Report: AI Use in Newspapers Is Widespread, Uneven and Rarely Disclosed

More than 9% of all news in U.S. newspapers contains at least some text created by artificial intelligence (AI), according to new research led by University of Maryland computer scientists.

The study, based on analysis using a tool developed by Pangram Labs, also found that AI use in published articles is rarely disclosed, and that usage is much higher in smaller local outlets than nationally circulated papers. It was posted Thursday on the preprint server ArXiv. 

“As a reader, you generally don’t have a way to know if the news or the opinions are coming from a human or from AI,” said Mohit Iyyer, an associate professor of computer science and the study’s senior author. “It makes it hard to trust the factuality of the content,” he said, both because generative AI chatbots frequently get facts wrong and often engage in “hallucination,” or fabrication.

The paper’s first author was Jenna Russell, who is advised by Iyyer and had previously compared the accuracy of AI detection systems, finding Pangram Labs’ product most effective. Other coauthors were Marzena Karpinska at Microsoft and Destiny Akinode, Katherine Thai, Max Spero and Bradley Emi with Pangram.

The team examined 186,000 articles published by 1,500 newspapers in the summer of 2025 and found that 9.1% were either AI-generated or mixed. Only 1.7% of articles at papers with circulation of more than 100,000 were partially or fully AI-generated. But at smaller ones, nearly 1 in 10 (9.3%) was shown to contain significant AI content. 

Some major companies’ papers contained significant AI content. Boone News Media had the highest percentage (20.9%), well above the second-highest, Advance Publications (13.4%), the paper found.

“This disparity—that communities served by smaller papers and some corporate owners get more AI-made content than people in larger cities with bigger papers or different owner groups—is worrying, and may be a consequence of collapsing news economies, the result of news deserts,” said Emi, co-founder of Pangram.

AI-generated content was not limited to news coverage; the report identified 219 articles containing AI content on the opinion pages of The New York Times, The Wall Street Journal and The Washington Post.

The use of AI in these pages was 4.5% overall—significantly higher than in those outlets’ news pages (0.7%). Most of the opinion articles containing AI content were from guest contributors, not regular columnists, the paper found. 

The team also manually analyzed 100 articles flagged for AI use by Pangram and found that only five disclosed AI use, while only seven of the newspapers have any public policies on AI use, “leaving readers largely unable to determine the role AI plays in article authorship,” according to the report.

Daniel Trielli, assistant professor of media and democracy at UMD’s Philip Merrill College of Journalism, called the study a fascinating look at how journalism is changing—not always for the better.

“What’s most jarring is how many newsrooms are actually using this technology without saying they’re using it,” said Trielli. “What we teach students at Merrill College is that part of the deal we as journalists make with the public is we’re going to be as transparent with them as we can be.”

Trielli said he expects more attention to be paid to proper uses of AI in the journalism field, where the technology might help with behind-the-scenes data analysis, generating ideas for stories and customizing content based on audiences. There may even be uses for public-facing AI writing itself, carefully considered and always disclosed.

These new tools can increase the capacity of journalists to do more and better work,” he said. “We don’t want to use it to create more—but worse—work.”

This article is based in part on a news release by Pangram Labs.