A third of web pages published since ChatGPT launched were written by AI, study finds

Pew Research analyzed nearly half a million English‑language pages from the Common Crawl archive, spanning roughly five years that include the period before and after ChatGPT’s November 2022 launch. Using Open Pangram’s detection technology, the researchers first examined a random sample of 10,000 pages collected in July 2026 and found that about one in ten displayed strong indicators of AI authorship. Because that sample inevitably contained pages created before generative models were available, the team removed all items dated prior to the ChatGPT release. Re‑evaluating the post‑ChatGPT subset revealed that roughly 35 % of the pages exhibited signs of being written or substantially edited by AI. The study notes that this figure aligns with other investigations reporting a growing share of AI‑generated or AI‑edited content on the newer web. It also follows Cloudflare’s observation that automated traffic now exceeds human‑driven visits, a threshold reached earlier than the company’s internal forecasts.

Why this matters

The finding that over a third of post‑ChatGPT pages carry AI authorship signals suggests that language models are already influencing the volume and provenance of online text at scale, which may affect downstream tasks such as information retrieval and model training data quality. While the study measures likelihood of AI involvement rather than confirming it for each page, the consistent detection across a large, temporally filtered corpus supports the inference that generative tools are contributing substantively to web content creation, a trend that could necessitate updated provenance‑tracking mechanisms and licensing considerations for training data sourced from the web.