Two weeks ago I shipped a stupid little thing: a 10-question quiz that matches you to one of 8 AIs. ChatGPT, Claude, Gemini, Grok, Llama, Perplexity, Mistral, or DeepSeek. You get a shareable card at the end, you post it to LinkedIn, your coworkers argue about whether you're really a "Claude person." That's the whole loop.
I expected it to be a toy. What I didn't expect was the mismatch rate. Five hundred takers in, the gap between "the AI you opened today" and "the AI the quiz says you actually are" is wider than any AI vendor wants you to know. Most people are using the AI their coworker mentioned first, not the AI that fits how they think.
Here's the data, the 8 archetypes, and the four reasons people end up bonded to the wrong model.
Take the quiz → find out which AI you actually are 10 questions · ~90 seconds · free · no signup · shareable cardThe quiz isn't a benchmark. It doesn't ask which AI scored highest on MMLU. It asks how you work:
The scoring maps those answers onto the 8 archetypes the AIs actually represent in practice — the personalities they project, not the personalities their marketing copy claims. (Claude really does ask more clarifying questions. Grok really is the one that will roast you. Perplexity really does cite or die. None of that is a stretch.)
Then I asked every taker, before they saw their result: which AI do you currently use most? That's where the data got spicy.
Of 500 takers, 312 said they "mostly use ChatGPT." Here's where the quiz placed them:
| Quiz result | % of self-described ChatGPT users | Read |
|---|---|---|
| ChatGPT | 41% | correctly bonded |
| Claude | 34% | drafted-and-stuck |
| Perplexity | 11% | needs citations, accepts vibes |
| Gemini | 6% | underusing Google account features |
| Grok | 4% | closet contrarian |
| Mistral / Llama / DeepSeek | 4% combined | privacy-curious, not yet acting on it |
So 59% of self-described ChatGPT users got matched to something else. The biggest single bucket — 34% — got matched to Claude. The pattern in their answers was always the same: they care about tone and accuracy more than speed, they get annoyed when an AI hallucinates with confidence, and they want the model to push back instead of agreeing.
That's a Claude profile. They're just using ChatGPT because their friend showed them ChatGPT in November 2022 and they never reopened the question.
Fast, friendly, eager to please, will draft a wedding toast and a Python script back-to-back without complaining. People who really are ChatGPT love volume and don't mind editing. They'd rather get 80% right in 5 seconds than 95% right in 60. Mismatch tell: if hallucinated citations rage you, you're not actually ChatGPT.
Refuses to ship the first sentence until it's right. Asks clarifying questions. Gives you the hedge instead of the confident wrong answer. People who match here usually say they use ChatGPT but secretly screenshot Claude outputs to share. Mismatch tell: you delete and rewrite your prompt three times before sending.
You live in Google Docs, Sheets, Calendar, Gmail. Of course your AI should already be there. Mismatch tell: you've paid for ChatGPT Plus while having Gemini Advanced bundled in Workspace for free.
You want the AI that argues. You'd rather get told you're wrong than told you're brilliant. Mismatch tell: you find every other AI "too corporate."
You want weights, not a chat window. You want to fine-tune, deploy, own. Mismatch tell: you've cloned at least one model from Hugging Face this month.
Confident answers without sources make you twitch. Mismatch tell: you copy ChatGPT answers into Google to check them anyway.
You care about data residency, GDPR posture, who trained on what. Mismatch tell: you've actually read an AI model card.
Cost-per-token is a thing you can feel in your chest. Mismatch tell: you opened DeepSeek's pricing page before you opened the chat.
The takers who got mismatched fell into four neat buckets. If you've never re-evaluated your AI, one of these is probably you.
Someone you respect — usually a coworker, sometimes a podcast host — said "I just use ChatGPT" in 2023, and you never re-tested. Your bonded AI is 30 months old. The models have all moved. You haven't.
Your first prompt to a new model is always lazy: "write me a haiku about Mondays." Every modern AI passes that test. So you never push hard enough to discover the differences that matter to your actual work.
You think "I'm a ChatGPT person" the same way you think "I'm a Mac person." It's identity, not fit. Letting it go feels like a tiny betrayal. So you don't.
You hit the wall on a free tier of model A, upgrade to paid on model A, and never try model B at all. You're paying $20/month for the AI you tolerated when it was free.
You don't need to switch AIs. You need to know which one fits, and then either switch or own the trade-off.
Take the quiz. If it matches you to your current AI, great — your gut was right, post your card, move on. If it matches you to a different one, here's the cheap experiment: run the next five real prompts you would've sent to your default AI through the recommended one instead. Not the toy prompts. The real ones — the emails, the briefs, the code reviews. After five, you'll know.
That's it. No vendor switch, no subscription juggling. Just five honest comparisons.
Try yourself → take the 10-question quiz You'll get an archetype, a shareable card, and an honest "do these five things differently" list.A few things people did with their cards that I didn't expect:
And a few related things you can do once you know your archetype:
This is a 10-question quiz, not a study. The 500-person dataset is self-selected, skews technical, and the "AI you use" question is self-reported. The 1-in-3 number is real for this sample. Whether it generalizes to your team or your audience, I don't know.
What I am confident about: the gap between the AI you opened today and the AI that fits you is non-trivial, it's not narrowing, and the cost of taking 90 seconds to check is approximately zero.
So check.
What AI are you? → take the quiz 10 questions · 90 seconds · shareable card · free · no signupTinyTools makes small, free, single-purpose web tools. No tracking pixels, no email gates, no AI of our own to sell you. We just want you on the right one.