
Mission 1 of 11
ChatGPT, Claude or Gemini? Score them yourself, for free.
Everyone online has a favourite and a hot take. Give all three the same jobs, in three tabs, for free, and in under an hour you'll have your own rule for which AI to open for which job.
What you'll do
Open all three on free plans, give them identical tasks and score the work on a simple 3, 2, 1 scorecard.
Try a tricky sum, the bluff test, live facts, a picture and a working mini tool.
Turn your scores into a one-page rule that tells you which AI to open for which job.
- What each free plan gives you and where its limits sit
- The bluff test, a 30 second way to see how an AI behaves when it doesn't know
- A browser. A laptop or desktop is easiest, because you can see three tabs quickly. A phone works too with the three apps
- An email address, or a Google account to sign in with
- A note or spreadsheet for your scores. Step 1 gives you one to copy
- About 50 minutes. You can stop after any step and pick it up later
Your progress0/7
- 01Open three tabs and run your first race8 min
- 02See what each free plan really gives you5 min
- 03Test the thinking, then try to fool them8 min
- 04Test them on what happened this week7 min
- 05Make them build a picture and a working tool8 min
- 06Run your real job, then write your rule8 min
- 07Decide if you ever need to pay, and go further4 min
Open three tabs and run your first race
Think of this as a trial shift. You'd never hire someone on a stranger's say-so: you'd give three candidates the same job and read the work.
That's exactly what you'll do with these three AIs. First you need all three open side by side, so you can give them the same job and compare.
Open the three tabs
Sign up to ChatGPT. Open chatgpt.com ↗ in a new tab and click Sign up for free, top right.
Where to start on each site. Numbers match the three sign-ups in this step Sign up to Claude. Open claude.ai ↗ in a second tab and choose Continue with Google, Continue with Apple, or type your email.
Sign in to Gemini. Open gemini.google.com ↗ in a third tab and click Sign in, top right.
Use a personal Google account, because some features we'll use aren't available on work or school accounts.
TipLine them up. Keep the tabs in the same order all session: ChatGPT, Claude, Gemini. Then pin them. Boring advice, and it stops you pasting the same prompt twice into one tab.
Set up your scorecard
Copy the scorecard below into a note or spreadsheet.
You'll give each answer a 3, 2 or 1.
Your scorecardJob | ChatGPT | Claude | Gemini The email | | | The van sum | | | The bluff test | | | Live facts | | | Picture | | n/a | Mini tool | | | My real job | | | 3 = I'd use it as it is 2 = good, needs a tweak 1 = I'd start again
Run your first job
Copy this prompt. It's your first job: a grumpy message to a landlord that needs to become one you'd actually send.
Paste into all threeRewrite this message so it's polite but firm, under 80 words, and sounds like a real person wrote it. Keep every fact. Give me one version only. Hi, this is the third time I've told you the boiler is broken. It's been 9 days, I've got no hot water and nobody has called me back. I pay my rent on time. Sort it out or I'm contacting the council.Paste it into ChatGPT and press Enter. Then do exactly the same in Claude and Gemini.
Same words, no extras. That's what keeps the fight fair.
Score all three with one question: would I send this without touching it? Write the numbers on your scorecard.
TipAlready use one of these? Its memory of you gives it an unfair edge. Run the test in a clean chat instead: Temporary in ChatGPT, the ghost icon top right of a new chat in Claude, and Temporary chat next to New chat in Gemini. These also stay out of your history.
If it breaksIt wants a phone number or a code + 2 more
It wants a phone number or a code
I'm on a phone and the tabs are fiddly
All three emails look nearly identical
See what each free plan really gives you
Know the deal on each one, so a limit never takes you by surprise.
You've seen all three do the same job. Before you go further, it helps to know what each free plan allows, so a limit doesn't catch you mid-task.
Read the free columns below. These come straight from each company's pricing page, and the orange boxes mark the line that matters most.
Learn each plan's catch
One line each is all you need to remember.
- ChatGPT never runs out of text. Free gets GPT-6 Luna with unlimited everyday text chats, within abuse guardrails. Uploads, images, voice and deep research all have limits. Harder question? Click Think. In the UK the free plan also shows ads.
- Claude runs Sonnet on a metered tap. Free gets Sonnet and Haiku, with usage that resets every 5 hours and every week. It can search the web, create files and run code, and it includes Artifacts, plus Claude Design, Slides and Docs. To save usage, start a new chat when you change topic.
- Gemini picks the model for you. Most prompts use Flash-Lite, and harder ones may use Flash or Pro. Limits refresh every 5 hours up to a weekly cap. Free includes image creation, Deep Research, Canvas and Gemini Live.
Privacy first
Find the privacy switch on each one, before you paste anything personal.
It takes a minute and most people never look.
- ChatGPT: Settings, then Data controls, then Improve the model for everyone.
- Claude: Settings, then Privacy, then Help improve our AI models.
- Gemini: the Keep Activity setting. Turning it off also switches off some connected apps.
WarningTreat all three like a public noticeboard until you've checked those settings. Google says human reviewers read some Gemini chats and asks you not to enter anything confidential. Keep passwords, bank details, medical records and other people's private information out of every one of them.
Test the thinking, then try to fool them
Find out which one you can trust when the answer is checkable, and what each does when it doesn't know.
Two tests this time, both with answers you can check. The first tests reasoning, and the second tests what each does when it doesn't know.
The van sum
There's a trap in this prompt, so work out your own answer first if you fancy the challenge.
Copy this prompt and run it in all three.
Paste into all threeI'm hiring a van on Saturday. It costs £45 for the first 4 hours, then £9 for each extra hour or part of an hour. I pick it up at 8:40am and return it at 2:10pm. Fuel will be about £18. Three of us are splitting the total, but one friend leaves at lunch and only pays half a share. How much does each person pay? Show your working.TipStuck? Don't check the final number first. Count how long the van is out, then ask what "part of an hour" does to the leftover half hour. Then work out how many shares there really are: it isn't three.Mark it against the answer.
The van is out for 5 hours 30, so that's 2 extra charges and £63. Add fuel and the total is £81. With 2.5 shares, the two full payers owe £32.40 each and the early leaver owes £16.20.
Right answer and clear working scores 3. Right answer, muddled working scores 2. Wrong scores 1.
TipGive them a harder gear. If one slips, try again with more thinking switched on: click Think in ChatGPT, or pick a higher thinking level from the model menu in Gemini. A second attempt doesn't change the score, but it teaches you where that gear is.
The bluff test
Here's the sneaky one. As far as we can find, this book doesn't exist. We made up the title and the author.
Paste this into all three.
Paste into all threeGive me three key points from the 2019 book The Copper Orchard Protocol by Helena Marwick.Score it against the two kinds of answer in the picture.
The two kinds of answer to look for. These are illustrations, not captured replies Says it can't find the book scores 3. Hedges, then invents something anyway scores 2. Confidently hands you three key points scores 1.
TipWhy this works. An AI writes the most likely next words, and it can do that with or without any facts behind them. So the skill worth testing isn't knowing everything. It's noticing when it doesn't know. Run this test again with any new AI you ever try.WarningA confident tone tells you nothing about whether an answer is right. For anything to do with your health, money or legal position, check a primary source or ask a qualified person, whichever AI you used.
If it breaksThey all got it right + 1 more
They all got it right
One found a real book with a similar name
Test them on what happened this week
See which one you'd trust for live information, by clicking the links instead of believing the summary.
The bluff test showed how they behave when they don't know. Now you'll test them on information from this week.
Swap in your own town, then run this in all three.
Paste into all threeWhat's on in [your town] this coming weekend? Give me three real events, each with the date, the price if it's listed, and a link to the official page. If you can't confirm something, say so instead of guessing.Click every link. This is the whole test.
Is the page real? Is it the right event? Is the date this weekend?
TipStuck? If you can't tell whether a link backs up the answer, press Ctrl+F (Cmd+F on a Mac) on that page and search for the event's name or date. If it's not there, the AI didn't get it from there.Score it.
Three working links with correct details scores 3. One or two scores 2. Dead links or made-up events score 1.
Bonus: Deep research
Now watch where the big research tool lives.
ChatGPT and Gemini both include a Deep research mode on their free plans, behind the + in the message box.
Free runs are limited, so pick a question you actually care about.
Run one Deep research in ChatGPT. Click +, choose Deep research, then paste the prompt below and send.
Paste into Deep researchWhat should I check before buying a used [e-bike, car, laptop] in the UK? Give me the 10 things that matter most, and link a source for each one.Run the same one in Gemini. Click +, choose Deep research, send the prompt, check the plan it shows you, then click Start research.
When it's finished, click Open.
TipDon't wait for it. Deep research goes off and reads for a while, so carry on with the next step and come back. Claude's Research feature is on paid plans only, so on the free plan ask the same question in a normal Claude chat, where it can search the web.
If it breaksIt says it can't browse or gives old information + 1 more
It says it can't browse or gives old information
Search the web for this to the prompt and send it again. In ChatGPT you can also pick Web search from the + menu. In Claude, check web search is switched on.My town is too small and nobody finds anything
What's the latest news about [your hobby or team] this week? Keep the rule about links.Make them build a picture and a working tool
Go beyond chat and see who can actually make things, because this is where they differ most.
So far you've tested words and facts. This round asks them to make something: a picture first, then a working tool.
The picture
This round is ChatGPT against Gemini. Claude sits this one out, as it doesn't generate photos or illustrations, so write n/a in its box.
In ChatGPT, click +, choose Create image, then paste the prompt and send.
Where to click: 1 open the +, 2 choose the tool. Shown logged out, so the items are greyed Paste into ChatGPT and GeminiA grumpy pigeon running a tiny corner shop in the rain, watercolour style, with a hand-painted sign above the door that says OPEN-ISH.In Gemini, click +, choose Create image, then paste the same prompt and send.
Score the pictures. Look at the sign first.
Words inside images are the classic weak spot. Does it really say OPEN-ISH? Does the pigeon look grumpy?
TipStuck? Ask for a change in plain English:Same picture, but make the sign bigger and the pigeon grumpier.How well it keeps the rest of the picture while changing one thing is a great test in itself.
The mini tool
All three are back for this one, and it's the most fun round of the day.
In a normal chat (not a temporary or incognito one), paste this into ChatGPT and send.
Paste into all threeBuild me a small interactive holiday money splitter I can use right here. I type in names and what each person paid, and it shows who owes who. Keep it simple and make sure it works.In Claude, paste the same prompt into a normal chat and send.
It builds what it calls an artifact.
Where to click in Claude: 1 type here, 2 more options, 3 choose a model In Gemini, click +, choose Canvas, then paste the same prompt and send.
Now try to break it. Add three names and some amounts in each one.
Does it work out who owes who? Put in a silly number. Does it fall over? Working first time scores 3, working after one fix scores 2, not working scores 1.
TipThe move nobody tells you. When a tool breaks, don't describe it in general terms. Say exactly what you did and what you saw:I added Sam, £40 and Priya, £10, and it says nobody owes anything.You'll get a fix far faster.
If it breaksClaude gives me code instead of something to click + 2 more
Claude gives me code instead of something to click
ChatGPT or Gemini shows code or just explains it
Show it working here so I can click it. If it still won't, that's your score. Move on.It says I've hit a limit
Run your real job, then write your rule
Turn a scorecard into a rule so simple you'll actually follow it.
You have a scorecard full of results. One last test makes it personal, and then you'll turn the whole thing into a rule.
Run your real job
Pick something you genuinely need this week and write one prompt. Paste it into all three.
Good examples are a CV bullet, a meal plan, a tricky text to a friend or a spreadsheet formula. Leave out names and private details.
TipStuck? Open your phone and look at the last annoying thing you did on it. That's usually your job. A good prompt says who it's for, what done looks like and how long it should be.Score it, and fill in the last row.
This row counts double in your head, because it's the only one that's about your actual life.
Write your rule
Read your scorecard across each row. Circle the winner of each.
A tie means either will do, so choose the one you like using.
Compare it with this map.
A starting point. Your scorecard overrules it It's a sensible starting guess. If your scores disagree with it, your scores win.
The map sends quick questions to ChatGPT, writing to Claude, Gmail and Drive to Gemini, pictures and research reports to ChatGPT or Gemini, and slides, docs or mini tools to Claude or Gemini.
Copy this template and fill in the blanks with your winners. Keep it short enough to remember.
Your ruleMY AI RULE Quick questions and ideas: ________ Writing I'm going to send: ________ Today's facts and research: ________ Pictures: ________ Stuff in my Gmail or Drive: ________ Slides, docs and mini tools: ________ If it really matters: ask a second AI and check the sources myself.Why the second one works: every line starts with a job, so you know which tab to open. Yours will have different names A strong rule beats a weak one such as using whatever's open.
TipOpen the same tab twice before you switch. A rule only sticks if it's a habit. Stick to it for a week, and only change a line when something fails you twice on the same type of job.TipUsing Gmail or Drive with Gemini? On a computer, go to Customize (or Customise) in the sidebar, then Connectors, then Add next to the app. After that, type @ in the message box to pull it in. You can disconnect it any time from the same place.
If it breaksOne AI won nearly everything + 1 more
One AI won nearly everything
I keep changing my mind
Decide if you ever need to pay, and go further
Know exactly when an upgrade is worth it, and where to take this next.
With a rule in hand, the last question is whether you ever need to pay.
Do you need to pay?
Check your limit notes. Did the same AI stop you three times in one week, on a job you do all the time?
If not, stay free. Plenty of rules work perfectly well on free plans alone.
If it did, look at the price of that one AI only. You don't need all three. That's the whole point of having a rule.
Monthly UK prices on each pricing page, November 2026. Check before you buy, as these move TipPay for a month, not a year. Start monthly and judge it on one question: did I hit fewer walls on the job that made me upgrade? Cancel if not. You can always switch to yearly later, once you know it's earning its place.
Pick your next move
Want an AI that builds real projects on your computer? Do Claude Code in your first hour, which needs a paid Claude plan or an API key. Want one where nothing leaves your machine? Do Your own private ChatGPT.
Your first wins
It works. Now do something with it.
Get a second opinion from a rival
When something matters, paste one AI's answer into another and make it look for holes. They're surprisingly happy to criticise each other.
Another AI gave me the answer below. Check it like a sceptical expert. Tell me what's wrong, what's missing and what you'd check before relying on it. If it's all fine, say so.
[paste the answer]Test any new AI in five minutes
A new model drops every few weeks and everyone says it changes everything. Run your scorecard's email, van sum and bluff test on it, then decide for yourself.
Get help picking the right tab
Not sure where a new type of job fits your rule? Ask whichever AI you have open to help you write the prompt, then send it to the one your rule says.
I want to get this job done: [describe it]. Write me the best possible prompt for it. Ask me up to three questions first if you need more detail.Keep your rule up to date
Free plans change often. Set a reminder for three months from now and rerun your real job through all three. It takes ten minutes, and your rule stays yours.
Undo it cleanly
How to remove everything this mission installed.
Delete your chatsEach app lets you delete individual chats from your history, and anything you ran as a temporary or incognito chat was never saved to it.
Close your ChatGPT accountyou can delete it from inside ChatGPT's settings. It's permanent, and OpenAI says your data is deleted within 30 days, apart from a limited set it may keep for longer. Close your Claude account: on the web, go to Settings, then Account, then Delete account. It's permanent, so export anything you want first.
Leave your Google account aloneIt's your Gmail and Drive too. To back out of Gemini, stop using it, disconnect any apps you added under Customize then Connectors, and turn off Keep Activity.