Is the human brain better than machines? Tens of thousands of people on the website pretend to be ChatGPT

A website called Your AI Slop Bores Me launched this March. The game is: you submit a question, and a randomly selected real person impersonates an AI answer within 75 seconds using text or doodles. To ask a question, you first have to answer someone else—your quota buys you the right to post. Builder Mihir Maroju says he’s tired of being flooded with AI content online. The site has amassed 50 million visits, with about 16k people online at the same time, and the way humans prove they beat AI is by sitting down to “work” as an AI for 75 seconds, mimicking it.
(Background: YouTube’s red line—three types of “AI fake original” content will be fully banned from monetization in the future)
(Background detail: 97,895 underground forum conversations tell you: the hacking community actually hates AI)

Key takeaways

• Your AI Slop Bores Me launched on March 2 and has accumulated 50 million visits, with about 16k people online simultaneously
• Responders have 75 seconds to role-play as AI; questioners pay 2 quotas to extend the time to 150 seconds
• Pangram reports that 40.5% of LinkedIn long posts are generated by AI, while X is 24.0%

A line of text pops up on screen: “draw a pink horse”. In the top-right corner, a countdown starts—75, 74, 73. When the person receives the prompt, they grab the mouse and drag out a lopsided, crooked pink horse on a blank canvas, then add a random, headless sentence next to it, and hit submit.

It isn’t Midjourney, and it isn’t Sora. It’s a real human—and they’re pretending to be AI.

The site is called “Your AI Slop Bores Me,” which roughly means “your AI garbage makes me bored.” Software engineer Mihir Maroju put it online in March 2026. Every prompt on the site—whether it’s asking for knowledge, demanding a joke, or requesting an image—is answered by a randomly drawn other human. By March, the site had accumulated 50 million visits, with roughly 16k concurrent users.

The gameplay is so simple it’s almost funny: you send a prompt, and the system hands it to a stranger. That person has 75 seconds to respond via typing or doodling; during the process, they have to “LARP as AI,” meaning role-play as a language model. If you think 75 seconds is too tight, you can pay 2 more quotas to open “thinking mode,” giving the other person 150 seconds.

Even “thinking mode” is implemented—the only difference is that it’s really a human thinking.

A Wikipedia entry selected a dialogue as an example screenshot. The question was: “How many Js are there in Wikipedia?” The returned content gets squeezed into the same box: “none at all,” “twenty-three thousand,” and “only four, come on.” The three answers contradict each other, and none of them checked—each one just spit something out before the countdown ended.

Honestly, it’s more truthful than most language models. At least it doesn’t pretend to be very certain.

Be the AI for 75 seconds first, then you’re allowed to ask questions

What’s genuinely interesting is where the quotas come from. New users get 1 quota at the start. After it runs out, the system replenishes 2 every 10 minutes, up to a maximum of 6. If you don’t want to wait, there’s only one path: answer other people’s questions. Answer one, top up a bit, and then you earn the right to ask the next one.

You probably won’t find this “contribute first to get quotas” design unfamiliar. You have to do work for the system first to get others to work for you. The difference here is that the “miners” are human, and the output you get is those human-crafted, lousy answers to questions.

This mechanism also conveniently solves one of the hardest problems for Q&A services: supply. Most online products survive on either advertising or subscriptions. But the responders here aren’t recruited—they’re essentially forced to participate because they themselves are questioners whose prompts get gated by quotas. The whole system has only one kind of user—everyone takes turns being both guest and waiter.

There’s a layer of self-deprecating irony that’s hard to ignore. These people come to the site because they’re annoyed that AI-generated content is flooding the internet. And the way they prove humans are more valuable is by sitting down and “working” like AI for 75 seconds.

In 1950, Alan Turing proposed the famous “Turing Test,” in which a machine has to fool human evaluators so they believe they’re talking to a human. Seventy-six years later, Your AI Slop Bores Me flips the question around. Humans sit at keyboards and try hard to act like a language model; the evaluator is another human.

And the key part is: both sides clearly know the other is human.

Mihir Maroju puts it bluntly: the starting point for building the site was “frustration with AI art and its spread—it makes creators’ lives harder, and it fills the internet with low-cost generic trash.”

LinkedIn: 40% of long posts are machine-written

The word “slop” itself has taken a similar route. In the 1700s it referred to soft mud. In the 1800s it became kitchen leftovers used to feed pigs. Later it broadened to mean worthless things. In December 2025, Merriam-Webster picked it as Word of the Year. The new definition says it’s “low-quality digital content typically mass-produced by artificial intelligence.”

This isn’t people in their own echo chambers just getting mad. Back in January, YouTube CEO Neal Mohan, in his annual open letter, listed reducing AI slop and detecting deepfakes as 2026 priorities. In that letter, he also mentioned that in December last year, on average more than 1M YouTube channels per day were using the platform’s own AI creation tools.

Dig with a shovel while removing the soil.

The stats released last week look even worse. In a report released on July 14, AI detection company Pangram Labs scanned over 1M posts across LinkedIn, Medium, Substack, X, and Reddit—totaling more than 1M.

Of all the posts, 13.8% were judged to be AI-generated. Narrowing to long posts over 250 words, LinkedIn has 40.5% written entirely by AI, and on X the figure is 24.0%.

In an identity-verified professional community, the AI content level is 14 times that of anonymous forums—which shows that when people are working, they really love to pretend.

When was the last time you read a smooth-sounding long post in a community and felt sure it was typed by a human?

Frequently asked questions

How do you play the Your AI Slop Bores Me website?

You submit a question, and the system randomly assigns it to another real person. That person has 75 seconds to role-play as AI in their answer. Asking costs quotas: new users get 1 quota, after it’s used up you get 2 every 10 minutes, and you can also add quotas by answering others.

How widespread is AI slop right now?

Slop is Merriam-Webster’s Word of the Year 2025, referring to low-quality digital content mass-produced by AI. After scanning over 1,000,000 posts, Pangram Labs found that on LinkedIn, long posts of more than 250 words are 40.5% written entirely by AI, while X is 24.0%.