A third of webpages published since ChatGPT’s launch show signs of AI authorship, study finds

Over one-third of webpages published aft the merchandise of ChatGPT amusement signs of being written by AI, according to a caller survey from Pew Research released connected Thursday. The study corroborates other studies that item however overmuch of the web’s newer webpages are present either written by oregon “substantially edited” by AI, the steadfast says.

It besides arrives soon aft net infrastructure supplier Cloudflare reported that bot web postulation had overtaken quality web postulation — a milestone that was reached sooner than the institution had estimated.

Pew’s data, however, is focused not connected who oregon what is browsing the web, but connected what is being browsed. And apparently, overmuch of it is bots speechmaking webpages written by different bots.

To compile the report, Pew said it utilized the Common Crawl web archive to cod astir fractional a cardinal English-language webpages from the past 5 oregon truthful years, starting a mates of years earlier ChatGPT’s November 2022 release. Pew past utilized Open Pangram’s exertion to observe however galore were apt written oregon heavy edited by AI.

In a random illustration of 10,000 webpages collected successful July 2026, astir 10% showed “significant signs of AI authorship,” Pew said.

However, Pew pointed retired that a random illustration similar this would inevitably see older webpages — ones published earlier AI penning tools adjacent existed, and truthful couldn’t person been AI-written.

% of webpages showing important AI editing oregon authorship
Image Credits:Pew Research

To get a amended consciousness of however overmuch of the web is present being written by AI, Pew filtered retired the older webpages and focused lone connected those published aft the merchandise of ChatGPT.

In this aforesaid snapshot with the older pages removed, signs of AI authorship were recovered successful implicit one-third (35%) of the pages.

ScreenshotImage Credits:Pew Research

Drilling down into domains themselves, Pew recovered that URLs with a .com domain showed signs of AI authorship astatine astir 10 times the complaint of a .edu oregon .gov domain (both of which were astir 1% AI-authored. In addition, .org domains had lone a 4.6% complaint of AI authorship.

Of course, this investigation is not perfect. Pangram, similar different AI detection tools, tin misclassify pages arsenic AI-written erstwhile they weren’t. But astatine scale, the information is apt astatine slightest directionally correct.

In addition, Pew recovered that different expected tells of AI authorship had besides accrued implicit the years, similar the usage of em dashes, Oxford commas, and phrasing similar “it’s not X, it’s Y,” among different things.