Why we're pulling back the curtain on AfC
Spoiler alert: We analyzed 1,009 submissions and want to share our data
This newsletter is not all fun and games. Occasionally we do real research! To wit: Lumino, the digital agency behind the Wikify empire, just published the first large-scale empirical audit of Wikipedia’s Articles for creation (AfC) workflow, which is where newbie users + brands & businesses go to submit new articles for publication.
Our crack team of researchers analyzed 1,009 draft submissions that survived initial triage during two sampling windows in late 2025. What we uncovered about approval rates, LLM usage, and notability guidelines will shock and amaze you.
The full report can be downloaded here and we discuss key findings ahead.
But first, let’s explain why AfC matters—and why we’re pulling back the curtain on how brands get involved.
What’s driving AfC submissions?
There is an enormous demand for new article creation, particularly from companies and prominent individuals (especially CEOs, founders, and other executives). A Wikipedia article not only conveys credibility, it also substantially influences Google search results and generative AI answers.
To satisfy this market need, dozens of vendors—from individual Upwork contractors to SEO firms to large public relations agencies—either offer article creation services directly or outsource the work through white-label arrangements.
A substantial percentage of these Wikipedia vendors appear to be illicit operators who promise results they cannot deliver. There is an entire Wikipedia information page about these scams, and you can search across Reddit or Quora to uncover horror stories of how these vendors operate. In some cases, the vendor ends up publishing the draft to a lookalike domain.
Vendors like this are proactive in drumming up business, including by monitoring Wikipedia’s Articles for creation (AfC) queue for rejected submissions and then reaching out to the brand or business in question to offer their services.
When someone contacts us about creating a new article, we immediately outline the challenges involved and note that Wikipedia editors decline the overwhelming majority of AfC draft submissions. That claim reflects our years of experience with the site, but it is still anecdotal. In truth, we’ve never known the exact percentage of article submissions that are declined, as there’s limited research on the subject.
Just guessing wasn’t good enough for us, so we started collecting a dataset of viable AfC submissions. Once we had 1,000+, we aimed to calculate the approval percentages, identify how those percentages varied by submission category, and assess how relevant content guidelines were applied in practice.
Here’s what we learned.
Most submissions are declined
The first thing we learned was that a majority of AfC submissions (68%) are indeed declined.
We were more surprised by the relatively high approval rate (27%) for new articles, given Wikipedia editors’ general “deletionist“ disposition. However, we should note that our dataset excluded submissions that failed the initial quick-fail review that screens out glaringly low-quality drafts. If you included those lemons the approval percentage would be even lower.
We also dug deeper into AfC approval rates by category and even sub-category (download the full report to see those). Arts & Culture submissions had the highest approval rate of any category (48%), while Business submissions had the lowest (15%). Narrow in even further on just tech companies, and the approval rate drops to 6%. Ouch.
The top reason for rejection is a lack of notability
Notability is a core test among Wikipedia editors for deciding if a topic merits its own article, and a core test of humility among finance executives and tech startup founders across the globe.
Wikipedia editors cited “lacks notability” as the primary reason for decline on over half of declined submissions in our dataset. The second-most cited reason for decline was that the submission lacked reliable sourcing.
A small group of editors manages a large number of requests.
A total of 219 editors reviewed the AfC submissions in our sample, but a majority of those (112 editors) reviewed just one submission.
The bulk of the work reviewing AfC submissions was carried out by 11 especially active editors who reviewed 20 or more submissions each. This core group was responsible for approving or declining approximately 40% of submissions in the sample. #SquadGoals
They weren’t lenient, either—this group approved 23% of submissions, compared to the 27% approval rate for all submissions in the sample.
Speaking of editors, they’re sick of reading AI slop.
Last month, Wikipedia’s editor community took the bold step of banning the usage of AI to generate new articles or rewrite existing ones. Our dataset underscored the challenge this technology poses to the encyclopedia: LLM use was the third-most cited reason for decline in our sample, after “lacks notability” and trailing only slightly behind “lacks reliable sourcing.” And if we consider editor feedback on past submissions, then 16 percent of AfC drafts were flagged for AI/LLM concerns.
AI usage was especially prominent in technology-focused business drafts: 40 percent of these showed evidence of LLM usage. A reviewing editor left a comment on the submission for a German financial technology company that referred to the draft “AI garbage.”
Interestingly, being flagged for LLM did not doom a draft. Thirteen of the approved submissions in our dataset had previously been flagged for LLM usage.
Notability is a highly subjective assessment
The AfC Workflow flowchart lays out a schematic, well-defined, and seemingly replicable process for assessing whether submissions meet Wikipedia content guidelines.
But those guidelines—particularly regarding notability—are interpreted and applied differently by each reviewing editor, resulting in judgments that can feel arbitrary and inconsistent when viewed in the aggregate, especially for submissions on the borderline of notability. In practice, Wikipedia guidelines can be cited to explain AfC approvals or rejections retroactively, but they don’t always regulate or predict those decisions.
For instance, the report details how (and why) a submission for a Georgian businessman and philanthropist was approved despite only citing three sources—and no media coverage. Conversely, the submission for Katseye superstar Manon Bannerman was repeatedly denied, despite citations to the numerous (and often glamorous) profile articles about her.
Some degree of inconsistency is unavoidable given the sheer volume of AfC submissions and the operational necessity of having different editors with varying backgrounds, biases, and levels of experience reviewing drafts. It would be uncharitable and misleading to cite a handful of outlier examples of what we perceive as low-quality drafts being approved or seemingly notable drafts being dismissed and then suggest that this is evidence of administrative failure.
That said, the problems we identify with the AfC process do reflect long-simmering concerns about the encyclopedia’s size and scope. The English-language Wikipedia is staggeringly large (7 million-plus articles) and thus extremely difficult to administer, especially in a consistent fashion. The influx of LLM-generated content and submissions only exacerbates this challenge.
In this study, we observed a small number of editors trying to review large numbers of submissions in a relatively short amount of time, all in an effort to enforce content guidelines and ensure Wikipedia remains a valuable source of information.
Read the full report (pretty please)
Want to dig deeper into our findings and methodology? Download our full AfC audit report, available here.
Questions? We’re happy to provide answers! Just shoot us an email at shout@luminodigital.com