Microsoft: USA TODAY uses AI to accelerate public records requests

What the article is about

Published by Microsoft in June 2026, this case study describes how USA TODAY Network built an AI agent to accelerate Freedom of Information Act (FOIA) and state public records requests. FOIA requests are central to investigative journalism but procedurally demanding: each letter must identify the right agency, cite the relevant statute, specify the documents requested in legally appropriate language, and follow formatting requirements that vary by jurisdiction.

Context

USA TODAY Network — one of the largest news organizations in the United States, with hundreds of local and national publications — integrated an AI agent into Microsoft Teams and Outlook. The agent operates as a workflow assistant: a journalist identifies the documentation a story requires, the AI shapes it into a proper legal request, assists with routing to the appropriate government agency, and the journalist reviews, edits, and sends the final letter.

Senior Product Manager Stephen Harding led the implementation alongside colleagues at Newsquest, a related publishing group. The case study quotes Harding on the core design principle: “AI handles the mechanics. Journalists keep the judgment.”

Key takeaway

At least five or six front-page stories have come from requests enabled by this agent. Composition time for public records letters, which previously took approximately one hour per request, has been substantially reduced. Journalists report that the freed time goes into actual reporting — interviews, source development, community engagement — rather than administrative work.

The most significant aspect of the implementation is its scope. The AI is not generating story ideas, drafting articles, or making editorial decisions. It is applied to one specific bottleneck where accuracy and format consistency matter more than voice or judgment: the procedural work of asking government agencies for documents in the right way. This narrow application, focused entirely on a task with clear inputs and verifiable outputs, is what makes the results credible.

For newsrooms that are resistant to broader AI adoption, this model offers a starting point: identify the administrative tasks that consume journalist time and apply AI there first, before addressing anything that touches editorial content directly.

Who it is useful for

Editors, investigative reporters, and news organization product managers evaluating where AI can produce measurable impact in newsroom operations without requiring changes to editorial standards or reporter autonomy.