{"id":77702,"date":"2026-08-24T16:15:42","date_gmt":"2026-08-24T20:15:42","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=77702"},"modified":"2026-08-24T16:15:42","modified_gmt":"2026-08-24T20:15:42","slug":"ai-authorship-pew-study-77702","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/ai-authorship-pew-study-77702\/","title":{"rendered":"1 In 10 Webpages Shows Signs of AI Authorship"},"content":{"rendered":"

The internet is undergoing a quiet but radical transformation. Nearly a decade after generative AI entered public consciousness, the technology has woven itself into the very fabric of the web so deeply that, according to new research from the Pew Research Center<\/a>, approximately one in every ten webpages now carries signs of AI authorship. That figure jumps to more than one in three for pages published after the launch of ChatGPT. These findings, drawn from a scan of nearly half a million webpages, provide the most comprehensive look yet at how extensively AI is reshaping online content.<\/p>\n

Pew’s Scan of Nearly Half a Million Pages Reveals a 10% AI Footprint<\/h2>\n

The Pew Research Center’s Data Labs team published its analysis on August 20, 2026, after running nearly 500,000 webpages through an AI detection system. The study found that roughly 10% of all pages in the sample showed signs of AI authorship or AI-assisted editing. Among pages published specifically after ChatGPT’s launch, that share climbed sharply to 35%. The research arrives less than a year after two other major attempts to estimate the volume of AI-generated<\/a> text on the web, adding another powerful data point to a fast-evolving picture.<\/p>\n

The Pew study does not claim that 10% of the entire internet is written entirely by AI. Its detection threshold captures both text generated from scratch by a language model and text written by a human that was subsequently edited or polished with an AI tool. That distinction matters, because it means the 10% figure encompasses a spectrum of AI involvement rather than marking only fully automated content. But even with that caveat, the scale is striking. A decade ago, AI-authored text on the open web was essentially nonexistent. Today, it is pervasive enough to alter the character of search results, the economics of content production, and the trust readers place in what they find online.<\/p>\n

Commercial Domains Drive the AI Writing Wave at Nearly Ten Times the Rate of Educational and Government Sites<\/h2>\n

The concentration of AI-influenced content is not uniform across the web. Pew’s data reveals a stark divide between commercial domains and institutional ones. On .com domains, the six-month average detection rate stood at 9.35%. On .org domains, it was 4.59%. On .edu domains, the rate dropped to 1.03%, and on .gov domains, it fell further to 0.76%. In other words, pages on .com domains show signs of AI authorship at roughly ten times the rate of pages on .edu and .gov domains.<\/p>\n

This divergence did not always exist. In the sample of pages collected before ChatGPT launched, all four domain types sat at or below 1% detection. The separation began after the technology became widely available. .com domains climbed fastest and have continued to rise in every reading Pew has taken since. .edu and .gov domains, by contrast, have stayed close to the 1% mark, barely budging. The pattern strongly suggests that commercial incentives are the primary engine driving AI adoption in content creation. When the goal is to publish at scale, to rank in search results, or to capture audience attention, the pressure to use AI tools is far greater than in academic or government settings, where editorial standards, review processes, and institutional norms act as natural brakes.<\/p>\n

Em Dashes, Oxford Commas, and the Linguistic Fingerprints of AI Text<\/h2>\n

Pew’s analysis did not rely on a single detection method. Instead, the research team examined specific linguistic markers that have become statistically associated with AI-generated text, tracking how their frequency has changed over time. The results show a clear shift in the stylistic texture of the web.<\/p>\n

Em dashes, a punctuation mark that language models tend to use more liberally than many human writers, increased from 5.79 uses per 10,000 words in early 2023 to 11.19 uses per 10,000 words in early 2026. That is nearly a doubling in frequency. Oxford commas, which are common in AI training data<\/a> and often appear in machine-generated text, saw a 63% rise over the same period. Certain vocabulary words that have become signature markers of AI writing, including “delve,” “interplay,” and “testament,” more than doubled in usage across the sample. The negative parallelism structure, sometimes described as the “it’s not just X, it’s Y” construction, increased from 0.87 to 2.36 uses per 10,000 pages, though it remains relatively rare overall.<\/p>\n

Pew itself is careful to note that none of these traits alone can identify a single document as AI-generated. Human writers use em dashes. Human writers use Oxford commas. Human writers have employed the word “delve” long before ChatGPT existed. The claim the research makes is about rates across large populations of text, not about individual pages. When these markers all move in the same direction at the same time, across hundreds of thousands of pages, the pattern becomes difficult to explain away as coincidence or natural stylistic drift. Something structural is happening to the way the web is written.<\/p>\n

This question of detection is not merely academic. In a separate analysis covered earlier this year, Ahrefs ran its own detector analysis and ended with an open question about how useful a detector score really is as AI editing becomes more common in everyday writing tools. Pew’s data adds new dimensions to that problem. The markers it tracks are appearing more frequently across the web, which means they are becoming more common in human-written text as well, if only because human writers are increasingly exposed to AI-generated prose and may unconsciously absorb its patterns. Whether that trend makes any single marker more or less helpful for identifying AI text is an open question that Pew did not explore, but it is one that publishers, search engines, and content platforms will need to grapple with.<\/p>\n

What Percentage of Webpages Show Signs of AI Authorship? The Estimates Diverge<\/h2>\n

Pew’s 10% figure is not the only estimate on the table, and it is not the highest. Different research teams using different methodologies have arrived at substantially different numbers, and the lack of consensus highlights how difficult it is to measure something that is itself changing rapidly.<\/p>\n

Graphite, an SEO firm, estimated that the share of newly published English-language articles that are primarily AI-generated reached 49.9% in the first quarter of 2026. That is nearly half of all new articles, and it suggests that AI is not just touching content but in many cases becoming the primary author. Graphite’s methodology differed from Pew’s in important ways. It focused on newly published articles rather than a broader cross-section of existing pages, and it used a combination of detection tools, including Copyleaks and GPTZero, alongside Pangram.<\/p>\n

A separate preprint from researchers at Imperial College London, the Internet Archive, and Stanford University found that by mid-2025, 35% of newly published websites were detected as AI-generated or AI-assisted. That study, which has not yet been peer-reviewed, aligns more closely with Pew’s finding of 35% for pages published after ChatGPT’s launch.<\/p>\n

All three estimates lean on Pangram in some form, which introduces a common dependency. Pangram is a widely used AI detection framework, but no detector is perfect, and different detectors can produce conflicting results on the same text. The variation among the three estimates, from 10% to 35% to 49.9%, reflects real differences in scope, sampling, and methodology, but it also reflects the fundamental difficulty of drawing a clean line between human and machine writing in an era when the two are increasingly blended.<\/p>\n

Why the Concentration of AI Text on Commercial Domains Matters for Search and Publishing<\/h2>\n

The domain-level data from Pew carries practical implications that extend far beyond academic interest. Signs of AI text are most prevalent on .com domains, which is precisely the part of the web that search engines index most heavily and that SEO work touches most directly. If roughly one in ten pages on .com shows AI<\/a> markers, and if that share is still climbing, then the commercial web is already deeply interwoven with machine-written content.<\/p>\n

This concentration creates a cascade of effects. Search engines must decide how to treat AI-authored content in their rankings. Google has stated that it rewards quality content regardless of how it is produced, but quality is a subjective measure, and the sheer volume of AI-assisted content creates new challenges for ranking systems designed to surface the most useful results. For publishers, the economics of content production are shifting. If competitors can produce articles with AI tools at a fraction of the cost, the pressure to adopt similar tools becomes intense, even for organizations that would prefer to maintain a purely human editorial process.<\/p>\n

Pew’s threshold catches AI editing as well as full generation. A page that a person wrote and then cleaned up with an AI tool lands in the same category as one that an AI produced from start to finish. That means the 9.35% figure for .com domains likely includes a large amount of content that was substantially human-authored but AI-polished. Whether that distinction matters depends on the use case. For a reader evaluating credibility, a lightly edited page may be perfectly acceptable. For a search engine trying to determine originality and authority, the distinction becomes more significant.<\/p>\n

AI Editing Is No Longer Optional, It Is Embedded in the Tools Writers Already Use<\/h2>\n

One reason the detection problem is becoming more complex is that AI editing has moved from standalone tools into the core features of mainstream writing software. Google Docs and Microsoft Word both now include native AI editing capabilities. A writer can draft a paragraph and ask the software to rephrase it, shorten it, or adjust its tone, all without leaving the document and without any external tool being involved. Pew’s detection threshold already considers this type of AI-assisted writing as a signal, but none of the current studies meaningfully distinguish between text that was entirely generated by a language model and text that a human wrote and then refined with AI suggestions.<\/p>\n

This blurring has consequences for how the industry thinks about AI authorship. If a journalist writes a story from scratch and then uses an AI tool to fix a few awkward sentences, is the resulting page AI-authored? Most readers would say no. But a detector that flags em dashes and Oxford commas might classify it as AI-influenced nonetheless. As AI editing becomes a standard part of the writing process, the baseline for what counts as human-authored text will shift, and detection tools will need to evolve accordingly.<\/p>\n

The question that remains open is whether any of this matters for the end user. A page that is accurate, useful, and worth publishing is still a page worth publishing, regardless of whether it was written by a person, a machine, or a collaboration between the two. The real test is not how the text was produced but whether it serves its purpose. That, ultimately, is a judgment that no detector can make. It is a judgment that still belongs to readers, editors, and the human beings who decide what to publish and what to trust. As the web becomes increasingly written by machines, that human judgment becomes more important, not less.<\/p>\n","protected":false},"excerpt":{"rendered":"

The internet is undergoing a quiet but radical transformation. Nearly a decade after generative AI entered public consciousness, the technology has woven itself into the very fabric of the web so deeply that, according to new research from the Pew Research Center, approximately one in every ten webpages now carries signs of AI authorship. That […]<\/p>\n","protected":false},"author":7,"featured_media":82731,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/77702.png","fifu_image_alt":"1 In 10 Webpages Shows Signs of AI Authorship","footnotes":""},"categories":[31],"tags":[],"class_list":["post-77702","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/77702.png","fifu_image_alt":"1 In 10 Webpages Shows Signs of AI Authorship","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/77702","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=77702"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/77702\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/82731"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=77702"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=77702"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=77702"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}