Are you reading more text written by an AI bot? A Pew Research study published this week says it’s very likely.
That’s not breaking news if you’re aware of the Dead Internet Theory that dates back to 2021. CNET writer Trisha Jandoc dove into that theory last year, which basically says people believe the internet died 10 years ago in 2016 and that much of what we see online now is AI-generated.
There is some truth to that. Pew researchers scanned nearly half a million English webpages from the past five years using Common Crawl, a web archive. That dates back to before ChatGPT was made publicly available in November 2022. Researchers ran those pages through Open Pangram, an AI detection tool, to see if it was written or edited by AI. Then the researchers looked at a sample of 10,000 pages from last month to spot signs of AI-written content. They looked for phrases, words and language that are more common among AI bots than humans.
Researchers made it clear that AI detection models don’t always get it right. Sometimes they flag text written by humans as AI authorship and AI-written text as being written by a human.
Nonetheless, the study found that 10% of the samples “show significant signs of AI authorship.”
“One in 10 webpages as of July 2026 may seem modest. But the internet includes a mix of new and old material. A lot of pages in these random samples couldn’t have been written by AI,” the study findings said. “If we filter old webpages out of our samples and look at only the pages published after the release of ChatGPT, the trend is even more pronounced.”
Researchers say this trend began in 2022 when ChatGPT was released and increased as other AI bots launched, like Claude and Google’s Gemini.
Where is AI-written content spotted the most?
The Pew Research study shows that AI-derived text is most common on .com domains, with about 1 out of every 10 pages. Other domains have less AI-written work — 4.6% for .org and around 1% for .edu and .gov webpages.
Looking to spot AI-written text? Researchers say certain types of punctuation, words and phrases are more common in AI-generated text compared to humans. For example, em dashes and Oxford commas in lists. This is mainly because they’re trained on datasets of human text, so they mimic that style and, as a result, use them more than humans do. The researchers found that em dashes appear twice as frequently, negative parallelisms are more common and Oxford commas are used 63% more than in human-written text. AI is also more likely to use certain words, like “delve” and “interplay.”
“They also learn to recognize more subtle statistical patterns in word choice and sentence structure, which help them determine whether a particular piece of writing was likely AI-authored,” researchers said.
A few other common AI-written flags include key terms from your prompt being used repeatedly, repetitive explanations and a bot-like tone instead of a natural one. More importantly, AI bots even make it clear that their responses could be inaccurate, so it’s best to double-check elsewhere online. But in this case, Pew’s findings show that we may not be able to trust the sources that we’re verifying it against if that content is AI-generated, too.