By Matt Busel, who built the apps, crates and game he writes about · every number links to its source and every dataset is downloadable (methods) · corrections · featured in Kirino, gamebiz, This Week in Rust, New Mobile Life, By-Press
Wikipedia banned AI text. Then AI tags tripled: 1,618 in July
On this page · 9 sections
Short answer: English Wikipedia banned LLM-written article text on March 20, 2026. Its own cleanup tags say the ban did not slow the flood: editors flagged 532 articles as suspected AI text in March and 1,618 in July, three times as many. October already has 1,232 with four days left. Nobody had added these numbers up, so we did, from Wikipedia's monthly tag categories and 1,000+ Wayback Machine snapshots of them.
TL;DR: We count at least 13,484 English Wikipedia articles tagged as suspected AI-generated text between May 2025 and October 27, 2026, and 8,100 of those tags are still open today. In the six months after the ban (April to September 2026) editors tagged 5,513 articles, 31% more than the 4,216 in the six months before it. The ban gave volunteers a rule; it did not give them hands, and the cleanup rate for the summer peak sits at 47%.
Key numbers
| What | Number | Source, checked Oct 27 2026 |
|---|---|---|
| Articles tagged as suspected AI text, May 2025 to Oct 27 2026 (floor) | 13,484 | Wayback snapshots + Wikipedia API |
| Tags still open today | 8,100 | Wikipedia API |
| Tagged in March 2026 (ban month) | 532 | Wayback, May 6 2026 |
| Tagged in July 2026 (record) | 1,618 | Wayback, Aug 1 2026 |
| Tagged in October 2026 so far | 1,232 | Wikipedia, live |
| Apr-Sep 2026 vs Oct 2025-Mar 2026 | 5,513 vs 4,216 (+31%) | Our sum, data.csv |
| RfC vote that passed the ban | 44 to 2, closed Mar 20 2026 | MediaNama |
The ban passed 44 to 2. The tags went up anyway
The Request for Comment closed on March 20, 2026, with 44 votes for and 2 against. The new guideline bans using LLMs to generate or rewrite article content, with two narrow exceptions: copyedit suggestions on your own writing and a first-pass translation.
The coverage stopped there. Rule passed, story over. Nobody went back to check what the rule did.
Wikipedia keeps the answer in plain sight. Every time an editor adds the "AI-generated" cleanup template to an article, the article drops into a dated category: "Articles containing suspected AI-generated texts from July 2026", and so on. Count the members and you have the monthly number of new flags.
How we counted (and why this is a floor)
A category only shows articles that still carry the tag. Clean an article and it leaves, so today's count for an old month shrinks over time. Comparing today's numbers across months would make old months look quiet and recent ones look like a crisis.
So we went to the Internet Archive. Its CDX index (the list of every saved copy of a URL) holds 1,095 dated snapshots of these 18 monthly category pages. We read the "out of N total" line from the copy closest to each month's end plus samples across the year, and kept the highest count we saw. For October 2026, where the month is still running, we used the live API count from this morning.
That makes every number a floor. Tags removed before a snapshot never show up. For January, February and March 2026 the archive has no copy right at month end, so their peaks come from later copies and run low. That cuts against our headline, not for it.
Month by month: the summer flood
Before the ban, the busiest month on record was September 2025, at 1,340 tags. Through the spring of 2026 the count sat near 530 a month, and May dipped to 406.
Then June hit 1,099, July set the record at 1,618, and September came in at 1,173. October has 1,232 tags with four days left: already the second-biggest month ever, and on its recent pace (about 36 new tags a day since October 15) it lands near 1,370 by Halloween. That projection is our estimate.
Our read: the ban changed what editors were allowed to do about AI text, so they went looking. Some of that jump is enforcement, not just more slop. Either way the job got bigger, not smaller, and the people doing it are volunteers.
The backlog: 8,100 tags nobody has cleared
Here is the number that should worry anyone who reads Wikipedia, and every AI lab that trains on it: 8,100 articles carry an open suspected-AI tag right now.
The old months mostly got cleaned. May and June 2025 are at 98% and 99% cleared. But that was when a month meant 80 to 108 tags. The bigger the month, the slower it drains: July 2026 is 47% cleared after nearly three months, and November 2025 is only 28% cleared after eleven.
The ban debate said it out loud: generating text takes seconds, verifying it takes hours. The tag data puts a number on that gap. Tagging is fast. Fixing is slow.
Check any page yourself in 30 seconds
You do not need our pull to see this. Open any article and scroll to the bottom: if it sits in a category named "Articles containing suspected AI-generated texts from" a month, an editor flagged it and nobody has cleared it yet.
For the whole picture, the Wikipedia API call in our sources returns every monthly category and its live size in one request. Pair it with the Internet Archive CDX link and you can rebuild our numbers line by line.
A machine can do it too. An agent that cites Wikipedia can check the category list of a page before quoting it, and skip or warn on flagged text. That is a few lines of code, and it is the cheapest quality filter on the internet right now.
Who wins, who loses
The losers are the readers who land on one of 8,100 flagged pages from a search box, and the AI companies that scrape Wikipedia as their cleanest source and now pull in their own echo. A 2024 Princeton study estimated detectors flag over 5% of new English Wikipedia articles as AI-made; the encyclopedia's own count now shows the volume rising after the rule meant to stop it.
The winners are whoever builds the tooling that turns a tag into a fix. Wikipedia did the cheap part, a rule and a template. The expensive part, checking claims against sources at scale, has no owner. That is a gap a builder can fill, and the dataset below is where to start.
More on the AI economy in our AI industry coverage, on the news desk and in the LLM hub.
What this means for you
Download our data.csv: every monthly tag category, its highest archived count, the snapshot date, how many are still open and the share cleared. Use it to track the backlog, to filter flagged pages out of a training set, or to point a cleanup bot at the oldest open months first. We will rerun it as the months close.
If you publish anything people read, the same test applies to you: copy that sounds machine-made loses trust fast. Our $49 Launch Check runs 26 checks on your page and tells you what to fix; if a problem we checked turns up later, we keep fixing it at no extra cost until it does what it promises.
Sources
- English Wikipedia API, category sizes, Oct 27 2026
- Internet Archive CDX index, monthly AI-text categories, Oct 27 2026
- MediaNama, English Wikipedia bans AI-generated text, Mar 26 2026
- Princeton researchers (arXiv 2410.08044), The Rise of AI-Generated Content in Wikipedia, Oct 2024
- Tensorust data.csv (18 monthly categories), Oct 27 2026
First published here on tensorust.com on Oct 27, 2026. Related: .
Up next
More on AI industryTim Sweeney told Kenney "up yours" on 2.5M views. Kenney is rightTim Sweeney told Kenney 'up yours' over Epic's AI-first pivot. The receipts: 1,000+ layoffs, a vibe coded launcher, and an AI engine that broke ROLLBACK.Read it next →Cite this
Free to cite and reuse under CC BY 4.0 with a link back. How it was collected: methods. Spotted an error? Corrections.
Did this answer your question?
Get the next one by email.
What my AI agents tried this week, with the real numbers: what worked, what flopped, what made money. One email when there is something worth reading.
No spam, never sold. Every email has a one-click unsubscribe.
What are people saying about this? Read and post on The Floor →
More: the author, related reading, share this, a free game
Try it yourself · Following AI?
Check what the models really cost
71 API prices, each copied from the provider page with the date, plus every change.
In The Manual: on this topic
Explainers answer one question, short answer first. Guides walk you through it step by step.
- Explainer · AI, LLMs and agentsWhat is the KV cache in an LLM, in plain English?
- Guide · Money and the economyHow to search USAspending.gov for any company's federal contracts (and catch the resellers)
- Explainer · Money and the economyWhat is an IPO lock-up expiration, and why do stocks drop around it?
- Explainer · Money and the economyWhat is enhanced geothermal, and how is it different from regular geothermal?
Read next
- AI · Oct 28, 2026 · 6 min readAI layoffs: firms blaming AI spent $28.8B on buybacks, 12x the cuts
- AI · Oct 28, 2026 · 6 min readAI layoffs: firms that told the SEC AI cuts jobs lobby on AI for $37M
- Economics · Oct 28, 2026 · 6 min readAIAI's founder promises $125M from one drone deal. Stock: $120M
