About 24% of crawlable web pages contained AI-generated or AI-edited text in August 2026, according to a study posted to arXiv on 27 October 2026 by Adam Visokay, Tyler H. McCormick and 11 co-authors from the 2026 Summer Institute in Computational Social Science at the University of Washington. The paper says the share “rose from near zero before 2022 to nearly 25% by August 2026.”
In plain words: on the authors’ estimate, about one in four pages of longer text that a web crawler can reach now has text written or edited by AI, and the share is rising faster than before. The figures come from 5,118,347 pages on 505,974 domains across 71 Common Crawl snapshots from January 2019 to August 2026, scored by several AI detectors and corrected for their errors with a statistical model.
Short explainer · AI web text brief
A vertical news Short narrated from the published headlines and opening lines of this piece and four related FourWeekMBA pieces on AI, data and the web.
Business Pill · Watermarking
An evergreen Business Engineer explainer on watermarking AI output. It explains the general idea only and does not discuss this study.
The key insight: As we read it, the study puts a number on how much of the open web’s longer text is now AI-written, about one page in four, and shows the share is growing fastest on small, poorly linked sites rather than on the most central ones.
Estimated share of crawlable pages with AI-generated or AI-edited text in August 2026, by top-level domain: .com 29.5%, .net 20.4%, .org 14.6%, .edu 6.1%, .gov 5.8%; about 24% across all pages. Detector-corrected model estimates. Source: Visokay et al., arXiv 2610.36631, 27 Oct 2026.
Share now. About 24% of crawlable pages in August 2026, with a 95% credible interval of 22.9% to 24.4%.
Speed. The average annual change was 0.2 percentage points from January 2019 to February 2023, 4.6 points from February 2023 to January 2025 and 8.1 points from January 2025 to August 2026, the paper says.
By domain type. In August 2026 the estimate was 29.5% of .com pages, 20.4% of .net, 14.6% of .org, 6.1% of .edu and 5.8% of .gov pages.
By language. 32.8% of English-only pages against 17.0% of pages with no English detected; the gap was 0.6 points in February 2023 and 15.8 points in August 2026.
Average annual change in the estimated share of crawlable web pages with AI-generated or AI-edited text, in percentage points: 0.2 (Jan 2019-Feb 2023), 4.6 (Feb 2023-Jan 2025), 8.1 (Jan 2025-Aug 2026). The share in August 2026 was about 24% (95% credible interval 22.9-24.4). Bayesian, detector-corrected estimates from 5.1 million pages in 71 Common Crawl snapshots. Source: Visokay et al., arXiv 2610.36631, 27 Oct 2026.
What Was Posted on X
Highly recommended read. AI-written text is everywhere on the web now. This study measured how much.
Lots of interesting insights here.
They found that across more than 5 million crawlable pages from 2019 to 2026, about 24% contained AI-generated or AI-edited text in August 2026, up from near zero before 2022.
The growth rate nearly doubled, from 4.6 points per year between early 2023 and early 2025 to 8.1 points per year since.
About four-fifths of the growth comes from pages on the edge of the web graph, linked from a few well-connected domains. The most central pages are the least likely to contain AI text. English pages have more than other languages, and .gov and .edu pages have far less than .com.
The team uses a Bayesian model to correct for false positives and false negatives in the detector.
Elvis Saravia (@omarsar0) of DAIR.AI shared the study on X on 28 October, writing that across more than 5 million crawlable pages from 2019 to 2026, about 24% contained AI-generated or AI-edited text in August 2026, up from near zero before 2022.
Where the Growth Comes From
The authors sort domains into six tiers by PageRank, a measure of how well connected a site is in the web’s link graph. “About four-fifths of the increase in the prevalence of pages with fully AI-generated or AI-edited text comes from those belonging to domains on the periphery of the web graph that are not highly connected to other well-connected domains,” the paper says.
The two lowest tiers hold 99% of domains and 70.5% of eligible pages; the paper’s examples include “a Swedish rabbit breeder, a Louisville dog-walking service, a German hair salon” and a pizza restaurant. In August 2026 the estimate was 27.5% in the least-connected tier and 11.6% in the most central one, which the paper says is mostly “the platforms and infrastructure of the web itself” plus a few institutions and publishers such as the Financial Times.
How It Was Measured
Single detectors make mistakes that vary by context, the authors write: in their tests the open Pangram model had fewer false positives than Binoculars-Falcon on .gov and .edu domains but more on .com domains. The study combines several detectors in a Bayesian model that corrects for false positives and false negatives. Without that correction, the paper says, “the uncorrected detector puts the prevalence substantially higher.”
The estimates cover pages with at least 500 characters of extracted text, about 80 English words. Pew Research Center has reported a figure of 10 percent; on all English-language captures with no text extraction, the same detector would flag “roughly 13 to 19 percent,” the authors write.
Why Training Data Is in the Frame
Common Crawl is the open archive the authors sample, and filtered text from it is used to train widely used language models, the paper notes, so “its composition could have substantial implications for model collapse, depending on how it is weighted in the training process.” The authors say AI text is growing in what people see on the web “and in text ingested for training by contemporary and future language models.”
Similarweb data put AI ahead of search across the purchase journey, as we reported in June. What’s new here: a corrected, all-language estimate of how much of the open web’s longer text is now AI-written, and where it sits.
The Structural Read
As we read it, the gap by domain type matters for anyone who uses the web as a source: the paper estimates 29.5% for .com pages against 6.1% for .edu and 5.8% for .gov.
The authors note that filtered Common Crawl text is used to train widely used language models, and say the web’s composition could matter for model collapse depending on how it is weighted in training.
The paper estimates English-only pages at 32.8% against 17.0% for pages with no English detected, a gap that has widened since February 2023.
Visokay et al., arXiv 2610.36631
“the role of AI-generated text is growing and accelerating in the content that people see on the public-facing web”
Three Implications
FOR SEARCH AND AI ANSWERS Systems that rank or cite web pages draw on a web where, by the paper’s estimate, about one crawlable page in four has AI-generated or AI-edited text.
FOR MODEL BUILDERS The paper says Common Crawl’s composition could have substantial implications for model collapse, depending on how it is weighted in training.
FOR PUBLISHERS The paper estimates the lowest shares on the most central domains, a tier that includes a few publishers such as the Financial Times.
The framework describes a Corpus Moat “built by pretraining, eroding as the public web is exhausted and synthetic data closes gaps”.
As we read it, this study measures one side of that erosion: the authors estimate that about 24% of crawlable pages in August 2026 had AI-generated or AI-edited text. The framework was published in July 2026 and does not discuss this study.
What Is Not Established
The figures are statistical estimates from a sample, not a count, and they rest on AI detectors whose error rates the model estimates. The study groups AI-generated and AI-edited text together. It covers only pages a crawler can reach, so content behind logins, inside apps and on sites that block crawlers is outside it, as are short posts, comments and reviews on their own; the authors say their estimates “say nothing about the prevalence of fully AI-generated or AI-edited text on social platforms.” The paper is a preprint on arXiv; we did not contact the authors.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
A University of Washington-linked team estimates that about 24% of crawlable web pages carried AI-generated or AI-edited text in August 2026, up from near zero before 2022, with the yearly rise climbing from 4.6 to 8.1 percentage points. The growth is concentrated on small, poorly linked sites; .gov and .edu pages sit near 6%, .com pages near 30%.
A note on sourcing. Every figure and quotation comes from the paper by Visokay et al. posted to arXiv on 27 October 2026 (arXiv 2610.36631); the estimates are the authors’ detector-corrected model estimates, not a count. We found the paper through a post on X by Elvis Saravia of DAIR.AI on 28 October 2026. We did not contact the authors.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
