Updates

New data and upgraded models

We've added articles published through June 12, 2026 and moved to newer models for our analysis. AI detection now runs on Pangram 3.3, and we sort articles into topics using the Qwen 3.5 language model. Both the Recent News and Opinions datasets have been fully re-scored and re-tagged. We've also published the data on HuggingFace — the Recent News dataset is ready now, with Opinions alongside it — and every author byline has been anonymized with a stable alias.

AI News Audit goes live

We have launched AI News Audit to help track the use and disclosure of AI in American newspapers. The first batch of data was collected roughly twice a week between June 15, 2025 and September 15, 2025. It includes 186,507 articles from online editions of 1,528 American newspapers published during that summer. At launch the articles were tested with Pangram 2.0. (These articles were later re-scored with Pangram 3.3 in May 2026 — see the update above.)