Smartphone and laptop side by side on a wooden desk with glasses, for comparing AI chatbots

How to Compare AI Chatbots: 6 Tests to Find the Best One

Updated 20 November 2026. The best way to compare AI chatbots is to give two or three of them the same real tasks and score the answers. Run six tests: accuracy, following instructions, writing quality, working with your files, privacy controls and price. The "best" chatbot is the one that wins on the jobs you actually do, not the one that tops someone else's leaderboard.

Why "which AI chatbot is best?" has no single answer

ChatGPT, Gemini, Claude and Copilot are all capable general assistants, and each company updates its models and plans often. A ranking that was true six months ago may not be true today.

What changes less is what you need. A student revising for exams, a sole trader writing quotes and a parent planning meals all want different things. So instead of trusting a league table, test the tools yourself. It takes about an hour, and you can do it on the free plans.

The field is getting more competitive: Ofcom's Online Nation 2025 report found ChatGPT had 1.8 billion UK visits in the first eight months of 2025, up from 368 million a year earlier, while rivals such as Gemini, Claude and Perplexity were also growing fast.

The main AI chatbots at a glance

Here is a factual overview of the four best-known general assistants. Prices are from each company's own pricing page at the time of writing (November 2026). They change often, so check before you buy.

ChatGPT (OpenAI)

  • Free plan: yes.

  • Paid plans: Go, Plus and Pro. OpenAI's help center lists Plus at $20 a month, with local pricing in some countries. Full details on the ChatGPT pricing page.

  • Good fit if: you want a general all-rounder with a large range of features and a big user community sharing tips.

Gemini (Google)

  • Free plan: yes, with a Google account.

  • Paid plans: Google's subscriptions page shows Google AI Pro at $19.99 a month in the US (£18.99 in the UK), with higher Ultra tiers.

  • Good fit if: you already use Gmail, Google Docs and Google Drive.

Claude (Anthropic)

  • Free plan: yes.

  • Paid plans: Pro, Max and others. Claude's pricing page lists Pro at $20 if billed monthly; Anthropic's help center notes local currency pricing where supported.

  • Good fit if: you do a lot of writing, long documents or careful reasoning, and want to test it on those.

Microsoft Copilot

  • Free plan: yes.

  • Paid plans: Copilot comes with Microsoft 365 plans for individuals. Microsoft's pricing page lists Microsoft 365 Personal at $9.99 a month and Premium at $19.99 a month in the US (£8.49 and £18.99 in the UK).

  • Good fit if: you already use Word, Excel, PowerPoint and Outlook.

We don't give these tools star ratings here. Your six tests will tell you more than any score we could invent.

Before you start: set up a fair test

  1. Pick 2 or 3 chatbots. Testing all four is fine, but it takes longer.

  2. Use the same plan level. Compare free with free, or paid with paid.

  3. Use the same prompt, word for word. Copy and paste it into each tool.

  4. Start a fresh chat for each test, so earlier messages don't affect the answer.

  5. Use your real tasks. A test is only useful if it looks like what you'll actually do.

  6. Keep notes. Use the scorecard at the end of this guide.

Hands holding a smartphone with a chat app open, next to a coffee cup and notebook

The six tests in one table:

Test

What to try

What good looks like

Accuracy

A question you already know the answer to

Correct, with sources you can open

Instructions

A prompt with 3 clear rules

Follows all 3 rules

Writing

A short email in your tone

Sounds like you, needs few edits

Files

A PDF or spreadsheet you know well

Accurate summary, no invented details

Privacy

The settings page

Clear opt-outs for training and history

Price and limits

A normal week of use

You rarely hit the limit

Test 1: Accuracy

What it checks: whether the chatbot gets facts right, and admits when it doesn't know.

How to run it: ask three questions you already know the answers to, from your job, hobby or local area. Add one question about something recent, and one that is a little tricky.

Example prompt: "What are the current rules for [a topic you know well] in your country? Give your sources."

What good looks like: correct facts, sources you can open, and clear signs of uncertainty where it is unsure. The UK's National Cyber Security Centre warns that these tools can make things up and be biased, so a tool that admits gaps is a good sign.

Test 2: Following instructions

What it checks: whether it does exactly what you asked.

Example prompt: "Write a 3-sentence product description for a handmade oak chopping board. Use American spelling. Don't use the word 'perfect'. End with a question."

What good looks like: exactly three sentences, American spelling, no banned word and a question at the end. Count the misses. Small rule-breaks add up when you use a tool every day.

Test 3: Writing quality and tone

What it checks: whether the writing sounds natural and fits the reader.

Example prompt: "Write a short, warm email to a customer apologizing that their order is 3 days late. Offer 10% off their next order. Under 120 words."

What good looks like: something you could send with only light edits. Watch for stiff phrases, too many exclamation marks and American spelling. Then ask for a change ("make it more casual") and see how well it adjusts.

Test 4: Working with your files

What it checks: how well it reads and summarizes a document, spreadsheet or image.

How to run it: upload the same non-sensitive file to each tool, such as a public report, a menu or a spreadsheet of made-up numbers. Remove anything personal first.

Example prompt: "Summarize this document in 5 bullet points, then list 3 questions it doesn't answer."

What good looks like: a summary that matches the file, with no invented details. Check one or two points against the original yourself.

A woman working on a laptop while sitting on the floor beside a sofa

Test 5: Privacy and data controls

What it checks: how much control you have over your data.

How to run it: open each tool's settings and look for:

  • whether your chats can be used to train its models, and whether you can opt out

  • whether you can turn off or delete chat history

  • whether there's a temporary or private chat mode

  • which apps and accounts are connected

For example, OpenAI's data controls guide explains that personal ChatGPT plans can turn off Improve the model for everyone under Settings > Data controls. Find the matching settings in each tool you test.

What good looks like: clear, easy-to-find controls that match what you're comfortable with. If you'll use it for work, also check your employer's rules.

Test 6: Price and limits

What it checks: whether the free plan is enough, and what an upgrade really gets you.

How to run it: use each free plan for a normal week. Note when you hit a limit, and what the tool offers when you do.

What good looks like: a free plan that covers your normal use, or a paid plan whose extras you'd actually use. Check whether the price includes sales tax or VAT, whether it's billed monthly or yearly, and whether it bundles anything you already pay for, such as cloud storage or Microsoft 365.

Your free AI chatbot scorecard

Score each chatbot from 1 (poor) to 5 (excellent) on each test, then add them up.

  • Accuracy: __ / 5

  • Following instructions: __ / 5

  • Writing quality and tone: __ / 5

  • Working with files: __ / 5

  • Privacy controls: __ / 5

  • Price and limits: __ / 5

  • Total: __ / 30

If one test matters most to you, such as accuracy for research or writing quality for a small business, count that score twice.

Mistakes to avoid when comparing AI chatbots

  • Testing with one question. One lucky or unlucky answer tells you very little. Run all six tests.

  • Changing the prompt between tools. Tiny wording changes can change the result. Copy and paste.

  • Comparing a free plan with a paid one. Paid plans often use more capable models and have higher limits.

  • Trusting screenshots online. Other people's results reflect their prompts and settings, not yours.

  • Testing once and never again. These tools change often. Re-run your favorite tests every few months.

Frequently asked questions

Which AI chatbot is the most accurate?

It depends on the topic and changes as models are updated, so no single tool is most accurate for everything. Run Test 1 with questions from your own field, and always check important facts against an official source.

Is ChatGPT better than Gemini?

Neither is better for everyone. Gemini fits well if you live in Google's apps, while ChatGPT is a strong general all-rounder. Run the same six tests in both and compare your scores.

Should I pay for an AI chatbot?

Only if the free plan keeps running out, or a paid feature saves you real time. Use the free plan for a week first, then decide.

Can I use more than one AI chatbot?

Yes. Many people use one for everyday questions and another for a specific job, such as writing or working with documents. Free plans make this easy.

3 tips before you go

  1. Use the exact same prompt in every chatbot you test, or the comparison won't be fair.

  2. Keep a simple score out of 5 for each test in a notes app so your decision is based on evidence, not first impressions.

  3. Re-run your test every few months; chatbots change models often and the winner can change too.

Related reading

Sources

Written and checked under our editorial policy. Spotted a mistake? Tell us and we will fix it.