More than a third of web pages published since ChatGPT's launch show signs of being written by artificial intelligence, according to a new study from Pew Research Center — one of the most rigorous measurements yet of how thoroughly generative AI has reshaped the content of the open web.
The finding, released Thursday by the nonpartisan research organization, lands amid growing concern that the internet is filling with machine-written text faster than anyone can track it. The scale of the shift is a defining story of the AI era, and one tracked continuously in AI industry coverage.
How Pew Measured the AI-Written Web
To compile the report, Pew used the Common Crawl web archive to collect nearly half a million English-language web pages spanning roughly the past five years — starting a couple of years before ChatGPT's November 2022 release. Researchers then ran the pages through Pangram, an AI-detection technology, to estimate how many were written or heavily edited by AI.
The headline numbers are striking. In a random sample of 10,000 web pages collected in July 2026, about 10% showed significant signs of AI authorship. But that random sample inevitably included large numbers of older pages published before AI writing tools existed — pages that could not possibly have been machine-written.
When Pew filtered those out and looked only at pages published after ChatGPT's launch, the share showing signs of AI authorship jumped to 35% — more than one in three.
Commercial Domains Lead the Way
The study's domain-level breakdown may be its most revealing finding. Pages on .com domains showed signs of AI authorship at roughly ten times the rate of .edu and .gov domains, both of which sat at around 1% AI-authored. Nonprofit .org domains fell in between, at 4.6%.
The pattern tells a story about incentives. Where content is produced to rank in search engines and generate advertising or affiliate revenue, AI writing has proliferated. Where institutions attach their reputation to what they publish — universities, government agencies — human authorship still dominates.
Pew also found that classic stylistic tells of machine writing have grown more common across the web over the years: em dashes, Oxford commas, and constructions like "it's not X, it's Y" have all increased in frequency.
The .com finding also lines up with what search engine optimization practitioners have reported anecdotally for years: content farms, affiliate sites and low-cost publishers were the earliest and most aggressive adopters of generative writing, because for them the economics of producing another page — any page — had already collapsed to near zero. Institutional publishers with editorial standards moved slower, and it shows in the data.
Bots Writing for Bots
The Pew report arrives shortly after internet infrastructure provider Cloudflare disclosed that bot web traffic had overtaken human web traffic — a milestone reached sooner than the company had estimated. As TechCrunch noted in its coverage of the study, the two findings together sketch an unsettling picture: much of the web is now bots browsing pages that were, in many cases, written by other bots.
Commentators have been quick to connect the data to the "dead internet theory" — the once-fringe idea that much online content is generated artificially for an audience of automated crawlers. What began as a conspiracy-minded meme is increasingly looking like a measurable demographic reality, at least for the commercial web.
Caveats in the Detection
Pew was careful to note the limits of its methodology. Pangram, like other AI-detection tools, can misclassify human-written pages as AI-generated, and detection accuracy varies across writing styles and genres. At the scale of hundreds of thousands of pages, though, the data is likely at least directionally correct, the organization said — and multiple independent studies have reached similar conclusions about the share of newer web content showing signs of AI authorship.
The study also only measured English-language pages and drew from Common Crawl's archive, which captures a snapshot of the web rather than its entirety.
Why the Numbers Matter
The finding has implications far beyond internet trivia. Search engines are already grappling with how to surface trustworthy results when a growing share of indexed content is machine-generated. Publishers and educators face a harder time establishing provenance. And AI model builders face a quieter problem: training data scraped from the web increasingly contains AI output, raising the specter of model degradation through feedback loops of synthetic text.
For everyday readers, the practical takeaway is a new default posture of skepticism. A product review, a how-to guide, or a news-adjacent blog post published on a commercial domain in the post-ChatGPT era now carries roughly a one-in-three chance of bearing an AI's fingerprints — and on some categories of sites, likely much higher.
The web has always been a mix of the authentic and the manufactured. What Pew's data shows is that the balance tipped, and quickly: in under four years, machine writing went from nonexistent to a substantial fraction of everything published online.
Stay Ahead of AI
For the latest AI research, industry data, and trend analysis, bookmark AI Buzz Wire.
Read more AI news →
