AI

AI Slop Is Strangling the Hobby Forums That Trained It

Moderators of woodworking, aquarium, and ham radio communities describe a losing war against generated posts designed to farm ad revenue.

Moderators of woodworking, aquarium, and ham radio communities describe a losing war against generated posts designed to farm ad revenue.

This is synthetic placeholder content. All companies, people, and events in this article are fictional.

The moderators of a 19-year-old woodworking forum noticed the invasion before they could prove it. New accounts were posting plausible, confident, completely wrong advice — recommending interior glue for outdoor furniture, inventing a “standard” mortise depth that doesn’t exist — at a pace no human hobbyist could match.

“The tells used to be obvious,” said one moderator, who has run the forum since 2007. “Now the only reliable tell is volume. Nobody who actually builds furniture has time to post forty times a day.”

Dead Pixel spoke with moderators of eleven hobbyist communities — woodworking, aquariums, ham radio, sourdough, vintage synthesizers — who described the same pattern: a flood of AI-generated posts and replies, seeded by accounts that later edit links into their post history or harvest the community’s trust for affiliate spam.

Eating the seed corn

The irony is structural. Forums like these are precisely the high-quality, human-written corpora that large language models were trained on. Now generated text is flowing back into them, polluting the archive that made the communities valuable.

“Twenty years of real answers from real machinists, and it’s being buried under statistically average noise,” said the operator of a metalworking board who has started requiring photos of members’ actual workshops to register.

Moderation doesn’t scale, but slop does

Every moderator interviewed said detection tools don’t work. AI-text classifiers misfire in both directions, flagging non-native English speakers while missing lightly edited machine output. What works, marginally, is friction: registration interviews, photo verification, posting limits for new members — measures that also strangle the open, searchable nature that made forums useful.

Several communities have given up on growth entirely, going invite-only or unlisted. The public internet, one moderator said, no longer feels worth defending: “We built the commons. The robots grazed it bare. Now we’re fencing off what’s left.”

aiplatformscommunities