Is AI Writing the News?
Recent analysis shows 9% of recently published news contains AI-generated text.
Introduction
Researchers from Univ. of Maryland, using AI detection from Pangram, found that more than 9% of all news coverage contained AI-generated text, while AI text has also appeared on the opinion pages of The New York Times, The Wall St. Journal, and The Washington Post.
AI is profoundly reshaping the content landscape of the internet. It has given content creators the ability to generate, expand, summarize, and paraphrase pieces at scale. This new, powerful technology, combined with the disruptive trend of collapsing newspaper business models, has resulted in the rise of AI-generated content in U.S. newspapers. We see an urgency to track how much AI-generated content is being created and where readers are most likely to encounter it.
We estimate that ∼ 9% of newly-published newspaper articles are either partially or fully AI-generated.
AI in the News
9% of newly-published newspaper articles are either partially or fully AI-generated.
| Category | Percentage |
|---|---|
| Human | 90.9% |
| AI | 5.2% |
| Mixed | 3.9% |
Areas of focus for AI-generated text
Local newspapers are AI adopters
Local newspapers have faced economic pressure that has resulted in reduced resources and staff. Even prior to the rise of LLMs, local newspapers have been under pressure from consolidation and have seen difficulty monetizing their content giving rise to “news deserts.” As a result, many have adopted AI to operate more efficiently (Northwestern Report). What is the scale of this increase in AI usage for news articles?
Only 1.7% of articles at papers with circulation >100,000 are partially or fully AI-generated, versus 9.3% at papers below 100,000. These patterns suggest AI adoption is far more common among small and mid-sized outlets than among the largest national newspapers, which appear to enforce stricter editorial constraints on automation.
AI use by newspaper size
Share of articles with AI-generated text
| Category | Percentage |
|---|---|
| <100k circulation | 9.3% |
| >100k circulation | 1.7% |
Data-heavy articles are more likely to have AI-generated text
AI is more likely to be published in specific topic areas. For example, weather articles in our dataset exhibit the highest average AI likelihood (27.7%). Other topics with elevated AI usage include science and technology (16.1%) and health (11.7%). By contrast, content on more sensitive issues such as conflict and war (4.3%), crime, law, and justice (5.2%), and religion (5.3%) shows significantly lower rates of AI usage.
AI likelihood by topic
Average AI likelihood across categories
| Category | Percentage |
|---|---|
| Weather | 27.7% |
| Science & Technology | 16.1% |
| Health | 11.7% |
| Religion | 5.3% |
| Crime, Law, & Justice | 5.2% |
| Conflict & War | 4.3% |
Opinion pieces aren’t immune to AI
An analysis of Opinion Pieces from WSJ, NYT, and Washington Post reveals 4.5% articles contained AI-generated text in the June – September 2025 sample. Guest contributors published AI-generated text at a higher rate than staff writers. We found 428 pieces articles containing AI written text in our analysis and that number is rising. Opinion articles published at the NYT, WaPo, and WSJ are 6.4 times as likely to contain AI content than contemporaneous news articles from the same three newspapers.
A significant lack of disclosure
Because audience trust depends in part on whether AI involvement is made visible, we examined how often newspapers disclose such usage. In a sample of 100 AI-flagged articles from unique newspapers, 95% of authors and 97% of publishers did not acknowledge AI. The few disclosures we observed appeared only in environmental reporting, such as weather forecasts and air-quality alerts. Among three publications with formal disclosure policies, just two articles complied while another violated its own policy by omitting disclosure. Notably, two newspapers with explicit bans on AI nonetheless published articles our classifier flagged as AI- generated.
Explore the data
The AI News Monitor Data Dashboard allows readers and news observers to track trends in AI usage across news and opinion pages. This tool facilitates the exploration and monitoring of how AI use evolves over time.
Methodology
Researchers at the University of Maryland, collected these datasets of American newspapers:
- We collected 186,507 articles published by U.S. 1,528 newspapers with active RSS feeds (both local and national) from June–September 2025. The list of newspapers was gathered from onlinenewspapers.com.
- We collected 44,803 opinion articles: 16,964 from WSJ, 15,977 from WP, and 11,862 from NYT between August 2022 and September 2025.
We evaluated each article with Pangram AI detection software to obtain both an AI likelihood (from 0–100%) and a categorical label. We also performed topic classification on each article and linked each newspaper to circulation and ownership meta‑data when available.
Pangram's AI detection tool has been independently evaluated by staff at the Becker Friedman Institute at the University of Chicago. In the September 2025 study, Pangram had 0.08% a false‑positive rate for News and a .0.33% or less false-negative rate for News content. (University of Chicago Study)
