AI fan wikis are eating the search results games rely on, and the errors are the point

A “Nighthawks Wiki” sits at the top of Google results for the game, and it is not a wiki. It takes no edits, only emailed “corrections,” and its pages list companions and skills that do not exist, plus casting notes for the 1981 film and references to the Edward Hopper painting. In a Eurogamer feature published October 27, Jay Castello traced the domain to Spaceship, a site-building service that advertises an “AI Website builder.” The email address on the wiki bounced.

That single bad site would be a curiosity. What Castello found is a system. Dozens of these pages target indie games too small to have a thriving community wiki, and they compete for the same search queries that real fan projects depend on. One such site, for Hooded Horse’s Corsair Cove, turned out to be a node in a collection of more than 20 games run by a site called GameDB, whose creator admitted to using generative AI to build it. Hooded Horse CEO Tim Bender describes the dynamic as “play whack-a-mole.”

The take here is not that AI writes bad wikis. It is that the economics of the open web have flipped against the people doing real work, and AI is the accelerant.

The cost floor collapsed, and the search results filled up

Bender’s diagnosis is the sharpest line in the piece. “There’s not that much ad revenue, but there’s going to be a little bit and that’s all it takes,” he says. “Because the barrier is so low, why not just spew out a mountain of it and flood the search results?”

That is the mechanism. Fan wikis were never lucrative. FlamingMojo, a member of the Independent Game Wiki Network, pays for the Dune: Awakening Community Wiki out of his own pocket and runs no ads. The legitimate projects are labors of love funded by donations and stubbornness. The fake ones need only a trickle of ad impressions to justify a domain and a template. When the marginal cost of producing a plausible-looking page drops to near zero, the low-margin end of the information web gets buried in garbage. The good stuff does not lose because it is worse. It loses because it cannot outproduce a script.

And the harm is not just wrong facts. Bender notes that some of these sites collect data he cannot account for, and at least one pushed users toward a paid subscription for software that Castello’s malware scanner flagged as “potentially unsafe.” A Eurogamer editor who tried the Nighthawks wiki got redirected to spam URLs. The fake wiki is sometimes a front for something worse than a bad answer.

The real wikis are also getting DDoSed by crawlers

Here is the part that should worry anyone building on the web, not just games fans. The Dune: Awakening Community Wiki’s own internal data shows 7m requests from AI crawlers in a 30-day window. Five million of those came from Meta alone. FlamingMojo says Meta “hammers every page multiple times per second” and ignores sensible rate limits, while other scrapers can be tamed with files that point them at real pages instead of brute-forcing every link. Server costs for the wiki have risen between three- and ten-fold, again because of genAI.

Read that again: the same volunteer project that is losing search traffic to AI slop is simultaneously paying more to serve the crawlers that will feed the next generation of models. The people maintaining the actual knowledge are subsidizing the machines that are about to replace them in the results page. That is not a fair trade, and nobody negotiated it.

What this says about the AI pipeline

The industry likes to talk about data quality as a training problem. This story is about data quality as a supply chain problem, and the supply chain is being poisoned at the source. If the models that scrape the web are ingesting the AI-generated wikis alongside the real ones, the next round of models learns from the fakes. The Nighthawks wiki listing nonexistent companions is not just a nuisance for a player. It is a data point in somebody’s crawl, and the crawl does not check whether a page was written by a fan who played the game or a template that scraped a movie database.

Fandom, the dominant hosting platform, already rolled back AI features after inaccuracies and backlash, per Castello. That is the tell. Even the company with the most to gain from automated content found that automated content did not hold up. The umbrella sites have the resources to retreat. The independent maintainers do not.

The people maintaining the actual knowledge are subsidizing the machines that are about to replace them in the results page.

The fix is not a better model

FlamingMojo’s answer is the one worth carrying out of the piece. “The way we survive the AI pollution is with community,” he says. “Many people chipping in a small bit, and feeling a sense of shared ownership.” The Dune: Awakening wiki gamifies contributions in affiliation with the developer, handing out in-game rewards for edits. That is a real mechanism, not a vibe. It aligns the incentive of the player with the health of the resource.

For AI builders, the lesson is less comfortable. Search ranking is the choke point, and whoever controls it decides which wiki a player lands on. Google has spent years fighting spam, and it is now losing ground to pages that are cheap to generate and hard to distinguish from the real thing at a glance. The crawler behavior matters too: a company that sends 5m requests a month at a volunteer wiki and ignores rate limits is externalizing its compute costs onto the smallest operators on the web. That is a policy question as much as an engineering one, and no regulator is currently looking at it.

Bender frames it as one front in a wider war. “These tools have made it very easy for bad actors to flood things and crowd out good information which people put dedicated time into,” he says, and he grants that journalists feel it worse. He is right about the breadth. The fake game wiki is the same pattern as the fake recipe site and the fake product review, just with a smaller audience and a more obvious tell.

The tell, in this case, is a page that cites a Hopper painting as a source for a video game character. A human fan would never make that mistake. A template will make it every time, and the search result will still show up first.