When forums cry 'AI slop,' the charge seems to track who belongs, not how the text was made.

This explores why 'that's AI slop' accusations act as social policing rather than real AI detection. The clearest evidence comes from a study of 25 million Hacker News and Reddit comments, each compared against similar comments that weren't accused. The writing features that actually separate AI text from human text did not predict which comments got accused Do AI slop accusations actually detect AI text?. So whatever people are reacting to, it isn't the statistical fingerprint of a language model. The label works like a bouncer at the door, not a lab test.

Part of the reason is simple: people can't do the detecting. A review of 30 studies found that human accuracy at telling AI content from human content, across text, images and voice, generally sits around chance. It hasn't kept up as AI output has become more realistic Can people reliably spot content made by AI?. If you can't actually tell, an accusation can't carry information about where a text came from. What it can carry is a judgment about tone, effort, fit or status, so the word 'AI' ends up standing in for 'doesn't belong here.' Machines make a related mistake. Fake-news detectors flag truthful LLM-written articles as fake and let human-written disinformation through, because they have learned to treat AI's writing style as a sign of deception Why do fake news detectors flag AI-generated truthful content?. Humans and classifiers alike end up reacting to surface style and treating it as evidence of something deeper.

There's also a structural reason why the question 'is this AI?' is hard to settle from the text alone. One note argues that AI output has the same shape as pre-Enlightenment hearsay: no clear origin, changed with each retelling, and impossible to check against a stable source. The usual verification tools, like citation and evidence chains, can't get a grip on it Does AI-generated knowledge have the same structure as hearsay?. A related point comes from work on persuasive AI explanations: intent and purpose are invisible in the finished text, so a helpful explanation and a manipulative one can look the same Can we distinguish helpful explanations from manipulative ones?. When what you care about (who wrote it, and why) can't be read off the page, communities fall back on what they can enforce, which is norms about how people are supposed to sound.

The part you might not expect is who pays for this. If accusations don't track AI use, most of the people accused are likely humans whose writing simply reads as off. One note frames this as testimonial injustice turned around: the usual worry is AI deceiving readers, but here readers' suspicion wrongs human writers by discounting what they say Do unfounded AI accusations harm human writers instead?. Even when AI involvement is real and openly disclosed, suspicion is a weak defence. Disclosure makes audiences more critical, yet 34–62% stay persuaded anyway Does telling people an AI wrote something actually stop them from believing it?. So the slop label fails in both directions: it lands on humans who did nothing, and it does little to blunt AI content that actually gets through.

One gap: the collection shows that accusations don't track AI features, and gives good reasons why detection is out of reach. It doesn't directly explain what accusers are responding to, whether that's newcomers, non-native English, a certain polish, or unpopular opinions. That question stays open in the corpus.