ChatGPT vs. Claude vs. Gemini: How to Choose an AI Assistant

ChatGPT, Claude and Gemini now overlap so much that the right pick depends less on which one is cleverest than on how it fits your work. The quickest way to settle ChatGPT vs Claude vs Gemini is to answer three questions, shortlist on the answers, then run one real task through all three:

1. Where does your work already live: Google apps, Microsoft 365 or a mix?
2. What will you hand the assistant: short text, long documents, photos, audio or video?
3. How much of your conversations are you willing to let the company keep?

There is no single winner for everyone. Gemini is the natural first try for people who spend their day in Gmail and Google Docs, Claude and ChatGPT both reach into Google and Microsoft tools, and all three handle everyday writing well enough that your own test decides; the four-step blind test near the end shows how. If you have never used one, begin with the basics of using an AI assistant before comparing them.

ChatGPT vs Claude vs Gemini, side by side

ChatGPT is made by OpenAI, Claude by Anthropic and Gemini by Google, and each is a chatbotchatbot: A program that holds a conversation in everyday language, replying to what a user types or says. Most of today's AI chatbots run on large language models, which can give fluent, confident answers that are partly or wholly wrong.Full entry in the glossary with a free plan and paid tiers. As checked on 28 September 2026, ChatGPT, Claude and Gemini each had a single privacy setting that was the main switch for whether a person’s chats could be used to improve the company’s AI models.¹²³

Each row is a trade-off, and every row changes as companies rename plans and move features between tiers, so the sections below explain what each difference means for you.

Everyday writing and reasoning: closer than the rankings suggest

For emails, summaries, explanations and first drafts, all three assistants are capable enough that the gap between them is usually smaller than the gap between a vague request and a clear one. Reviewers still describe different personalities: Zapier’s 2026 guide, by Miguel Rebelo (Zapier sells integrations for all three), calls ChatGPT the best general-purpose chatbot, rates Claude best for writing and coding, and picks Gemini for its links to Google’s own apps, while noting that ChatGPT’s sourcing from the web is uneven and Gemini’s answer quality inconsistent.¹⁷

All three are large language models trained on vast amounts of text and tuned to follow instructions, and they share many strengths and weaknesses. The differences you notice day to day are often in tone, length and formatting: one assistant answers in tidy bullet points, another in flowing paragraphs, a third asks a clarifying question first.

Public leaderboards look like a shortcut, but they measure something narrower than your work. A 2025 NeurIPS paper led by researchers at Cohere Labs (the research arm of an AI company whose models appear on these rankings) found that some large developers could test many private versions of a model on the popular Chatbot Arena leaderboard and publish only the best score, which rewarded fitting the leaderboard rather than general quality.¹⁸ We read its preprint. A ranking says which model pleased other people’s prompts last month, not which one suits your client emails.

Myth

The assistant at the top of this month's leaderboard will write your reports best.

Fact

Rankings shift with each release and reflect other people's prompts. How an assistant handles your own documents, tone and deadlines is a separate question that only your own test answers.

A project coordinator in Nairobi asks all three to turn meeting notes into a client update. One draft is warm but too long, one is crisp but drops a deadline, one is accurate but stiff. The right one is the draft that needs the least fixing for that client, so judge assistants on your own material.

Files, photos and live web results on the free plans

All three free plans read uploaded files, but each rations how much you can hand over at once, and each counts differently: ChatGPT by how much text fits in one message, Claude by files per chat, Gemini by files per prompt and minutes of audio and video. The unit matters more than the headline number, because it decides what you will have to cut.

At work, a 40-page policy report pasted in as text is well past what ChatGPT’s free plan takes in one message, while Claude and Gemini count an uploaded report as one file of their allowance, within their size limits. An hour-long recorded interview is far past Gemini’s free audio limit. Photos, such as a whiteboard after a meeting or a screenshot of an error message, count as uploads too, within the same limits.

Web search is the feature to treat with most care. OpenAI’s own help page says search results and citations can be incomplete, outdated or incorrect, and advises opening a cited source to check it.¹⁹ That caution applies to all three: a link beside an answer shows where the assistant looked, not that it read the page correctly. A confident answer with a source attached can still be wrong, for reasons covered in what makes AI invent facts.

Before choosing an assistant for document work, upload one real file of the size you usually handle, and see whether the free plan takes it whole. Then open two of its web links and confirm they say what the answer claims.

Google or Microsoft: where your work already lives

An assistant becomes far more useful once it can see the email, calendar and documents you already use, so the ecosystem you work in is often the deciding factor. Gemini starts with an advantage for Google users because it is Google’s own assistant, built to work with the Gmail and Calendar you already use. For people whose work runs on Microsoft 365, or on a mix of services, the table shows that Claude and ChatGPT reach further.

They do it through connectors: permissions you grant so the assistant can search, read or act in another service, such as finding last month’s proposal in your Drive or drafting a reply in Outlook.

Two warnings come with that convenience. A connected assistant can read far more than the one document you meant to share, so connect only the accounts you need. And at work, your employer’s rules come first: if your company runs Microsoft 365 or Google Workspace, an administrator may have already approved one assistant and blocked the others, which settles the choice for your work account.

Who keeps your conversations, and for how long

The privacy difference that matters most is what each company does by default and how long it keeps your chats. A 2025 study by Jennifer King and colleagues, published in the proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, reviewed the privacy policies of six leading US AI developers and concluded that all six seemed, by default, to train and improve their models on people’s chats, and that some kept the data indefinitely.²⁰ It examined written policies, not what the companies do in practice, and we saw its abstract alone.

Each assistant names its control differently. In ChatGPT it is “Improve the model for everyone” under Data controls; turning it off stops new conversations being used for training but does not delete saved chats, and rating an answer with a thumbs up or down can still send that whole conversation for training.¹ Claude’s choice sits in its Privacy Settings: Anthropic says that if you allow your chats to be used for training, they can be kept for up to five years, and if you do not, its standard 30-day retention applies. Those consumer terms do not cover its business, education and API services, which follow separate rules.²¹ Google’s Gemini ties training to its Keep Activity setting: with it off, Google says it keeps chats for 72 hours and does not use them to improve its models unless you send feedback; with it on, activity is deleted after 18 months by default, and human reviewers read some chats.³

Please don’t enter confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services, including machine-learning technologies.

Picture a hypothetical freelance accountant in Manila pasting a client’s figures into a free chat to check a formula. If the training setting is on, that spreadsheet may be kept for months or years and, on some services, read by a human reviewer. Set the training control before your first sensitive chat, not after, and keep client, health and financial details out of any consumer plan.

Prices, languages and features change at the border

Which assistant you can use, what it costs and which features you get all depend on where you live. When we opened Google’s plans page on 28 September 2026, one visit showed prices in Turkish lira and another, on the page’s US version, in US dollars, so the price you see in a review may not be the price you pay.⁶

Europe and the UK sometimes get new features later than other regions, or not at all. Google’s Bard, the chatbot later renamed Gemini, reached the EU in July 2023, several months after it opened in the US and UK, once Ireland’s data protection regulator had asked Google for a data protection impact assessment.²² The pattern continues: Google’s tool for importing your chat history and memory from another assistant into Gemini was not available in the European Economic Area, Switzerland or the UK when we checked.²³

Wherever you live, check three things on each company’s pages before you commit: that the assistant is offered in your country, that the feature you need is switched on there, and what the plan costs in your currency. Language matters too: all three work in many languages, so test yours with a real task rather than assuming results in English carry over.

Which assistant to try first, by need

Start with the assistant that fits your main need and keep the others as a second opinion. The matrix below is our editorial reading of the companies’ pages, not a test result. Where your work lives sits on one axis because it controls which of your files and messages the assistant can reach without copying and pasting; the kind of material sits on the other because it decides which free plan can take it.

Work mostly in Google appsWork in Microsoft 365 or a mixMostly writing and documentsAlso audio and video

Mostly writing and documents, Work mostly in Google apps: Gemini firstthen compare one task with Claude or ChatGPT

Mostly writing and documents, Work in Microsoft 365 or a mix: Claude or ChatGPTwhichever your employer approves

Also audio and video, Work mostly in Google apps: Gemini firstits free plan takes short audio and video

Also audio and video, Work in Microsoft 365 or a mix: Gemini for mediayour approved assistant for the rest

For a general writing partner, start from the reviewer’s picks above and let the test below decide.

Test all three on your own task

A practical way to choose is a small blind test: give all three the same real task, hide which answer came from which assistant, and pick the one you would actually use. The method borrows from how researchers rank chatbots at scale.

The study

Limited evidence

Blind votes as a yardstick: the Chatbot Arena method

Researchers mostly at the University of California, Berkeley built Chatbot Arena, where people ask a question, see answers from two unnamed models side by side and vote for the better one. After collecting more than 240,000 votes, they reported that the crowd’s votes agreed well with those of expert raters.²⁴

The method’s strength is that you judge the answer without knowing the brand; its limit is that crowd votes reflect other people’s questions.

1. One real task and one clear prompt, the same for all three assistants
2. Three answers with the names hidden; label yours only A, B and C
3. Your own score: facts right, tone right, minutes of editing needed
4. Only then reveal which assistant wrote which answer

To run it:

1. Pick one task you do every week, such as a client email, a summary of a report or a plan for next week, and write one clear prompt with the material it needs.
2. Paste the same prompt into all three free plans on the same day, and copy each answer into a document labelled only A, B and C.
3. Score each answer on what matters to you: facts right, tone right, and minutes of editing it would need.
4. Reveal which assistant wrote which answer, then repeat with a second, different task before you decide.

Before you settle on one assistant

• Ran the same real task through all three
• Scored the answers before knowing which was which
• Uploaded one file of the size I usually use
• Checked the privacy setting that controls training
• Confirmed the plan and features offered in my country

Download PDF ↓

Get this checklist as a one-page PDF. We'll email you the download link.

Your emailAlso send me the weekly newsletter (launching soon)Email me the PDF

We use your address to send this link and, only if you tick the box, the newsletter. Privacy policy.

The bottom line

Your own test should decide between ChatGPT, Claude and Gemini, because they are close on everyday writing and differ mainly in the apps they reach, the files they take and how they handle your chats. Run one real task through all three blind, set the training control before your first sensitive chat, and check the pages for your country, because the details in any comparison, this one included, will change.