A Third of Web Pages Published Since ChatGPT's Launch Show Signs of AI Authorship, Pew Study Finds

More than one-third of English-language web pages published since ChatGPT's public debut in November 2022 show signs of having been written or substantially edited by artificial intelligence, according to a new analysis from the Pew Research Center, offering one of the most concrete measures yet of how quickly generative AI has reshaped online content.

To conduct the study, Pew researchers used the Common Crawl web archive to collect nearly half a million English-language web pages spanning roughly the past five years, beginning a couple of years before ChatGPT's release. Researchers then ran the text of those pages through an AI detection tool called Open Pangram to assess how many were likely written or heavily edited by artificial intelligence, according to Pew's own published findings.

In a random sample of 10,000 web pages collected in July 2026, approximately 10% showed what Pew described as "significant signs of AI authorship." However, that headline figure includes a substantial share of older pages published well before AI writing tools existed and therefore could not possibly have been AI-generated. When researchers filtered out those older pages and focused exclusively on content published after ChatGPT's release, the picture shifted dramatically: signs of AI authorship appeared in more than one-third, or 35%, of that more recent content, according to Pew's report.

"In the July 2026 snapshot, signs of AI authorship can be found in over one-third of pages published after ChatGPT was released," Pew wrote in its analysis, adding that the finding aligns with other research indicating that a large share of recently published internet content has likely been written or substantially edited using AI tools.

The distribution of AI-authored content across the web is far from uniform. According to Pew, commercial .com domains showed signs of AI authorship at roughly 10 times the rate found on .edu or .gov domains, both of which registered around 1%. Nonprofit .org domains fell in between, with an AI-authorship rate of about 4.6%, according to Pew's data. That gap has widened considerably since ChatGPT's initial release; Pew noted that when the chatbot first launched, the kinds of linguistic patterns that can signal AI authorship appeared at broadly similar rates across the major top-level domains, meaning the current divergence reflects a meaningful shift in how different types of websites have adopted AI writing tools over the past several years.

TechCrunch, which reviewed Pew's findings, cautioned that AI-detection tools such as Pangram are not infallible and can occasionally misclassify genuinely human-written pages as AI-generated. Even so, the outlet noted that "at scale, the data is likely at least directionally correct," suggesting the overall trend captured by the study reflects a genuine and substantial shift in online authorship patterns rather than a statistical artifact.

Beyond the headline percentage, Pew's research identified specific stylistic markers that have become increasingly common across web content and are frequently associated with AI-generated writing. According to the study, the use of em dashes, Oxford commas and certain characteristic phrasing patterns, including constructions such as "it's not X, it's Y," have all become noticeably more frequent in recently published web content, further supporting the broader finding that AI tools are playing an increasingly significant role in how online text gets produced.

Digital Trends, in its coverage of the study, emphasized an important nuance in how the findings should be interpreted. The outlet noted that Pew's research does not claim one-third of web pages were written entirely by AI with no human involvement, but rather that the pages showed indicators suggesting AI played some role in the content's production. Modern online writing increasingly blends human and machine contributions in varying proportions, the outlet explained, with writers frequently using AI tools to generate initial drafts, rewrite specific sections, correct grammar, expand shorter pieces into longer articles, or produce search-optimized versions of existing content, meaning a single published page can reflect a complex mixture of human and AI authorship rather than falling cleanly into one category or the other.

The findings have already drawn attention from the digital advertising industry, given the significant overlap between AI-generated content and the broader ecosystem of low-quality, high-volume web content sometimes referred to as "Made-for-Advertising," or MFA, sites. According to B&T's coverage of the study, industry observers are urging advertisers and publishers to shift their focus away from simply determining whether a given piece of content was created by AI, and toward evaluating whether that content delivers genuine value to readers regardless of how it was produced. The publication noted that generative AI has dramatically reduced both the cost and time required to produce large volumes of low-value content at scale, a dynamic that has significant implications for how digital advertising inventory gets evaluated and purchased going forward.

Separate research cited alongside Pew's findings has reached broadly similar conclusions using different methodologies. According to Shelly Palmer's coverage of related Stanford research, researchers there found that 35% of new websites created since 2022 were AI-generated or AI-assisted, drawing on Internet Archive data to track how the composition of the web has shifted in the years following ChatGPT's launch, a finding that closely mirrors Pew's own 35% figure for post-ChatGPT content despite the two studies relying on different underlying data sources and detection methods.

Digital Trends pointed to a notable irony embedded within the broader trend the research documents. AI companies require enormous volumes of human-created text to train their language models, the outlet observed, even as those same AI systems are now responsible for producing an increasing share of the content that eventually gets published online, raising longer-term questions about a web increasingly characterized by humans writing content intended for machines to read, machines producing content intended for human readers, and, potentially, future AI models eventually training on content that earlier AI systems helped generate.

As generative AI tools continue to become more sophisticated and more deeply embedded in everyday content production workflows, researchers, publishers and advertisers alike are likely to continue closely tracking how the balance between human and AI-assisted authorship on the web evolves in the years ahead, particularly given how dramatically the share of AI-influenced content has grown in just the roughly three and a half years since ChatGPT's initial public release in November 2022.