I Built a Game to Test Whether You Can Tell Human Writing from AI
You probably cannot. That is the point.
This post is brought to you by Matiks, Slow AI’s first paid sponsor.
I chose Matiks because it fits what this post is about: testing your own judgement under pressure. Matiks runs one-minute duels that make you think smartly and carefully. The same pattern recognition this game asks of you with quotes, Matiks asks of you with numbers. No mindless scrolling. No brain rot. Just sharp, focused play. Free to download.
Here is a question I have been sitting with for months: if you cannot reliably tell whether a piece of writing was produced by a human or a machine, what does that mean for how you read?
Not how you detect. How you read.
I built a small browser game to explore this. It is called Bot or Not. You read ten quotes. Some were written by humans. Some were generated by AI. You guess which is which.
Most people score close to a coin flip.
In this post we will:
Offer a browser game that tests whether you can distinguish human writing from AI.
Examine why AI detection tools fail and what that means for how you read.
Share how the game was built entirely with Claude Code by someone who has not written production code in a decade.
Play it
Bot or Not runs in your browser. No accounts, no tracking, no cookies. It takes about five minutes.
If you score badly, sit with that. The discomfort is more useful than a high score.
What the game actually does
Each round shows you a single quote. You press Human or Bot. Then you see whether you were right, who wrote it, and a short reflection on what made that particular quote hard to place.
The reflections are the point. The score is secondary.
One reflection might examine why a Borges quote reads like something a chatbot would say about itself. Another might ask why an AI-generated sentence about loneliness feels so convincingly human. Each one picks apart a specific rhetorical pattern: the borrowed metaphor, the performed vulnerability, the missing specificity, the suspiciously tidy structure.
In total there are 200 quotes, 100 from humans and 100 from AI. The game only asks you a random selection of 10 in any one play-through.
The game does not teach you a detection method. There is no reliable trick for spotting AI text, despite what some people claim. What it does is slow you down. It asks you to notice what you are actually responding to when you read something.
Why AI detection does not work
I keep seeing two unhelpful reactions to AI-generated text.
The first is panic. People want a way to catch it, flag it, filter it out. They want AI detectors that work. The detectors do not work. They are biased against non-native English speakers and they produce false positives at rates that make them useless for any serious purpose.
The second is indifference. If you cannot tell the difference, who cares? The text is fine. Use it.
Both reactions skip the interesting question. The interesting question is: what happens to your judgement when you cannot tell whether what you are reading was thought or generated? What do you trust? What do you doubt? On what basis?
Bot or Not does not answer that question. It puts you inside it for five minutes and lets you feel how uncertain it actually is.
How I built a browser game with Claude Code
I wrote code during my PhD. Atmospheric physics, satellite data, that sort of thing. That was over a decade ago. Since then I have not written a line of production code. The skills rusted. The ambition for building things did not.
I built Bot or Not entirely with Claude Code. The whole thing: HTML, CSS, JavaScript, quote verification, game logic, the lot. I described what I wanted. Claude Code wrote it. I tested, broke things, described what was wrong, and it fixed them. The entire game, from first file to itch.io upload, was built in a few sessions.
I want to be honest about what that felt like. It felt like having a competent collaborator who never gets tired and never judges you for forgetting how arrays work. It also felt like something I need to be careful about. I am building a game about critical AI literacy using AI. The irony is not lost on me. But the game exists now, and it would not exist without the tool.
I also want to acknowledge the people in this community who gave me the confidence to actually try. Daria Cupareanu, Elena | AI Product Leader, Karen Spinner, Karo (Product with Attitude), & Jenny Ouyang have all been building with AI in ways that made me think: I could do that. Not because they made it look easy, but because they were honest about the process.
I also want to give a special thank you to ToxSec, whose peerless writing on cybersecurity informed the routine that I used to red team Bot or Not until it was safe and stable.
What to share
Share your score either in the comments or as a Substack note and tag me in. The game generates an image that lets you do this directly from the results screen. But more than the number, share which quote fooled you and why. Notice whether you trusted a certain kind of sentence structure or distrusted a particular emotional register. Observe what your wrong answers tell you about the assumptions you bring to anything you read.
The work here is not improving your detection rate. It is seeing what you already believe about writing before you know who wrote it.
If you play Bot or Not, I would like to hear which quote caught you out and what you noticed about your own reading habits.
I read and respond to all your comments.
Go slow.
A paid subscription to Slow AI provides access to The Slow AI Curriculum for Critical Literacy. This twelve-month programme is for individuals who want to understand how to engage critically with AI rather than just use it to generate outputs.
If you join now, you join the founding cohort. That group will shape the tone and direction of the curriculum going forward.
The 2026 Handbook for Critical AI Literacy provides the necessary detail for those who wish to make an informed decision before subscribing.


Really interesting Sam, thank you for the game and the write-up!
I’ll argue that the game actually *did* teach me a detection method as I went, though not 100% reliable:
The model-generated responses, though eloquent and well-reasoned, lacked any sense of what I would call personal experience.
I missed 3 near the beginning then went on a hot streak to finish with 7/10. I asked myself, “does this quote have any indication that someone did, or felt, anything?”
Despite this working today, I don’t expect it to be durable. Which is the lesson I’ve taken away after reflection - that any detection methods we think might work today won’t last for long.
Thanks again for the interesting exercise!
Is it odd of me to think that I prefer the clear logical process that AI follows as compared to human beings and the possibility of written words by human beings more performative and sometimes "abstract" only for the purpose of exhibiting nuance? I wonder, is this shift "right" or "wrong"? I find myself increasingly dropping dense sentences written by theorists, researchers into AI to help me unpack words increasingly... Is that a well placed thought process?