A recent study has uncovered that more than one-third of webpages published since ChatGPT’s launch in November 2022 bear traces of AI-driven authorship or heavy editing. The research, carried out by Pew Research using data from Common Crawl and analysis tools like Open Pangram, suggests that AI’s footprint on the web has become both sizeable and measurable. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
How the Study Was Conducted
To assess how widespread AI authorship has become, nearly half a million English-language webpages were collected—spanning approximately five years, including periods before ChatGPT existed. Pew’s team then applied AI-detection methods to determine whether content was likely written or substantially revised by AI. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
In a sample of 10,000 pages from July 2026, roughly 10% displayed strong indicators of AI involvement. But those results include older content published well before AI writing tools were available, which naturally can’t bear those markers. Adjusting the sample to only pages published after ChatGPT’s release raised the share of AI-touched content to about 35%. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
Domain Differences & Detection Challenges
The prevalence of AI-authored content varies sharply depending on the type of domain. Sites using .com extensions showed notably higher rates—around 10-times more—of detected AI authorship than those operating under .edu or .gov extensions, both of which hovered at about 1%. .org domains fell somewhere in between, at roughly 4.6%. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
Pew’s researchers emphasize that AI detection tools like Pangram are imperfect. Misclassifications occur: some content flagged as AI-written could be human-generated, while others may have only slight AI influence. Still, the broader trends—higher prevalence, rising markers of AI style—are likely accurate. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
Additional indicators that suggest increasing AI influence over time include changes in writing style: greater usage of Oxford commas, em dashes, and certain phrasings like “it’s not X, it’s Y”. These stylistic traits are identified as consistent with AI editing or generation patterns. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
The attention this study draws comes amid ongoing concerns about how AI is reshaping online information. Just recently, Cloudflare reported that bot traffic to websites has overtaken human traffic, underscoring how both the creators and consumers of web content are entwined with automation. ([techcrunch.com](https://techcrunch.com/2026/08/20/a-third-of-webpages-published-since-chatgpts-launch-show-signs-of-ai-authorship-study-finds/))
What this signals: The shift toward AI‐authored or heavily edited content suggests a major transformation in how information is created online. For content creators and readers alike, this raises questions about authenticity, trust, and quality. The findings imply that distinguishing human-written work from AI influencers will become increasingly difficult. As detection tools evolve, we’ll need clearer transparency standards and perhaps regulatory oversight. Going forward, watch for how platforms label AI content, how algorithms respond, and whether education helps readers spot AI’s footprint in what they read.
