October 2025, U.S.‑headquartered research firm Graphite released an alarming study: more than half of internet articles are generated by artificial intelligence or heavily rewritten by AI. Researchers call this proliferation of low‑quality output 'AI slop'.
Graphite's analysis covered a broad sample of webpages and various site types. The study shows the problem is concentrated in three sources: content farms optimized for traffic and advertising; aggregator sites that automatically rewrite large amounts of existing text; and bulk‑produced localized or niche‑topic articles created with generative models. Many pages read fluently but lack depth or original insight, exhibit mechanical structural repetition, and exist mainly to occupy space and attract search clicks.
The report notes this is not an isolated technical issue but is altering the 'upstream material' used to train current and future large models. Graphite warns that as training corpora become increasingly polluted with AI‑generated low‑quality content, model outputs may fall into a self‑reproducing loop: models trained on AI text then generate more AI text, degrading novelty and reliability.
Especially noticeable among younger users, search experiences are becoming more superficial. Keyword optimization and automated writing tools push large volumes of repetitive information to the foreground; in‑depth reviews, investigative reporting and original analysis that require professional editorial work are being drowned out by templated AI pieces. The report calls on search engines and platforms to improve visibility labeling for generated content and impose ranking penalties.
Graphite also warns of an 'arms race' between detection tools and generation technologies: as generators improve, detectors based on surface features fail. Content watermarking and source authentication become more critical, yet these measures are not widely adopted across the industry. The report recommends three practical paths: first, promote transparency of content origins, including mandatory AI labeling; second, introduce quality signals into search and recommendation systems that prioritize human‑edited or verified originals; third, establish cleanliness standards for training data and limit inclusion of high‑risk automated sources.
For creators and small site operators who rely on traffic monetization, Graphite's findings are both an opportunity and a warning: AI tools can greatly increase output speed, but excessive supply‑side competition can plunge the ecosystem into a cycle of content glut and quality decline. The study suggests publishers, advertisers and academic institutions must jointly set rules to prevent the web from becoming a closed‑loop factory produced by AI for AI consumption.
The article concludes that the solution is not a blanket ban on generative AI but clear rules and accountability: labeling, provenance, quality assessment and governance of training data will shape the internet information environment in the coming years. For ordinary users, improving information discernment and preferring media with transparent sources will be an important everyday safeguard.