Half of all articles online now authored by AI (sherwood.news)

AI-powered SEO firm Graphite reports that about half of English-language online articles and listicles are now mostly or wholly authored by generative models. Using AI-detection software on a 65,000-article sample, Graphite says AI content overtook human-written pieces as early as November 2024 — almost exactly two years after ChatGPT’s debut — and the balance has remained roughly even since. The finding arrives amid broader bursts of synthetic-media activity (e.g., recent text-to-video models) and follows an MIT blind-study result that human evaluators often rate AI-augmented content as higher quality than work from human experts. For the AI/ML community this is a practical tipping point: it intensifies the data-feedback loop (models trained on increasingly synthetic corpora), accelerates the adversarial detection arms race, and complicates evaluation, provenance, and moderation strategies. Technical implications include pressure to improve detection robustness and provenance tagging, rethink training-data curation to avoid model drift, and adapt search and ranking systems to prioritize factuality over fluency. The shift also raises economic and ethical questions for content creators, publishers, and platforms about disclosure, quality metrics, and the long-term reliability of online knowledge.