"Your AI Slop Bores Me": what the joke site can and cannot tell us about machine voice

Your AI Slop Bores Me is a browser game in which people answer each other's prompts while pretending to be an AI. It shows that readers can recognise and imitate a machine voice. It does not measure whether the web's writing has converged on one.

By Jack Stovell · Published 13 October 2026

Short answer: the game is good evidence about readers and no evidence about the web. It shows that people carry a working model of the chatbot voice in their heads. Whether the web’s writing has actually converged on that voice is a separate question, and it needs a different kind of evidence.

What Your AI Slop Bores Me is

Your AI Slop Bores Me is a website that went viral in March 2026. It looks like a chatbot. You type a request for an answer or a drawing, and another user, picked at random, answers it as if they were the AI. Every reply comes from a person. Drawings are done with a mouse or trackpad, and they look it.

NPR reported in April 2026 that it was made by Mihir Maroju, a 17-year-old in Puducherry, India, and that the name comes from a social media meme aimed at AI-generated content. The homepage offers two roles, “human” and “larp as ai”. Answers run against a short timer, so the human has to keep pace with a machine.

Gizmodo’s Tom Hawking tried it for a piece published on 12 March 2026 and called it “weirdly compelling fun”. He also spotted the inverted economy: “you have to spend time asking questions to get more chances to answer them.” The site keeps its own blog, which describes its subject as “ai roleplay, human creativity, and the fight against generic slop.”

I have read the coverage and the site’s pages. I have not sat a long shift answering prompts, so nothing below rests on my own rounds.

Why the joke lands: readers know the voice

Nobody has to explain the premise. You read a prompt, you put on the voice, and the person at the other end recognises it at once. The cheerful opener, the tidy list, the closing offer to help further. Players can produce it on demand, often faster than they could name its parts.

That is two skills at work. Recognition: readers can tell the voice when they meet it. Imitation: they have absorbed its habits well enough to fake them. A parody only works when both are common, because you cannot send up something nobody has noticed.

Where those habits came from is a separate matter, and one the court has heard. The case Not an invention. An inheritance. traces much of the vocabulary to human writing that came long before ChatGPT.

What a game can show

Entered as an exhibit, the game supports a few modest claims. The chatbot voice exists as a shared idea. Ordinary readers can describe it by performing it. And enough people find that funny for it to spread.

Those are findings about people: what readers know and what they enjoy. They are worth having. They are also anecdote. You see a handful of answers, you laugh, and you come away feeling you have learned something about the state of writing. You have learned something about the state of readers.

What it cannot show: recognition is not measurement

A vivid example and a measured sample can look alike on the page. Both arrive as “some text”. They are different kinds of evidence.

A measured sample has a question fixed before anyone looks, a pool of text chosen by a stated rule, and a count taken the same way across the whole pool. A vivid example has none of those. Somebody picked it because it was striking.

The game fails each test, by design. Nobody chose the prompts by a rule. Players are chasing a laugh, so the answers exaggerate the voice. The answers that travel on social media are the funniest ones, picked by the people sharing them. And no part of the game touches the web’s actual writing. It samples people’s impressions of a chatbot, written in a hurry, for fun.

Recognition says “I know it when I see it”. Measurement says how often it appears, in what, compared with what, and when. The first is a reliable way to spot a voice. Only the second can tell you whether the voice has spread.

Prevalence, convergence and a funny website

Two questions get tangled here, and a joke this good tangles them faster.

Prevalence asks how much of the writing out there was produced by a machine, or carries a machine’s features. Convergence asks whether writing has grown more alike: different writers, on different subjects, drifting towards one sound. A study measures one or the other, and a careful reader checks which.

The game nudges you towards both without earning either. You see the voice parodied everywhere and conclude it is everywhere. Then you conclude everything must sound the same. Neither step follows from a parody. The theory page sets out the claim on trial and keeps the two questions apart.

How to read the next viral demo

There will be another one. Four questions sort most of them.

What does it show? People’s ability to spot or fake a style, or the state of the writing itself?
Who chose the examples? If the funniest ones reached you first, you are seeing a highlights reel.
Was a sample defined before anyone looked? If not, it is an anecdote, however good.
Which claim is in play? Prevalence and convergence need different evidence. A demo that seems to prove both has usually proved neither.

If the answer to the third question is “no, it’s a gag”, enjoy the gag. Just leave the conclusion in the room.

Where the evidence is

The cases on this docket are where the measuring was done. I will not restate their numbers here; the point is to see how each was reached.

On convergence, start with The stories grew alike. Each one got better., a preregistered experiment. Its authors put the finding in one line: “With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced.” For the defence, The fingerprints are there. The flattening is not. found machine-style words rising in English news without lexical diversity falling.

On prevalence, which is a different question, see Half the articles. Then the rise stopped.. It is a vendor study built on a detector, and the case page carries both caveats.

Read them alongside the game. The game shows what readers think the voice sounds like. The cases show what was counted.

Sources

Questions people ask

What is Your AI Slop Bores Me?

A website that went viral in March 2026. It looks like a chatbot, but every request for an answer or a drawing is answered by another user, picked at random, who replies as if they were the AI.

Who made Your AI Slop Bores Me?

NPR reported in April 2026 that it was made by Mihir Maroju, a 17-year-old in Puducherry, India.

Is Your AI Slop Bores Me real people or AI?

Real people. Every answer and drawing comes from a human playing the part of the AI, usually against a short timer.

Does the site prove that AI writing all sounds the same?

No. It shows that readers can recognise and imitate a chatbot voice. That is recognition. Whether the web's writing has converged needs a defined sample and a count, which a game does not provide.

How do researchers actually measure whether writing has converged?

They fix the question before looking, draw a pool of text by a stated rule, and compare how alike the texts are across writers or over time. The cases linked in this post show this in practice, and each one says whether it measures prevalence or convergence.