AI has matured from an instrument for efficiency into one that unlocks investigative discovery. The ultimate newsroom function of AI won’t be saving journalists a few minutes on rote tasks; it’ll be helping them uncover stories that would otherwise remain hidden.

The promise of AI is its ability to sift through massive amounts of information presented in any format (audio, video, text, images, etc.) to help reporters pinpoint what warrants a closer look. Take the Epstein files. A few years ago, a drop of 23,000 documents would have been considered flooding the zone because it would have taken journalists a long time to identify what was newsworthy. AI has changed that equation. Now news organizations are able to rapidly extract all the text, analyze it, and search for relevant topics and people. Flooding the zone is no longer an effective media strategy.

At The New York Times, we’ve used AI for dozens of reporting projects. This is always done in accordance with our principles for using A.I. in the newsroom, supplementing the work of reporters and editors. We rely on journalists’ expertise at every stage of the process, from identifying what to look for in a trove of documents to fact-checking the results.

Following Charlie Kirk’s assassination, we transcribed 2,500 of his podcasts and videos to get a sense of the breadth of his remarks. Our reporters and graphics editors watched many of his on-campus debates to identify aspects of his rhetorical style. Then the AI Initiatives team was able to scour dozens of these debates to find more examples of the ways that he spoke with students, resulting in an illuminating interactive. In 2022, The Times did a similar analysis of Tucker Carlson’s comments on his Fox News show. That project took one year to complete. The Charlie Kirk project took two weeks.

One tool that has enhanced our reporting is Cheatsheet, an A.I.-powered spreadsheet app that fuels large-scale investigative projects. Dylan Freedman, our AI projects editor, conceived of Cheatsheet as a way to search, summarize, classify, and translate information quickly. After ingesting vast datasets, from FOIA documents to video footage, Cheatsheet automates the heavy lifting of analyzing information. We’ve used it for projects that involved parsing hundreds of hours of Zoom meetings, photos of handwritten documents in Arabic, TV and podcast transcripts, and much more.

Here’s how Dylan illustrated Cheatsheet’s ability to make sense of all this data:

There are examples across the industry of how newsrooms are using AI for accountability journalism:

• Norwegian news organization iTromsø developed Djinn, an AI tool that sifts through government documents and municipal archives, extracting and analyzing data to find story ideas. The tool accelerated the newsgathering process, resulting in numerous stories by the 25-person newsroom.
• CalMatters launched its Digital Democracy portal, making it easy for California residents to see which bills their elected officials proposed, how they voted on legislation, how much money they raised, and much more.
• Helsingin Sanomat, a Finnish news organization, built its Watchdog tool to automate some of the newsgathering process. It scans public records, emails, and other documents to flag potential stories and insights to editors.
• Chalkbeat monitored school board meetings across 30 states with LocalLens, a database that uses AI to transcribe and summarize public meetings. Tools like this help local reporters stay informed about meetings they can’t attend in person, expanding the scope of their coverage.

Investigative reporting has historically been limited by the time it takes to review thousands of documents, watch hundreds of hours of video, or attend hundreds of public meetings. AI tools are empowering newsrooms to report stories that would’ve otherwise been impossible to pursue.

Rubina Madan Fillion is the associate editorial director of AI Initiatives at The New York Times.