A major University of Maryland study suggests that AI-written newspaper articles are no longer a fringe experiment in American journalism. Researchers found that about nine percent of newly published American newspaper articles were at least partly written by artificial intelligence, and readers were almost never told.
The findings point to a practical shift inside newsrooms: AI is being used most where stories are repetitive, data-heavy, or supplied by contributors outside the core staff. But the study also raises a basic trust problem. If readers cannot tell when AI helped write an article, they cannot judge the work with full context.
Where AI use is showing up most
The researchers analyzed over 250,000 articles from three datasets using the Pangram AI detector. They report that the detector has a false positive rate of just 0.001 percent for news content.
The overall number is striking, but the study shows that AI adoption is not evenly distributed across the news industry. Large newspapers, defined as papers with circulations above 100,000, had a much lower rate. Just 1.7 percent of their stories were flagged as AI-generated or hybrid.
Smaller local papers showed a different pattern. For those outlets, 9.3 percent of stories were flagged as AI-generated or hybrid. That suggests AI-written news is especially present in the part of the industry where staffing and production pressures can be most visible to readers.
There are also clear regional differences. Maryland had the highest rate listed in the study at 16.5 percent. Alabama followed at 13.9 percent, and Tennessee was close behind at 13.6 percent. The Northeast was lower, with New Hampshire at 2.9 percent and Massachusetts at 3.4 percent.
The sections most likely to include AI
The study found that AI use is concentrated in sections where articles often depend on structured information. Weather stories led the list, with a 27.7 percent chance of being AI-generated. Science and technology followed at 16.1 percent, while health reached 11.7 percent.
More sensitive beats showed lower rates. War stories were at 4.3 percent, and crime and justice stories were at 5.2 percent. That difference matters because it shows newsrooms may already be making informal judgments about where automated writing feels more acceptable.
The examples in the source point to a wide range of uses. AI appears in fully automated weather reports and in AI-written advice columns like “Dear Annie.” One outlet, the Argonaut, was found to be a fully AI-generated newspaper staffed entirely by fictional reporters.
That range matters because “AI-written” does not describe just one workflow. It can mean a narrow, repeatable format, a hybrid story, a contributed opinion article, or an entire publication model built around automation.
Ownership and opinion pages change the picture
The study also found that ownership is a major factor. Boone Newspapers led with over 20 percent of its articles attributed to AI. Advance Publications followed at 13.4 percent.
Other major chains showed far lower use. Nash Holdings, Lee Enterprises, and Digital First kept their AI use below two percent. The gap suggests that company policy and newsroom culture may shape AI adoption as much as article type or market size.
The study also notes a tension in the broader media business. Several major media groups are suing AI companies for scraping their articles to train language models, while some newsrooms are quietly publishing AI-generated stories themselves.
Opinion sections are another important area. An analysis of 45,000 op-eds and commentaries from The New York Times, Washington Post, and Wall Street Journal between 2022 and 2025 found a 25-fold increase in AI use, from just 0.1 percent to 3.4 percent.
At those publications, opinion pages are now 6.4 times more likely to feature AI-written content than regular news stories. Guest contributors, including politicians, Nobel winners, and CEOs, are much more likely to use AI than staff journalists.
Disclosure remains the weak point
The most direct reader issue is transparency. Of 100 articles flagged as AI-generated, only five told readers that AI was involved. The study also found that just seven out of 100 newspapers surveyed even allow AI-generated content in their public editorial guidelines.
Public bans do not fully solve the issue. The New York Post and Michigan Daily both publicly forbid AI use, yet both published stories Pangram identified as AI-generated.
The study also followed ten veteran reporters over time and found a sharp change after ChatGPT’s release. These journalists moved from almost no AI use before 2023 to an average of 40 percent of their 2025 stories written with AI assistance. One journalist rose to 90 percent AI-generated articles in just two years.
According to the study, AI-written articles tend to have fewer details, vaguer time references, and more formal language. None of the journalists disclosed their use of AI.
What the researchers recommend
The researchers argue that newspapers need clear public rules. Their recommendations distinguish between lighter assistance and full article generation.
- Grammar checks could go undisclosed.
- Summaries might require disclosure.
- Full article generation should be off-limits.
- Opinion pieces should include mandatory author declarations about AI use.
- Guest columns should face stronger checks.
Those recommendations focus on the reader’s right to understand how a story was made. The study does not frame every AI use as the same. Instead, it separates routine editing help from automated production and undisclosed authorship.
For publishers, the central question is now less about whether AI is entering newsrooms. The study indicates that it already has. The harder question is whether newspapers will explain that use clearly enough for readers to trust what they are reading.