About half of all published articles on the internet are generated by AI systems, a new research from the analytics firm Graphite.io shows.
The findings, based on a study of 65,000 English-language articles published between 2020 and 2025, reveal that the share of AI-generated material rose sharply after the public release of ChatGPT in November 2022.
AI-written content increased from 5% in 2020 to around 48–52% in 2025, while human-written articles declined from 95% to about 52% over the same period.
Graphite’s data indicates that AI content briefly passed human writing between November 2024 and March 2025, reaching a peak of 55% in January 2025.
Since then, the percentage has remained stable, suggesting that the rapid expansion of machine-written text may have reached its limit.
“There’s a danger of ‘recycled knowledge’ that comes with AI-generated contents,” the editor of Myrtle Pen Publishing said to a TNC Correspondent.
“AI may be beneficial for research or checking for technical information like weather reports, but for generating original ideas, I doubt that.“
She added, “That is why Google search will always favor human-written contents because they come with a person’s learning journey and the exposure one has with life’s experiences evetyday- clearly, AI cannot keep up with that.”
Researchers believe this slowdown is linked to how AI-generated content performs in search results.
A separate study by Graphite found that 86% of top-ranking articles on Google were written by humans, compared with only 14% produced by AI.
Similar patterns appeared in responses from AI chatbots such as ChatGPT and Perplexity, which cited human-written sources 82% of the time.
Graphite stated that publishers may have learned that AI-generated material “does not perform well in search.”
Many content producers are adjusting their strategies as automated writing tools fail to attract readership or search visibility.
Experts note that identifying AI-written content is increasingly complex. Graphite used an AI detection tool called Surfer to classify articles and reported a false positive rate of 4.2% and a false negative rate of 0.6% when testing against known samples.
Still, many new articles combine both human and AI input, making clear distinctions difficult.
The rise of AI content has also influenced public trust. A Pew Research Center survey published in October 2025 found that only 2% of Americans regularly receive news from AI sources, while 75% say they never do. Of those who encounter AI-generated news, few say they fully trust it.
Economics have played a central role in the shift. It costs less than one cent to produce an AI-generated article, compared with $10 to $100 for human-written work.
This cost gap has driven widespread adoption of AI in online publishing, especially among content farms and marketing sites.
However, researchers warn that continued reliance on AI could create long-term risks. If new AI systems train mainly on machine-written material, the quality of future models could decline, a process known as “model collapse.” In such a scenario, online information may become repetitive and less reliable as originality fades.
Despite these concerns, most experts agree that the internet has not been overtaken by AI writing.
The proportion of AI-generated content appears to have stabilized at around half of all new articles, and publishers are increasingly turning to human–AI collaboration rather than full automation.
