Over one-third of net pages revealed after the discharge of ChatGPT present indicators of being written by AI, in line with a new study from Pew Research launched on Thursday. The report corroborates other studies that element how a lot of the net’s newer net pages are actually both written by or “considerably edited” by AI, the agency says.
It additionally arrives shortly after web infrastructure supplier Cloudflare reported that bot net site visitors had overtaken human net site visitors — a milestone that was reached earlier than the corporate had estimated.
Pew’s information, nonetheless, is concentrated not on who or what’s shopping the net, however on what’s being browsed. And apparently, a lot of it’s bots studying net pages written by different bots.
To compile the report, Pew mentioned it used the Common Crawl net archive to gather almost half one million English-language net pages from the previous 5 or so years, beginning a few years earlier than ChatGPT’s November 2022 launch. Pew then used Open Pangram’s expertise to detect what number of had been probably written or closely edited by AI.
In a random pattern of 10,000 net pages collected in July 2026, round 10% confirmed “vital indicators of AI authorship,” Pew mentioned.
Nevertheless, Pew identified {that a} random pattern like this could inevitably embrace older net pages — ones revealed earlier than AI writing instruments even existed, and due to this fact couldn’t have been AI written.
Picture Credit:Pew Analysis
To get a greater sense of how a lot of the net is now being written by AI, Pew filtered out the older net pages and targeted solely on these revealed after the discharge of ChatGPT.
On this similar snapshot with the older pages eliminated, indicators of AI authorship had been present in over one-third (35%) of the pages.
Drilling down into domains themselves, Pew discovered that URLs with a .com area confirmed indicators of AI authorship at round 10x the speed of a .edu or .gov area (each of which had been round 1% AI authored). As well as, .org domains had solely a 4.6% price of AI authorship.
In fact, this evaluation will not be good. Pangram, like different AI-detection instruments, can misclassify pages as AI written once they weren’t. However at scale, the info is probably going at the very least directionally right.
As well as, Pew discovered that different supposed tells of AI authorship had additionally elevated through the years, like the usage of em dashes, Oxford commas, and phrasing like “it’s not X, it’s Y,” amongst different issues.
If you buy by means of hyperlinks in our articles, we may earn a small commission. This doesn’t have an effect on our editorial independence.
