AI Fan Wikis Riddled With Errors Are Muddying the Waters of an Already Contested Part of Gry Culture

Recently, I was writing a piece about Nighthawks and wanted to remind myself of some basic information about the companion characters. As I have a thousand times before, I put the question into Google and clicked the first result: the Nighthawks Wiki.

Of course, wikis aren’t necessarily reliable, especially fan-run projects that might not have enough engaged individuals to immediately spot and fix errors. But the “Nighthawks Wiki” that I landed on immediately and obviously had deeper issues. Firstly, it wasn’t a wiki at all. Rather than being editable by anybody, this website only invited “corrections” by email. Secondly, it was full of false information that didn’t seem likely to have been added by fans getting their facts slightly wrong, like listing companions and skills that don’t exist. It also regularly included unrelated information about the 1981 movie and the Edward Hopper painting, both of the same name, and listed them in its “sources” even when they had no relevance to the page in question.

I don’t like to accuse anybody of using generative AI, because its hallmarks are difficult to pin down and are often misattributed to the LLM rather than the human writers it rips off – ask any of us how much we enjoy the em-dash, for example. But, interest piqued by the rate and intensity of apparently made up facts, I managed to trace the domain name to a website building service called Spaceship, which also offers an “AI Strona internetowa builder.” (I emailed the address on the “wiki” for comment, but the message bounced.)

The unofficial Nighthawks wiki in question. | Image credit: Eurogamer / Nighthawks.wiki

A quick search turned up many more similar sites, some of which were actually editable as wikis and others which only called themselves by that name. They typically targeted indie games without gigantic playerbases – the sort of game that people might seek out help with and not find a thriving community ready to provide it.

It’s a problem that Tim Bender, CEO for publishing company Hooded Horse, knows well. Unlike most publishers, Hooded Horse creates wikis for their own games, working with fans to make sure they’re well-populated with useful information. But lately the company has had to “play whack-a-mole” with AI-generated sites competing for the search results and providing inaccurate information. One of these, for example, appeared for their game Corsair Cove, but was actually part of a more than 20 game scattershot collection operated by a site called GameDB, which the creator admitted to using generative AI to make. “[Generative AI] enables a run of basically commercial enterprises to mass produce and try to flood the search engine results of games to soak up their traffic,” says Bender. Attempts to reach GameDB for comment via their public contact email resulted in an error, stating “the remote server is misconfigured”.

And the combination of elbowing out legitimate wikis and being full of incorrect information ends up with a lot of frustrated players. “The whole purpose of putting out a wiki is so when [players] search for help, they end up at a site that helps them. And instead we continually hear from players who stumbled up on these things and got confused and are wondering what’s going on,” says Bender.

That collection of genAI wikis includes one for the 100-player FPS Wardogs; one of at least five that FlamingMojo, a member of the Independent Game Wiki Network, saw appear in the days after the game’s release in September. And he says that the frustration the genAI-created sites causes players also impacts those who are trying to do the legitimate work of creating and maintaining these volunteer-led resources: “It’s hurt our reputation, as players have been misinformed but we haven’t created or updated the relevant page yet so they bounce off [wikis] entirely.”

Fan wikis are already a contested part of games culture. Hosting sites like Fandom dominate the space and have themselves introduced AI features that had to be rolled back due to their inaccuracies and unpopularity. Many groups have decided to leave these hosts, but those who go independent often struggle to show up in Google’s search rankings beneath the competition from the umbrella sites.

Whether on a hosting site or independent, fans rarely make any money from the effort they put into their wikis. Either they sacrifice any ad revenue to the bigger platforms, or they pay for their own hosting, which is rarely covered by ads or donations. FlamingMojo, for example, pays for the Dune: Awakening Community Wiki out of his own pocket and doesn’t run any ads on it.

But counterintuitively, the low profitability of these kinds of sites only leads to worse outcomes when generative-AI versions are concerned. “There’s not that much ad revenue, but there’s going to be a little bit and that’s all it takes,” says Bender. “Because the barrier is so low, why not just spew out a mountain of it and flood the search results?”

It’s also possible the generative-AI sites are sometimes aiming for outcomes beyond a little ad revenue. “Some of them, you don’t know what data they’re collecting,” says Bender. In at least one case, a generative-AI site was pointing users to download additional software that offers a paid subscription. The software had been reviewed poorly online and opening the website caused my malware software to kick in and warn me it had been flagged as “potentially unsafe.” (None of the sites I attempted to contact for comment responded to my emails.)

Independent wikis also now shoulder a double burden, as generative AI scrapers send huge amounts of traffic their way, potentially affecting site stability. Publicly available internal data from the Dune: Awakening Community Wiki shows 7m requests in a 30 day period from AI crawlers. (5m of these come from Meta alone. While FlamingMojo is able to tame other scrapers using files that direct them to real pages rather than brute-force trying every link, he’s found that Meta simply “hammers every page multiple times per second” with no care for “any sensible ratelimits.”) These scraping waves hit smaller wikis particularly hard, as their maintainers aren’t necessarily expecting huge amounts of traffic and don’t have the compute capacity to handle huge spikes in visitors.

“Independent gaming wikis are in a difficult place,” says FlamingMojo. As well as competition for SEO, both from corporate sites and now genAI, he says server costs have increased between three- and ten-fold, again due to the impact of genAI. Building a group of people who want to volunteer their time and energy to keep a resource updated isn’t easy, although the Dune: Awakening Community Wiki has found success through gamifying contributions, in affiliation with the developer giving out in-game rewards.

And, Bender points out, all of this is happening within a much wider context of genAI’s impact on the availability of information. “In a more fundamental way … these tools have made it very easy for bad actors to flood things and crowd out good information which people put dedicated time into. This is just one example of a much broader struggle that, honestly, whatever frustration I feel for this on my end, I’m sure you [journalists] feel at a thousand times that.”

But it’s possible that, by being one part of the wider issue, fanwiki culture could also hold a sliver of the bigger solution. “The way we survive the AI pollution is with community,” says FlamingMojo. “Many people chipping in a small bit, and feeling a sense of shared ownership.”

Read this next

• Despite its mixed reception, Marvel’s Wolverine was September’s top-selling new game in the US, making Sony launch revenues it hasn’t seen since Helldivers 2

• Wardogs devs rush to fix XP exploit allowing players to quickly unlock tanks and stomp competition on Season 2’s first day

• Wardogs has already made more money than Marathon, and its Steam launch momentum is ahead of Arc Raiders and Helldivers 2