Why you shouldn't use AI to make an edit request on Wikipedia
There are a lot of ways to make edit requests on Wikipedia.
Clearly so: this September, the outstanding queue of requests broke the threshold of 1,000, and has only gone upwards since. At this moment, there are almost 1,400 outstanding conflict of interest (COI) requests for editors to sort through. Don’t expect them to get to “inbox zero” anytime soon.
The site’s editors are its gatekeepers, so to speak, and they’re busier than ever—sorting through the wide range of asks, and the many ways they’re made. While the culture, policy, and protocol of Wikipedia remain starkly similar to the way they’ve long been, this backlog volume is a recent phenomenon.
Why is this happening?
No one is completely sure how it’s gotten this way, but most speculation about the root of the matter involves the popularization of Large Language Models (LLMs)..
There are two main AI-related factors at play here:
Companies are finding that the LLM-heavy internet has shifted away from Search Engine Optimization (SEO) concerns, and toward a Generative Engine Optimization (GEO) paradigm. In order for companies to get the desired results when users ask chatbots about their companies, they need a strong Wikipedia article, because LLMs scrape regularly from the internet’s largest encyclopedia. So the requests are pouring in.
Another possibility is that the increase in request quantity can also be attributed to the seeming ease of digital labor that LLMs present. In other words: a lot of these requests might be AI-generated.
Why not to use AI for COI requests
Wikipedia editors are, at baseline, on guard. They are steadfast in their mission to keep a neutral point of view (NPOV) consistent across the site, and all the more so when they’re sorting through COI requests.
COI protocol exists so that editors can determine the difference between what’s good for self-interested entities and what’s right for Wikipedia.
They are also, on average, very digitally literate. They’re going to sniff out AI-generated content much faster than most, and when they see it? They’ll be more likely to reject it; which, to be sure, they do a little more than half of the time, anyway, when it comes to COI requests.
Wikipedia is a human space
You might be thinking that this whole process is too dependent on the imperfect feelings of Wikipedia editors: their personal stances on COI-requesting companies, about LLMs, or about the overall state of the changing internet (and the world around it, for that matter).
The reality of the situation is that this is how Wikipedia works. While there are rules to what is and isn’t permissible on the site, those rules aren’t always black and white. They’re rich with grey area, and no one knows this subjective range better than the site’s editors.
Editors who get carried away with the gamesmanship available, here, are sometimes accused of “Wikilawyering.”
It’s an apt phrase. Because, like the legal systems of earth, the Wikipedia editorial process is a mix of hard rules and the art of interpreting and enforcing them.
AI is helpful for a lot of things, but your navigation of Wikipedia will go a lot better if you start with the understanding that this isn’t one of them. While LLM technology has advanced quite a bit in recent years, it’s not yet capable of playing in the highly nuanced, idiosyncratic realm of Wikipedia edits.
How editors know it’s AI
Let’s demystify—because, ultimately, AI detection isn’t just a bunch of hunchwork, intuition, or guessing.
Recently, a Wikipedia editor posted about their adventures pursuing WikiProject AI Cleanup (which is exactly what it sounds like).
In their dispatch, they shared that they were able to get an entire ChatGPT transcript from a user who fessed to using the program to construct their Articles for Creation (AfC) request.
What the transcript showed was that ChatGPT, in conversation with a user trying to draft a passable Wikipedia article, was peddling quite a few misconceptions and limp strategies for the AfC process. These included:
The unfounded notion that a micro-article is more likely to be accepted
A plethora of “citation needed” placeholders
Incoherent word salad, like the phrase “compliance-safe staging area”
The downplaying of the necessity for reliable third-party sources
…then, ChatGPT itself provided the most damning appraisal of its user’s efforts. It reminded them that LLM-generated edit and AfC requests are prohibited on Wikipedia, and that the more effective route would be to personally write the draft.
This is an especially transparent, demonstrative episode of Wikipedia edit requests gone wrong. That’s because this is a rare instance of the requester providing all the evidence. But don’t get it twisted: these are all details that Wikipedians are spotting every day in more covert requests, too.
So, where do we go from here?
The popularity of AI seems like a bigger shift rather than some kind of fleeting trend.
As things are, it’s hard to imagine a world where the Wikipedia COI backlog gets any smaller; it’s also just as hard to see a version of LLMs that are meaningfully more capable of writing effective edit and AfC requests.
If the COI backlog is indeed full of LLM-generated content, it’s a lot of dead wood that Wikipedia’s editors are simply too busy, at the moment, to have chopped through yet. And we don’t even really know how big the forest might get.
And, even if most requesters are doing things the right way? The need for a clean and accurate Wikipedia article is higher than ever, because of the new GEO landscape.
We’re going to need a bigger axe
The supply of active Wikipedia editors is relatively small and stagnant. Since 2020, the number of frequent English language editors (measured as those who make more than 100 edits a month, on average) has gone up and down, but always floated somewhere between 5,000 and 6,000.
Because this editorial hustle is purely a volunteer affair, there’s no reason to expect major change in the workforce, even as demand grows.
So the forest of requests, and all the time-consuming challenges they bring, is probably only going to grow further.
As the GEO internet expands, Lumino can help. Schedule a free consultation, right here!
