We analyzed 900,000 newly created internet pages in April 2025 and located that 74.2% of them contained AI-generated content material.
At Ahrefs, our machine studying workforce has constructed an AI content material detector (codenamed bot_or_not). We’re about to launch the AI content material detector for Ahrefs prospects to make use of, so we determined to place it via its paces with a query we’ve been dying to reply:
What share of recent content material is AI-generated?
We’re about to launch our AI content material detector as a part of the Web page Examine instrument in Website Explorer.
We used bot_or_not to research 900,000 English-language internet pages that have been newly detected by our internet crawler in April 2025. We analyzed one web page per area (so we examined content material from 900,000 totally different domains). Every web page was categorized in accordance with the share of the web page our mannequin detected as being AI-generated.
Right here’s what our content material detector discovered:
- 2.5% of pages have been categorized as “pure AI.”
- 25.8% have been categorized as “pure human.”
- 71.7% have been categorized as a mixture of the two.
Of people who contained a mixture of AI and human content material:
- 25.86% confirmed reasonable AI use (11%–40% of the web page content material was categorized as AI)
- 20.50% confirmed substantial AI use (41%–70%)
- 15.51% confirmed dominant AI use (71%–99%)
I used to be shocked that just about three-quarters of the pages we analyzed included AI content material. However the longer I sat with the information, the extra it made sense.
Free, quick AI content material technology is offered natively in Google Docs, in Gmail, in LinkedIn. AI can summarize Slack messages, conduct article analysis, shorten transcripts. It’s actively troublesome to keep away from the massive, shiny “generate” button, and can solely turn out to be harder as generative AI turns into additional embedded in functions.
An instance of generative AI obtainable natively in Google Docs.
And, as extra AI-generated content material is created, the brand new content material that references that AI-generated content material will, in flip, embody AI-generated materials. The presence of AI content material will unfold at an ever quicker price, tainting the whole lot it touches to a better or lesser extent.
If extra validation was wanted, after I surveyed 879 content material entrepreneurs, 87% reported utilizing AI to create or assist create content material. Simply 13% reported that they didn’t use AI in any capability.
And, unsurprisingly, weblog posts have been the commonest content material sort created by AI: 
LLMs are too handy and too low-cost to be ignored. Utilizing generative AI in content material creation is rapidly changing into the default.
No AI content material detector is ideal. Totally different detection strategies have totally different strengths and weaknesses, however all AI detection instruments share comparable struggles: they’re typically skilled on slim datasets, they wrestle with partial detection, and they’re susceptible to humanising instruments.
Our workforce understands these limitations and constructed our instrument with these issues in thoughts. We’ve had nice outcomes from testing our content material detector on Ahrefs content material, however—like each different market-leading content material detector—it’s going to by no means be 100% correct.
Support authors and subscribe to content
This is premium stuff. Subscribe to read the entire article.
