Study: Nearly half of online articles are now AI-generated
▼ Summary
– A study of 55,000 webpages found that nearly half of sampled English-language online articles were classified as mostly AI-generated by three commercial detectors.
– AI-classified content surged after ChatGPT’s launch in late 2022 and stabilized at roughly 50% of sampled articles through early 2026.
– The study did not measure whether AI-generated articles perform better or worse in search rankings, traffic, or visibility.
– Graphite tested detectors against pre-2022 human articles and AI-generated articles, finding average false-positive and false-negative rates below 2%.
– The researchers noted that mixed workflows, where humans heavily edit AI drafts, were not evaluated and may be harder for detectors to classify accurately.
Nearly half of all English-language articles published online are now classified as primarily AI-generated, according to a recent analysis of 55,000 webpages drawn from Common Crawl. The study, conducted by Graphite, reveals that the share of detector-classified AI-written content surged following the launch of ChatGPT in late 2022 and has since stabilized at roughly 50% of sampled articles through early 2026.
By the first quarter of 2026, an estimated 49.9% of sampled articles were flagged as mostly AI-generated. In late 2025, AI-generated articles briefly outpaced human-written ones in the dataset. Within the first year of ChatGPT’s November 2022 release, AI-generated pieces accounted for 35.9% of published content. That figure climbed to about 48% by late 2024 and has hovered near the 50% mark since early 2025.
The plateau suggests that publishers may be learning that heavily AI-generated articles often underperform in search results, according to Graphite. However, the study did not directly measure search rankings, traffic, or visibility, leaving open questions about how these articles actually fare with audiences.
Why this matters for SEO and publishing. The findings confirm that AI-generated or AI-assisted content is now a standard part of the web, especially in large-scale content operations. Still, the study does not indicate whether AI-written articles receive more or less traffic than human-written ones, nor whether users regularly encounter them in Google Search or AI-powered systems.
How the study was conducted. Graphite analyzed 55,400 English-language articles from Common Crawl, published between January 2020 and March 2026. Three AI detection systems were used: Pangram, Copyleaks, and GPTZero. The results from all three were averaged. Each detector defined “primarily AI-generated” slightly differently. Pangram and Copyleaks estimated the percentage of AI-generated text in each article, while GPTZero assigned labels like “Human,” “Mixed,” and “AI.” Articles labeled “Mixed” by GPTZero were counted as primarily AI-generated.
All three detectors showed the same broad trend: a sharp rise in AI-classified content after ChatGPT’s debut, followed by a plateau near 50%.
Testing the detectors’ accuracy. To estimate false positives, Graphite tested the detectors against 15,700 articles published before ChatGPT launched in November 2022, assuming those were mostly human-written. To estimate false negatives, the team generated 6,000 AI-written articles using GPT-5, Gemini 3.1 Pro, and Claude Opus 4.6. Average false-positive and false-negative rates were below 2% across all three detectors in these benchmark tests.
However, the study did not evaluate mixed workflows where humans heavily edit AI drafts , a common practice that may be harder for detectors to classify accurately. The researchers also cautioned that future AI models may become harder to detect as generation quality improves.
The study concludes that AI now produces roughly as many online articles as humans do, marking a significant shift in the digital content landscape.
(Source: Search Engine Land)
