1 In 10 Webpages Shows Signs Of AI Authorship, Pew Reports

Pew Research Center ran astir half a cardinal webpages done an AI detector and recovered signs of AI authorship aliases editing connected astir 10%.

Among pages published aft ChatGPT’s launch, the stock of AI-authored pages was moreover greater astatine 35%.

Pew’s Data Labs squad published their study connected August 20, little than a twelvemonth aft 2 different awesome estimates of AI-generated matter connected the web.

What The Data Shows

Pages connected .com domains thin to show signs of AI authorship astatine astir 10 times the complaint of pages connected .edu and .gov domains. In a six-month average, the complaint was 9.35% connected .com, against 4.59% connected .org, 1.03% connected .edu, and 0.76% connected .gov. 

All 4 domain types sat astatine aliases beneath 1% successful the sample collected earlier ChatGPT launched. They person separated since, pinch .com climbing fastest while .edu and .gov stayed adjacent to the 1% mark.

The Tells Are Spreading

Pew examined circumstantial markers successful pages published aft ChatGPT’s release. Em dashes accrued from 5.79 uses per 10,000 words successful early 2023 to 11.19 successful early 2026. Oxford commas saw a 63% rise.

The usage of words communal to AI writing, specified arsenic “delve,” “interplay,” and “testament,” has much than doubled. The antagonistic parallelism style, besides known arsenic the “it’s not conscionable X, it’s Y” structure, accrued from 0.87 to 2.36 uses per 10k pages, though it still remains rare.

Pew says nary of these traits tin pin down a azygous archive , since quality writers usage each of them. The declare is astir rates crossed ample sets of text.

I covered Ahrefs’ detector study successful July, which ended pinch an unfastened mobility astir really useful a detector people really is arsenic AI editing becomes much communal successful mundane penning tools. Pew’s information shows these markers appearing much often successful their samples. Whether this makes immoderate 1 marker much aliases little adjuvant for identifying AI is simply a different matter, and Pew didn’t research that.

Other Estimates Disagree

Graphite, an SEO firm, estimated that the stock of recently published English-language articles chiefly AI-generated was 49.9% in the first 4th of 2026. 

A preprint from Imperial College London, Internet Archive, and Stanford recovered that by mid-2025, 35% of recently published websites are detected arsenic AI-generated aliases AI-assisted (not peer-reviewed).

All estimates thin connected Pangram successful immoderate form, and Graphite besides utilized Copyleaks and GPTZero. 

Why This Matters

Signs of AI matter are astir prevalent connected the commercialized web. Pew’s .com discovery percent has climbed successful each reference since ChatGPT launched, while .edu and .gov person stayed nether 2%. Commercial pages are wherever the attraction sits, and that is the portion of the web astir SEO activity touches.

Pew’s period catches AI editing too, truthful a page a personification wrote and past cleaned up pinch a instrumentality lands successful the aforesaid class arsenic 1 AI produced commencement to finish.

Looking Ahead

AI editing is now a autochthonal characteristic successful Google Docs and Microsoft Word. Pew’s period already considers this type of writing, but nary of the studies separate it from matter wholly generated by AI.

Whether a page is accurate, useful, and worthy publishing is still decided by reference it.

Featured Image: Cater Image/Shutterstock