Google uses your chats to improve its AI and some are read by human reviewers. Google itself asks you not to enter confidential information you would not want a reviewer to see. Temporary chats are not used for training (Gemini Apps Privacy Hub).
• Grok: in the Grok app go to Settings, Data Controls, and switch off “Improve the model”. On grok.com it is under Settings, Data. Private Chat is not used for training, and content from business and enterprise customers is not used either. If you use Grok inside the X app, X has its own separate setting (SpaceXSI consumer FAQ).

A simple rule that keeps you out of trouble: client data, contracts and anything about colleagues only go into the version your company approved. Everything else, test wherever you like.

The 15-minute fresh-news test

Reviews tell you which chatbot is “best”. Your own work tells you which one is best for you. Because Grok’s selling point is live information, this test checks exactly that, on the free versions.

1. Pick one thing that changed in your field this week. A new product launch, a price change from a supplier, a software update your team relies on. Something you can check yourself.
2. Paste the same prompt (below) into ChatGPT, Gemini and Grok. Use temporary or private chat if you like.
3. Score each answer on three things: did it find the news from this week, did it link to a source you could open, and did it invent anything you know is wrong.
4. Open every link. A source that does not say what the chatbot claims is a fail, however good the answer sounds.
5. Run one writing task too, for example turning the result into a short email for your manager. Freshness and writing are different skills, and the winner is not always the same tool.

I work as [your job] and follow [your topic or industry].
What changed about [the product, price or update] in the last 7 days?
Give me:
1. Three bullet points, newest first, each with the date.
2. A link to the original source for every bullet.
3. One line on what it means for someone in my job.
If you cannot find anything from the last 7 days, say so. Do not guess.

The last line matters. A chatbot that admits “I found nothing this week” is more useful at work than one that fills the gap with an old story.

Which one for which job

• A difficult email or a first draft: ChatGPT. It is the most practiced all-rounder for writing, and the free plan is enough to start.
• A very long PDF or a pile of documents: Gemini Pro, with its 1,500-page upload limit. Claude is also strong here, see our guide to analyzing long PDF files with Claude.
• Work that already lives in Gmail and Docs: Gemini, because it works where your files already are.
• What happened today in your field: Grok, thanks to live web and X search. Always open the links before you repeat anything.
• A spreadsheet or Office file: ChatGPT with its Excel and Word extensions. If your company has Microsoft 365 with Copilot, read our ChatGPT vs Gemini vs Copilot guide first.

Using two of them is completely normal. Plenty of people keep one main assistant for writing and open another only when they need something it does better.

What about Claude?

Claude is the fourth name you will hear, and many people prefer its writing style. If you are choosing between ChatGPT and Claude, our summary of what Reddit users say about ChatGPT vs Claude covers it. This page sticks to the three you asked about, because Grok answers a different need: fresh information.

For the full comparison on writing, files, price and privacy, read ChatGPT vs Claude AI for work.

If what you need is research with a source on every answer, our ChatGPT vs Claude vs Perplexity guide shows when to look things up in Perplexity and when to write in ChatGPT or Claude.

Torn between Claude and Gemini instead? Our Claude vs Gemini Pro guide compares the two paid plans for everyday work.

Need code rather than fresh news? Our ChatGPT vs Gemini for coding guide compares the two for scripts and small tools at work.

So which one should you pick?

If you want one tool and nothing else, pick ChatGPT and learn it well. If your work runs on Google, add Gemini, which may already be in your Google plan. Add Grok when you need to follow what is happening this week, and try the free version before any paid tier, since the useful comparison tier costs $30.

The real skill is not choosing the “best” chatbot. It is knowing which job to give which tool, and checking the answer before it leaves your desk.

Quick answers

Is Grok free to use?

Yes, with limits. The Free tier of The X Subscription includes limited chat and agent requests, and you can try Grok on grok.com or in the apps (X Subscription pricing).

Do I need an X account to use Grok?

No. You can sign in with an email address instead of an X account (Grok product page).

Which is cheapest if I only want a little more than free?

Google AI Plus at $4.99 a month. ChatGPT Go and X Subscription Plus both cost $8 a month.

Can ChatGPT and Gemini also search the web?

Yes, both can look things up and both offer deep research. Grok’s difference is that it also searches X posts live, which helps when news breaks on X first.

Is it safe to paste work documents into these chatbots?

Only into the version your employer approved. On personal plans, switch off model training in the settings and remove names and confidential details first.

Want one practical AI lesson like this every week?

Sources

• ChatGPT plans and pricing (OpenAI)
• ChatGPT Learn: pricing for Free, Go, Plus and Pro (OpenAI)
• Data controls FAQ (OpenAI Help Center)
• Google AI plans for Gemini (Google)
• Google AI plans with cloud storage (Google One)
• Gemini Apps Privacy Hub (Google)
• Grok product page (SpaceXSI)
• The X Subscription pricing (SpaceXSI)
• X Subscription pricing docs (SpaceXSI)
• Introducing The X Subscription (SpaceXSI)
• Consumer FAQs on training and Private Chat (SpaceXSI)

You need a small script for work. Maybe it is a Python file that merges twelve monthly CSV exports into one clean sheet, or a macro your team keeps asking about. You are not a developer. You just want the code to work, and half the internet says “Claude is the coder” while the other half says “ChatGPT does everything”.

Here is the short answer. In the ChatGPT vs Claude for coding question, both write good everyday scripts, and both can run Python on a file you upload, right in the chat. The real difference shows up when you want an agent that works on files on your own computer. ChatGPT’s agent is Codex, and OpenAI gives even Free and Go accounts limited access to it. Claude’s agent is Claude Code, and it starts at the $20 Pro plan. So: try both chats for free, run the one-script test further down this page, and only pay for the agent you actually end up using.

Want one practical AI lesson like this every week?

ChatGPT vs Claude for coding at a glance

Plans and names change often, so here is what the official pages say in October 2026.

• Write and run code in the chat, free: ChatGPT puts code in code blocks, and supported Python blocks get a Run button with a console for output and errors (OpenAI Help Center). Claude Free can search the web, create files and run code, and code execution with file creation is on every plan (Claude pricing).
• Coding agent on your own computer: Codex, in the ChatGPT desktop app, a terminal (CLI), an editor extension and on the web. Limited on Free and Go, included in Plus (Using Codex with your ChatGPT plan). On Claude’s side it is Claude Code, included in Pro and up, not in Free (Claude pricing).
• Paid plan most people start with: ChatGPT Plus at $20 a month (Codex pricing). Claude Pro at $20 a month, or $17 a month billed yearly.
• How the limits work: both use a five-hour window plus weekly limits. On ChatGPT, Codex shares its allowance with ChatGPT Work. On Claude, your chats and Claude Code draw from one pool.
• Instruction file the agent reads every time: AGENTS.md for Codex (AGENTS.md guide). CLAUDE.md for Claude Code.

Notice what is missing: a clear winner. For a non-developer, the useful question is simpler. Do I need a chat, or do I need an agent?

Chat or agent? The question that decides it

A chat is enough when the job fits in one file and you can copy the result out. A formula, a regex to clean phone numbers, a Python script that merges exports, a VBA macro. You paste data or a file in, you get code or a finished file back.

You need an agent when the job lives in a folder. The script has to read 40 files on your laptop, try itself, read its own error and fix it. Copying code back and forth between a chat and your computer gets old fast, and that is exactly the loop an agent takes over.

Most people at a desk job start with the first and stay there for months. That is fine. Start where your real task is, not where the hype is.

ChatGPT for coding: what you actually get

Code blocks you can run. Ask for a script and ChatGPT places it in a code block. Depending on your plan and the block, you can edit it directly, ask ChatGPT to change it, preview HTML pages and charts, or press Run on Python and see the output or the error in a console. It all runs in a sandbox (OpenAI Help Center).

Python on your own file. Upload a .csv or .xlsx and ChatGPT can write and run Python in a notebook-style environment to clean, merge or chart it. Two limits worth knowing: that environment cannot make web requests or API calls, and scanned PDFs or tables inside images may not come through with exact values (Data analysis with ChatGPT). Free accounts get limited data analysis, paid plans get more (ChatGPT pricing).

Codex, the agent. Codex is OpenAI’s coding agent. You sign in with your ChatGPT account in the desktop app (Codex mode), the Codex CLI, the editor extension or Codex web. In a terminal you open your project folder, type codex, sign in and describe the task (OpenAI Help Center). Two details make it friendlier than it sounds:

• It starts careful in a plain folder. Codex recommends read-only mode for folders that are not under version control, and its Auto preset lets it edit and run commands only inside your working folder. It asks before it edits outside that folder or uses the network (Codex approvals and security).
• It remembers your rules. Put an AGENTS.md file in the folder (“never touch the originals, write results to /output”) and Codex reads it before it does any work.

The catch: on Free and Go, Codex access is limited, and Codex Cloud (tasks that run on OpenAI’s computers) is not included at all. Plus is the realistic starting point for weekly use.

Claude for coding: what you actually get

Artifacts and code execution in the chat. Claude can show a script in an artifact beside the conversation, and if it throws an error there is a “Try fixing with Claude” button (Claude Help Center: artifacts). It can also run code in a private computing environment, analyze uploaded CSV files and hand you back a finished Excel, Word or PDF file. That works on every plan, Free included, with a 30MB limit per file (Create and edit files with Claude). One setup note: to upload .xlsx files, code execution and file creation has to be turned on in your settings.

Claude Code (Pro and up). Claude Code reads the files in a folder, edits them, runs commands and fixes what breaks. It runs in the terminal, in VS Code and JetBrains, in the desktop app and in the browser (Claude Code overview). You pick how much it does alone. In Manual mode it asks before most edits, commands and network calls. In Plan mode it reads, writes a plan, and edits stay blocked until you approve it. Newer versions start in auto mode, where a second model reviews actions instead of you, so press Shift+Tab to switch if you want to approve each step (Claude Code permission modes).

The catch: no Claude Code on the Free plan, and your chats and coding share one usage pool. Pro gives at least five times the Free usage per five-hour session, with weekly limits on top. For a few scripts a week, that is plenty. For all-day coding it is not.

Want the full Claude basics first, from sign-up to your first Project? Start with our Claude tutorial for beginners.

The 2026 change most ChatGPT vs Claude guides miss

A lot of comparison articles still tell you to “open a canvas” in ChatGPT for coding. With the newer ChatGPT models, canvas is gone. OpenAI moved writing and coding into editable writing blocks and code blocks directly in the chat. Paid users can still reach canvas through legacy models for a limited time, until those models are retired (ChatGPT release notes).

What that means for you: if a tutorial shows a canvas side panel and you cannot find it, nothing is wrong with your account. Ask for the code, then use the Run, Edit and Preview buttons on the code block itself. And treat any comparison written before mid-2026 with care, on both sides.

Test ChatGPT vs Claude on one real script

Benchmarks test code you will never need. Your own task shows which tool fits your work. This test takes about 20 minutes and costs nothing, because both free chats can run Python on an uploaded file.

Step 1. Make a safe sample. Take two or three of your real CSV exports, delete the rows with customer names or anything private, and keep 20 rows each. Same columns, same messy dates. That is your test file.

Step 2. Paste the same prompt into ChatGPT and into Claude, with the same files attached:

I am not a developer. I attached sample CSV exports from our system.

Write a Python script that:
- Reads every .csv file in one folder.
- Combines them into one table with the same columns.
- Turns the "Date" column into one format: YYYY-MM-DD.
- Removes rows that are exact duplicates.
- Saves the result as combined.xlsx in a new folder called "output".
- Never changes or deletes the original files.

Then:
1. Run it on my sample files here and show me the first 10 rows of the result.
2. Tell me exactly how to run it on my own computer, step by step, for Windows.
3. List everything you assumed about my files.

Step 3. Score both answers on four questions:

• Did the script run on the sample in the chat on the first try? If not, did pasting the error once fix it?
• Are the dates actually all in one format, including the odd ones like “3/4/26”?
• Could you follow the “run it on my computer” steps without searching anything?
• Did it list its assumptions honestly, such as the date order (day first or month first) or the file encoding?

The one that wins on your own file is your answer. If both pass, pick the one you already pay for or already have open. And if you only wanted answers from the data, not a reusable script, you may not need code at all: see how to analyze a spreadsheet with ChatGPT.

Which one for which job

• A formula, regex or one-off snippet: either free chat. Use the one already open.
• A script that cleans or merges files you can upload: both run Python on your upload. Claude can hand back a finished .xlsx or .docx, ChatGPT gives you an interactive table and charts. Run the test above.
• A script that works through a folder on your laptop: an agent. If you want to try one without paying first, Codex has limited access on the free ChatGPT plans. If you already pay for Claude Pro, Claude Code is included. Start either one in its most careful mode.
• Understanding a long script a former colleague left behind: both agents can read a whole folder. Ask “explain what this does, file by file, before you change anything”.
• Google Sheets and Apps Script: that is a different comparison. Read our Claude vs Gemini for coding guide, which tests exactly that.

Choosing a tool for more than code? Our ChatGPT vs Claude on Reddit roundup covers what people who pay for both say about writing, long documents and usage limits. And if you are weighing Claude Code against a self-hosted agent, see OpenClaw vs Claude Code.

Picking one everyday assistant rather than a coding tool? Our ChatGPT vs Gemini vs Grok guide compares the three for writing, files and fresh news, with a short test on your own tasks.

Comparing the two for emails, reports and client copy instead of code? Our ChatGPT vs Claude for writing guide puts both through four everyday writing jobs and shows which one to pick for each.

Four habits before you run AI-written code at work

• Test on fake or trimmed data first. The sample from Step 1 exists for a reason. Real customer data goes in only when the script has proven itself and your company allows it.
• Ask for the plain-English version. “Explain this script line by line, and tell me what it deletes or overwrites” takes ten seconds and catches most surprises.
• Keep secrets out of the chat. Never paste passwords or API keys. Use a placeholder like YOUR_KEY_HERE and fill it in yourself.
• Check your data settings. On ChatGPT Plus, conversations, including Codex work, may be used to improve models unless you turn training off in data controls (OpenAI Help Center). Claude’s consumer plans also let you opt out of model training (Claude pricing). Business plans work differently, so ask IT which tool you may use for work files.

Free or paid: what to do this month

Start free on both. The chats write and run the kind of scripts most professionals need, and the test above costs nothing.

Pay when you hit a real wall. That usually means one of two things: you keep copying code between the chat and your computer every day, or you keep running out of usage. Then pick one plan, ChatGPT Plus or Claude Pro, both $20 a month, and use it for a month on real tasks. If you still open the other tool every day, that tells you something. If you do not, you saved yourself $20.

If you also use AI for writing, documents and everyday tasks, not just scripts, our ChatGPT vs Claude AI comparison for work covers those jobs and the plan prices.

Not sure an agent is even the right tool? Some jobs are better as a no-code automation that runs on its own. Our guide on building an AI agent without code shows that route.

If your other option is Gemini because your work runs on Google, our ChatGPT vs Gemini for coding comparison shows which one writes the better script for your setup.

Quick answers

Is Claude better than ChatGPT for coding?

Not in a way that holds for every job. Both write and run everyday scripts well. Claude Code is the bigger reason people pick Claude, and it needs a paid plan. ChatGPT’s Codex gives you a limited agent even on the free plans. Your own one-script test is the honest tie-breaker.

Can I use ChatGPT and Claude for coding for free?

Yes, for chat-based coding. ChatGPT can run Python in code blocks and on uploaded files, with limits on the free plan. Claude Free can run code and create files. For an agent on your own computer, ChatGPT Free and Go include limited Codex access, while Claude Code needs Pro or higher.

Do I need to know Python to use them?

No. You need to know what result you want and how to check it. Describe the input and the output in plain words, ask for setup steps and assumptions, and test on a copy. You pick up the basics as you go.

Codex or Claude Code: which is easier for a non-developer?

Both run in a desktop app, so you do not have to live in a terminal. Both let you start in a careful mode: read-only in Codex, Manual or Plan mode in Claude Code. Choose the one that comes with the plan you already pay for, and give it one small folder task first.

Pick the one script you have been putting off. Make a 20-row sample tonight, paste the test prompt into both chats, and keep the one that got it right.

Want the next one in your inbox?

Sources

• ChatGPT plans and pricing (OpenAI)
• Codex pricing and usage limits (OpenAI)
• Using Codex with your ChatGPT plan (OpenAI Help Center)
• Codex approvals and security (OpenAI Developers)
• Custom instructions with AGENTS.md (OpenAI)
• Working with writing blocks and code blocks in ChatGPT (OpenAI Help Center)
• Data analysis with ChatGPT (OpenAI Help Center)
• ChatGPT release notes (OpenAI Help Center)
• Claude pricing and plan FAQ (Anthropic)
• Create and edit files with Claude (Claude Help Center)
• What are artifacts and how do I use them? (Claude Help Center)
• Claude Code overview (Claude Code Docs)
• Choose a permission mode (Claude Code Docs)

You keep hearing colleagues say “just ask Claude”. So you open it, stare at an empty box, type something vague, and get something vague back. Was that it?

Here is the short answer. To start with Claude, sign up free at claude.ai (or in the desktop or phone app), then give it one real task from your work with three things in the message: who you are, what you need, and what the result should look like. Add your own file with the + button so it works from your material, not from memory. Reply with corrections instead of starting over. When a task comes back every week, put it in a Project with standing instructions. That is the whole beginner loop, and you can run it in the next 30 minutes. The steps are below.

Want one practical AI lesson like this every week?

Step 1: Sign up and pick where you will use Claude

Claude is the AI assistant made by Anthropic. You can use the same account in three places: in a browser at claude.ai, in the Claude desktop app for Mac or Windows, and in the Claude app for iOS or Android (Claude Help Center: Get started with Claude). Nothing to install if you use the browser.

My advice for work: start in the browser on your work computer. That is where your documents and emails already are, and copying text in and out is easiest there. Put the phone app on later for quick questions on the go.

The Free plan is enough to learn everything in this tutorial. It includes chat on web, desktop and phone, web search, file creation, memory and up to 5 Projects (Claude pricing).

Step 2: Write your first work prompt (the three-part message)

A prompt is simply your message. Anthropic’s own advice is to talk to Claude the way you would talk to a coworker, and to be specific (Claude Help Center). Their beginner course breaks a good message into setting the stage (your role and goal), the task itself, and the rules, like style and tone (Claude Academy: Your first conversation with Claude).

Here is what that looks like with a real task. Swap in your own details:

I'm a project manager at a 20-person marketing agency.
A client emailed that our campaign launch is two weeks late and they're unhappy.

Draft a reply that:
- owns the delay without grovelling
- gives the new date (12 November) and one reason (late approval of the visuals)
- offers a 15-minute call this week

Keep it under 120 words. Friendly, direct, no corporate phrases.

Compare that with “write an email about a delay”. The first one gives Claude your situation, your limits and your tone. The second makes it guess all three, and guesses read like a template.

Once the reply looks right, a second prompt can polish it further. We collected the ones that work best in make this email sound more professional.

Step 3: Give Claude your own material

Claude gets much more useful the moment it works from your document instead of its general knowledge. Click the + button in the lower left of the chat box, choose “Add files or photos”, and pick a file, or drag it into the chat (Claude Help Center: Upload files to Claude).

What it takes, according to that same help page:

• Documents: PDF, DOCX, CSV, TXT, HTML, ODT, RTF, EPUB, JSON and XLSX.
• Per chat: up to 20 files, up to 500 MB each.
• PDFs: up to 1,000 pages. Claude reads charts and images in PDFs of 100 pages or fewer; longer ones are read as text only.

A good first test: upload a report you already know well and ask a question you know the answer to.

Read the attached report.
1. Give me the 5 most important points in plain English, one line each.
2. For each point, name the page it comes from.
3. List anything that needs a decision from me, with the deadline if one is mentioned.

Asking for page numbers is the beginner’s best habit. You can check every claim in seconds, and you learn quickly how far you can trust it on your kind of documents. One tip from Anthropic: when you ask about a specific page, use the page number your PDF viewer shows, not the one printed on the page.

Want to go deeper with documents? These guides take it step by step:

• Claude AI analyze long PDF files for reports that run to hundreds of pages
• Summarize a long report and extract action items
• Extract data from a PDF into a spreadsheet

Step 4: Correct it, don’t start over

The first answer is a draft. Most beginners either accept it as is or open a new chat and try again from zero. Both waste the work Claude already did.

Reply in the same chat, short and blunt, like you would to a colleague:

• “Shorter. Cut the first paragraph.”
• “Less formal. This is a client I’ve worked with for three years.”
• “Give me two other versions of the opening line.”
• “You said the budget is 40k, the report says 45k. Fix it and tell me if anything else came from that number.”

Anthropic’s help page lists this as one of its four starter tips: iterate, refine, ask follow-up questions. In practice two or three corrections usually get you from “OK” to “I’d send that”.

Step 5: Use a Project for work you repeat

If you catch yourself pasting the same background into Claude every Monday, you need a Project. A Project is a workspace with its own files and its own instructions, and every chat inside it uses both automatically (Claude Academy: Introduction to projects).

Setting one up takes a few minutes:

1. Open Projects in the left sidebar and click “+ New Project”.
2. Give it a clear name, like “Weekly client updates”.
3. Click Instructions and write how Claude should work in this Project. Save.
4. Upload the files it should always know, such as a style guide, a template or last month’s report.

Example instructions for a weekly update Project:

This project is for weekly status updates to our clients.
When I paste my rough notes, turn them into an update with three headings:
Done this week / Next week / Needs your input.
Max 150 words. Plain English, no jargon. Use the client's name, never "Dear client".
If my notes mention a risk or a delay, put it first under "Needs your input".

Claude Academy also recommends naming your files clearly (“Q4-Brand-Guidelines.pdf”, not “document1.pdf”), because Claude uses file names to find the right information. On the Free plan you can make up to 5 Projects, which is plenty to start.

Five beginner tasks to try this week

Learning sticks when you use it on Tuesday’s real work, not on test questions. Pick two of these:

1. The email you keep postponing. Paste the email you received, say what you want to answer in one sentence, ask for a reply under 100 words.
2. Meeting notes into actions. Paste your messy notes and ask for decisions, action items with owners, and open questions. Our guide on turning meeting notes into a report uses ChatGPT, but the prompts work the same in Claude.
3. The long PDF nobody read. Upload it and use the five-points prompt from Step 3.
4. A contract or policy check. Ask for the obligations, deadlines and anything unusual, with page numbers. See key points from a PDF for a full prompt, and remember it is a first read, not legal advice.
5. A first draft from bullets. Write five rough bullets for a proposal or announcement and ask Claude to turn them into a draft in your tone. Then correct it twice, Step 4 style.

Free or Pro: what a beginner actually needs

Start free. Upgrade only when you hit a wall you can name. Here is what the pricing page shows right now:

• Free ($0): chat everywhere, web search, files, memory, up to 5 Projects, the Sonnet and Haiku models.
• Pro ($20 a month, or $17 a month billed yearly): at least 5 times the usage of Free per 5-hour session, Projects without the 5-Project cap, the Research feature, the Opus models, and Claude in Chrome and Microsoft 365.

Every plan has usage limits that reset on a rolling five-hour window, and there is no fixed message count, because long chats and big files use more. If you run out mid-task on Free most days, that is your sign Pro will pay for itself. If you don’t, stay free.

Beginner mistakes that cost the most time

• Trusting numbers without checking. Claude can sound sure and still be wrong. For anything that goes to a client or your boss, check the figures against the source. Asking for page numbers makes this fast.
• One endless chat for everything. A chat that started about invoices and drifted to holiday plans gets messy. New topic, new chat. Recurring topic, a Project.
• Pasting things you shouldn’t. Follow your company’s rules on client data. On the personal plans, whether your chats can be used to train models is a setting you can switch off (the pricing page lists model training as opt-out), so check your privacy settings once.
• Asking for “something good”. Vague in, vague out. Name the reader, the length and the tone every time until it becomes a habit.

Claude or ChatGPT for a beginner?

Both are good enough to learn on, and the habits in this tutorial work in both. Pick the one your company already allows, or the one you will actually open every day. If you want to see what real users say about the differences in writing and document work, we rounded it up in ChatGPT vs Claude on Reddit.

For a side-by-side look at price, privacy and which one fits which work task, see ChatGPT vs Claude AI: which one to use for work.

If you go with ChatGPT, our ChatGPT tutorial for beginners walks you through the same first steps there, from your first chat to real work tasks.

If your work lives in Gmail and Google Drive, our Gemini tutorial for beginners shows the same habits in Gemini, step by step.

FAQ

Is Claude free to use?

Yes. The Free plan costs $0 and includes chat on web, desktop and phone, web search, file uploads, memory and up to 5 Projects. Pro adds more usage and features for $20 a month (Claude pricing).

Do I need to be technical to use Claude?

No. Everything in this tutorial is typing plain English and clicking the + button. Claude also has tools for developers, but you can ignore them completely and still save hours on email, reports and documents.

Which Claude model should a beginner use?

The default one is fine. The model you are talking to is shown below the message box on web and desktop, and you can click it to switch (Claude Help Center). On Free you can choose Sonnet or Haiku; Opus comes with Pro.

Can Claude read my PDFs and spreadsheets?

Yes. You can upload PDF, Word, CSV, Excel and more, up to 20 files per chat. PDFs can be up to 1,000 pages, and charts are read in PDFs of 100 pages or fewer (Upload files to Claude).

How long does it take to get good at Claude?

The basics take one sitting: one real task, one file, two corrections. Getting fast takes a couple of weeks of using it on your actual work. Setting up one Project for your most repeated task is the step that makes it feel like a real assistant.

Ready for the next step after your first prompts? Here is how to use Claude for work by connecting it to your email, calendar and files, so it can prepare your day instead of waiting for you to paste things in.

Learning something new for work, like a certification or a new field? Here is how to use Claude to study without letting it do the thinking for you.

Still deciding whether Claude or Gemini fits how you learn? Compare them in Claude vs Gemini for studying.

Get one practical AI lesson for your work, every week.

Once you know the basics, the Claude AI course guide shows which free lessons to take next and how to turn them into a two week plan.

You keep hearing people talk about OpenClaw, and you want it on your Mac tonight. Then the docs show you a menu bar app, a Terminal command, npm and a word called Gateway. Which one is yours? And what if your Mac is a few years old?

Here is the short answer. If your Mac runs macOS Tahoe 26.2 or newer, the easiest way to install OpenClaw on a Mac is the menu bar app: download the DMG from the OpenClaw releases page, drag OpenClaw into Applications, open it, choose This Mac, and let it set itself up. No Terminal, no Homebrew, no admin rights for that part. On an older Mac, or if you like the terminal, one command does the job: curl -fsSL https://openclaw.ai/install.sh | bash. Either way you need a way into an AI model: an existing Codex or Claude login, or an API key. The docs talk about a first chat in minutes; give yourself 20 so you can read along calmly. The steps for both routes are below.

Want one practical AI lesson like this every week?

Before you start: two things to check on your Mac

Open the Apple menu in the top-left corner and choose About This Mac. That one window answers both questions.

• Your macOS version. The window shows the name and version number (Apple: which macOS is your Mac using). The OpenClaw app needs macOS 26.2 (Tahoe) or later. The Terminal route is more forgiving: the Node versions OpenClaw runs on are built for macOS 13.5 or later, and older releases “may still run them but are not supported” (OpenClaw docs: macOS app).
• Your chip. If you see Chip followed by a name like M1 or M3, you have Apple silicon. If you see Processor with an Intel name, you have an Intel Mac (Apple: Mac computers with Apple silicon). OpenClaw has a download for each, plus a Universal one if you are not sure.

One more thing: access to an AI model. OpenClaw itself is free and open source under the MIT licence. The model behind it is what does the thinking, and that is what you pay for. Setup can reuse a Codex or Claude login you already have, or an OpenAI or Anthropic API key.

Still deciding whether OpenClaw is the right tool at all? Read OpenClaw vs Claude Code or OpenClaw vs Claude Cowork first. This page assumes you have decided.

Which install route fits your Mac?

• The menu bar app (OpenClaw.app). A normal Mac app with a menu bar icon, a dashboard, a chat window and a Quick Chat bar. It carries its own runtime, so it needs no separate Node install. Best if you are on macOS 26.2 or later and would rather not touch Terminal.
• The Terminal installer. One command that installs Node if needed, installs OpenClaw and starts a guided setup. Best for older Macs, or if you like to see what is happening.
• npm. Only if you already manage Node yourself. Covered in one short block below.

A word you will see everywhere: the Gateway. That is the part of OpenClaw that runs in the background, holds your settings and connects your chat apps to the AI. Both routes end with a Gateway running on your Mac.

How to install OpenClaw on Mac with the menu bar app

1. Download the right DMG. Go to the OpenClaw releases page on GitHub and pick the Apple Silicon, Intel or Universal DMG. When this page was written, the docs pointed to release 2026.10.3. The docs also warn that some releases ship only CLI or Windows files; if the newest one has no Mac app, take the newest release that does.
2. Put it in Applications. Open the DMG and drag OpenClaw into your Applications folder. Do not run it from Downloads or from the disk image. The permissions page explains why: macOS ties permissions to the app’s path, and a release app should run from /Applications/OpenClaw.app so your grants stick (OpenClaw docs: macOS permissions).
3. Open it and approve two prompts. macOS asks if you want to open an app you downloaded: click Open. Then it asks if OpenClaw may find devices on your local network: click Allow. The app uses that to reach a Gateway on another machine (OpenClaw docs: Onboarding for the macOS app).
4. Read the security notice. It explains that OpenClaw is a personal agent for one trusted person by default. Continue when that fits you.
5. Choose This Mac. A window called Getting things ready prepares the included runtime, starts the Gateway and checks it. The docs say this step needs no runtime download, Node, Terminal, Homebrew or administrator access. Connecting an AI provider can still need internet (OpenClaw docs: Gateway on macOS).
6. Let it connect your AI. The app tries, in order: saved OpenClaw sign-ins, a Codex login, a Claude CLI login, an OPENAI_API_KEY or ANTHROPIC_API_KEY, and models already loaded in Ollama, LM Studio or llama.cpp. It tests each with a real request and keeps the first that works. If nothing works, it sets up the official Codex plugin and offers a ChatGPT sign-in. The dashboard shows what it picked, and you can switch.
7. Finish in the dashboard. A guided conversation helps with the rest: memory import, chat channels and permissions. All of it is optional, and you can come back to it later.

One setting to know on day one. On a fresh setup, the app runs the Gateway itself, so it stops when you quit the app. If you want OpenClaw to keep working with the app closed, turn on Keep OpenClaw running when the app is closed, next to Launch at login in the dashboard’s device settings. It is off by default. And try Quick Chat: press Option-Space for a Spotlight-style chat bar without opening a full window.

Install OpenClaw on Mac with Terminal

On an older Mac, or if you just prefer it: press Command-Space, type Terminal and press Return. Then paste:

curl -fsSL https://openclaw.ai/install.sh | bash

The script checks your Node version and installs Node 26 through Homebrew if it is missing. It installs Homebrew only when it needs it for Node or Git (OpenClaw docs: Installer internals). Then it installs OpenClaw and starts the setup wizard. Next:

1. Pick Quick start. It finds an existing Codex or Claude Code login or API key, tests it with a real request, saves your settings and opens the web dashboard (OpenClaw docs: Getting started).
2. Make it run in the background. Quick start keeps the Gateway running in that Terminal window. Press Control-C to stop it, then run openclaw gateway install. On a Mac this creates a LaunchAgent, so the Gateway starts at login and restarts if it crashes. Your settings stay saved.
3. Check it:

openclaw --version
openclaw doctor
openclaw gateway status

The status should show the Gateway listening on port 18789. Then run openclaw dashboard to open the Control UI in your browser. If it loads, everything works.

Already manage Node yourself? The docs give this route for npm 12 or 11.16 and newer (on npm 11.15 or older, leave out --allow-scripts=openclaw):

npm install -g openclaw@latest --allow-scripts=openclaw
openclaw onboard --install-daemon

OpenClaw needs Node 24.16+ or 26.1+, and Node 26 is the recommended one (OpenClaw docs: Install).

Mac permissions, without the confusion

This is where most Mac installs get messy, so take two minutes here. macOS guards your screen, microphone, files and keyboard control behind separate permissions. OpenClaw asks only when a tool needs one: you see a Grant card in the chat, and the agent waits for you before it tries again.

• Grant what you use, not everything. You will find the full list under Dashboard > Settings > This Mac > Permissions: Automation, Notifications, Accessibility, Screen Recording, Microphone, Speech Recognition, Camera and Location. For chat-based writing and research, you can leave the heavy ones off until a tool asks.
• Screen Recording still says missing? Click Grant, switch on OpenClaw under Privacy & Security > Screen & System Audio Recording in System Settings, then quit and reopen OpenClaw.
• Never give Accessibility to “node”. The docs warn that a node entry with Accessibility lets any script run through that Node control your Mac’s interface. Give it to OpenClaw.app instead, and remove a node entry if you added one by accident.
• Files on your Desktop or in Documents hang? macOS may block background processes from those folders. The docs’ simple fix: put the files you want OpenClaw to work on in its own workspace folder, ~/.openclaw/workspace.

And one storage tip: keep OpenClaw’s data on your Mac’s own disk, not in iCloud Drive or another synced folder. The docs say sync delays and file locks can break sessions and credentials, and openclaw doctor warns you if your data sits in a synced folder.

Your first 15 minutes after the install

An install is only worth something once it does a job. Do these three things right away:

1. Send a first message. Something small and real: “Turn these notes into a checklist for tomorrow morning: [paste]”. If you get an answer, the AI connection works.
2. Connect Telegram if you want to message it from your phone. The docs call it the fastest channel to set up, since it needs just a bot token. By default, anyone new who messages your bot gets an 8-character pairing code, and nothing they send is processed until you approve it. Codes expire after one hour (OpenClaw docs: Pairing).
3. Run the security check: openclaw security audit. It tells you if your setup has drifted from the safe defaults (OpenClaw docs: Security). Fix what it flags before you add more channels.

Then give it one job you repeat every week, like a Friday summary of your notes file. If you would rather build agents with drag-and-drop tools, see how to build an AI agent without code.

Can you run the AI model on your Mac too?

Yes, but go in with open eyes. If you already have a model loaded in Ollama, LM Studio or llama.cpp, the app’s setup can find it, and the docs say it never downloads a model by itself. For a guided route, install the llama.cpp plugin, run openclaw onboard and choose Managed local server. It shows the model, download size and backend before downloading, and tests a real tool call before making it your default. The smallest recipe needs at least 8 GiB of memory, and larger ones need more (OpenClaw docs: Local models). The same page is honest about the catch: local models do not have the safety filters of hosted providers. Start with a hosted model, get one real task working, then experiment.

Planning to buy a Mac just for this? See which Mac mini specs OpenClaw actually needs, for a cloud model and for a local one.

When the install does not work

• Not sure what is wrong: run openclaw triage. It runs read-only checks and writes a clean description of the problem, and can hand it to Claude Code, Codex or OpenClaw’s own agent. The docs say nothing leaves your Mac until you choose one.
• “command not found: openclaw”: close Terminal and open a new window first. If you used the app, check that it runs from Applications: the docs say copies running from Downloads, the Desktop or a disk image do not add the openclaw command to your shell. On the Terminal route it is almost always a PATH problem, and the Node.js page in the docs walks you through it.
• The app’s setup fails: choose Retry setup. You can also open Connection… from the menu bar icon, where the app offers Install, Update or Repair Gateway. If the app says its runtime is missing or broken, reinstall OpenClaw.app.
• A permission prompt never shows up: quit the app, remove OpenClaw under System Settings > Privacy & Security, open it again from Applications and grant again. The docs add that some permissions only come back after a full restart.
• The Gateway keeps restarting: run openclaw gateway status and openclaw doctor. The log lives at ~/Library/Logs/openclaw/gateway.log.

Should you put it on your work Mac?

Ask first. OpenClaw runs in the background, can be reached from chat apps, and can be given tools that act on your Mac. The setup’s own security notice says it is built for one trusted person, and shared setups need extra lock-down. On a company laptop, that is a conversation for your IT team. A personal Mac is the easier place to learn.

Wondering whether a desktop agent that clicks through your apps fits your office better? Our guide to the best AI agent for desktop workflows compares the options. Setting it up on a PC too? Here is how to install OpenClaw on Windows.

Quick answers

Does OpenClaw work on an older Mac?

The menu bar app needs macOS 26.2 or later. The Terminal install runs on Node, whose Mac builds are made for macOS 13.5 and later. Older versions may work but are not supported.

Apple silicon or Intel?

Both. There is a separate DMG for each, and a Universal DMG if you are not sure which you have.

Do I need Homebrew?

Not for the app. The Terminal installer uses Homebrew only if it needs it to install Node or Git.

Is OpenClaw free?

The software is free and open source (MIT licence). You pay for the AI model you connect, unless you run a local model on your own Mac.

How do I uninstall it?

Preview first with openclaw uninstall --dry-run --all, then run openclaw uninstall --all to remove the service, settings, workspace and app. Want to keep your agent’s files? Run plain openclaw uninstall and pick what to remove. If you installed the app, also delete it from Applications (OpenClaw docs: Uninstall).

Check your macOS version, pick your route, and send your first message before you close the lid tonight.

Want the next one in your inbox?

Want OpenClaw running all day without your laptop? Here is how to set up a Mac mini as an always-on OpenClaw box.

Sources

• macOS app (OpenClaw docs: requirements, downloads, first run, updates)
• Onboarding for the macOS app (OpenClaw docs: prompts, local vs remote, connecting your AI)
• Gateway on macOS (OpenClaw docs: automatic setup, background service, logs, iCloud warning)
• macOS permissions (OpenClaw docs)
• Install and Installer internals (OpenClaw docs)
• Getting started (OpenClaw docs: Quick start, Gateway service, dashboard, triage)
• Local models, Pairing, Security and Uninstall (OpenClaw docs)
• Find out which macOS your Mac is using and Mac computers with Apple silicon (Apple Support)
• OpenClaw releases and licence (GitHub)