Study Finds Over One-Third of Newly Published Web Pages Contain AI-Generated Text

Recent research indicates that the share of AI-written content on the internet has grown sharply following the arrival of Large Language Models (LLMs) in late 2022. According to an analysis by the Pew Research Centre of nearly half-a-million English-language webpages, about 10 per cent of a broad sample collected in July 2026 showed signs of AI authorship.

When researchers filtered the data to look strictly at content published after the public release of ChatGPT, the proportion increased significantly. The analysis found that over one-third of recently published pages were likely written or substantially edited by AI. The report notes, however, that AI detection models are not entirely infallible and can occasionally misclassify human-written documents.

A separate study titled The Impact of AI-Generated Text on the Internet, conducted by researchers from Imperial College London, the Internet Archive, and Stanford University, reported similar findings. By mid-2025, roughly 35 per cent of newly published websites were classified as AI-generated or AI-assisted, compared to near zero prior to late 2022.

PJ Narayanan, former Director of the International Institute of Information Technology-Hyderabad (IIIT-H), noted that the rapid generation pace of powerful AI systems allows individuals and companies to produce data at an exceptionally high rate. He cautioned that a direct consequence of this trend will be on future AI systems, as their training data increasingly incorporates synthetic material, which could potentially degrade quality over time.

Kashyap Kompella, an AI analyst, highlighted that the primary challenge is no longer access to text, but rather access to original, grounded, high-quality human knowledge. He explained that future AI systems will increasingly rely on data provenance to determine whether information originated from real human activity, expert work, proprietary workflows, or another AI system.

According to Kompella, this shift raises the market value of private and proprietary datasets. Real workplace communications, decisions, and workflows are becoming valuable training assets because they cannot be easily replicated from the public web. Consequently, the industry may transition from a data-labelling economy to a data-creation economy, where experts are compensated to generate new problems, examples, workflows, and evaluations specifically designed for AI systems.

"The rapid proliferation of AI-generated content on the web introduces a unique set of challenges for digital platforms and business enterprises alike. As synthetic text becomes more prevalent, the true value for businesses and AI developers will shift decisively toward verified, proprietary data and original human expertise. Founders and organizations must recognize that data provenance and high-quality, authentic insights will become critical competitive advantages in an increasingly automated digital ecosystem." — Dr. Shishir Gupta, Founder & CEO, StartupLanes