A growing share of the open web appears to be written, or at least heavily shaped, by AI. A new Pew Research study found that 35% of webpages published after ChatGPT's November 2022 release showed signs of AI authorship.
The finding adds a content-side view to a broader shift already visible in web traffic. Cloudflare recently reported that bot web traffic had overtaken human web traffic, while Pew's work looks at the pages those visitors are encountering.
What Pew Measured
Pew's report focused on English-language webpages gathered from the Common Crawl web archive. The sample covered nearly half a million pages from the past five or so years, including pages from before and after ChatGPT's release.
To evaluate those pages, Pew used Open Pangram's technology to detect material likely written or heavily edited by AI. The study does not claim perfect certainty for every individual page. Instead, it uses detection at scale to understand how AI authorship appears across a large slice of the web.
That distinction matters. AI detection tools can misclassify text, and Pew's analysis acknowledges that limitation. But across hundreds of thousands of pages, the results are presented as directionally useful: the web's newer content increasingly carries markers associated with AI writing.
The 35% Figure Comes From Newer Pages
In a random sample of 10,000 webpages collected in July 2026, around 10% showed "significant signs of AI authorship," according to Pew. That broader number includes older pages, including pages published before modern AI writing tools were available.
Pew then narrowed the view. By filtering out older pages and looking only at webpages published after ChatGPT's November 2022 launch, the rate changed sharply. In that newer group, signs of AI authorship appeared on over one-third of pages, or 35%.
The difference between those two numbers is the core point. A random crawl of the web still contains plenty of older human-written material. But among webpages created in the period after ChatGPT became available, AI-linked writing signals are much more common.
That suggests the visible web is changing not only because people and bots are visiting it differently, but because the material being published is changing as well. Pew's data is about what is being browsed, not who or what is doing the browsing.
Commercial Domains Stand Out
The study also found large differences by domain type. URLs with a .com domain showed signs of AI authorship at around 10 times the rate of .edu or .gov domains.
Both .edu and .gov domains were around 1% AI-authored in Pew's analysis. The rate for .org domains was also lower than .com, at 4.6%.
Those gaps are important because they point to where AI-written or AI-edited webpages appear most frequently in the sample. The source does not explain every reason behind the domain differences, but the distribution itself is clear: .com pages were far more likely to show AI authorship signals than the education, government or organization domains Pew measured.
- .com: around 10 times the rate of .edu or .gov domains
- .edu: around 1% AI-authored
- .gov: around 1% AI-authored
- .org: 4.6% AI authorship rate
For readers, publishers and searchers, the practical takeaway is simple. The type of domain may offer some context about how likely a page is to carry AI writing signals, but it is not a full guarantee of how a specific page was made.
Why Detection Is Complicated
Pew's findings rely on signals, not admissions from every publisher. That is why the report is framed around pages that show signs of AI authorship rather than a definitive list of webpages written by machines.
Open Pangram's detection technology, like other AI detection tools, can incorrectly classify some pages. A human-written page may look machine-written to a detector, and the source does not claim that each flagged page is unquestionably AI-created.
Still, the study's scale gives the data weight. When similar signals appear across a large body of webpages, the pattern can help describe the broader direction of online publishing, even if any single result should be treated with care.
Pew also found that several common supposed tells of AI authorship have increased over time. These included em dashes, Oxford commas and phrasing such as "it’s not X, it’s Y," among other things.
Those markers are not proof on their own. People use punctuation and familiar turns of phrase for many reasons. But in Pew's analysis, their rising frequency sits alongside the detection results as another sign that web writing habits have shifted since AI writing tools became more widely used.
What This Means For The Web
The study describes a web where automated systems are increasingly present on both sides of the page. Cloudflare's report points to bots becoming a larger share of traffic than humans. Pew's report points to many newer webpages showing signs of being written or substantially edited by AI.
Together, those facts create a plain but significant picture: more of the web is being produced with AI, and more of the web is being visited by bots. The source article describes the result as a situation where much of the web is bots reading webpages written by other bots.
That does not mean every AI-assisted page is low quality, false or useless. The source does not make that claim. It does show, however, that AI authorship is no longer a marginal feature of newly published web content in Pew's sample.
For anyone trying to understand online information, the study reinforces the need to look carefully at provenance, domain context and the substance of a page. The presence of polished language alone is not enough to know who wrote something, how it was edited or how much human judgment shaped it.
The web has always mixed personal publishing, commercial pages, institutional information and automated systems. Pew's research suggests that, since ChatGPT's release, AI authorship has become a much more visible part of that mix.