USA TODAY Network has deployed a custom-built AI agent powered by Microsoft 365 Copilot to automate one of journalism’s most laborious tasks: drafting and routing public records requests. Microsoft spotlighted the in-house tool on June 2, 2026, describing it as a template for how newsrooms can weave generative AI directly into their daily beat work.

The agent operates within the familiar Microsoft 365 environment, allowing reporters to initiate a request from Word, Outlook, or Teams and have the system generate a legally sound draft, pull relevant boilerplate from a curated library, insert case details from a connected SharePoint list, and even log the request in a centralized tracking sheet—all without leaving the apps they already use.

The Tedious Art of Public Records Requests

For reporters, the Freedom of Information Act and its state-level equivalents are vital tools for holding institutions accountable. But the process of crafting a request, citing the correct statutes, tailoring it to a specific agency, and then managing the subsequent correspondence often eats days of time that could be spent analyzing documents or cultivating sources. Many requests get rejected on technicalities, and keeping track of dozens of simultaneous requests across a large network like USA TODAY’s—which spans hundreds of local outlets—has been a logistical nightmare.

Investigative teams traditionally rely on shared spreadsheets, templated Word documents, and endless email threads to coordinate requests. Crucial institutional knowledge, such as which agency requires a notarized signature or what phrasing triggers a faster response, lives in the heads of veteran journalists or scattered across siloed folders. The Copilot agent aims to bake that hard-won expertise into every request, instantly.

Enter the Copilot Agent

The tool, built using Microsoft 365 Copilot’s extensibility framework and Copilot Studio, acts as a conversational assistant. A journalist can say or type, “Draft a public records request for the City of Phoenix email server logs between January and March 2026, related to the water contract negotiations,” and the agent retrieves the correct template, fills in dates and names from a SharePoint list that tracks ongoing investigations, and generates the body with legally appropriate language. It then prompts the user to review, edit if needed, and send via Outlook with the request automatically tied to a tracking number.

“The agent doesn’t just spit out a generic letter—it learns the nuances of different jurisdictions,” said Sarah Lin, Director of Newsroom Technology at USA TODAY, in a Microsoft blog post. “If you’re filing in Florida, it knows that the Sunshine Law has specific requirements for electronic records. If you’re dealing with a particular department that prefers email over fax, it adjusts the routing accordingly.”

Under the Hood: How the Agent Works

At its core, the tool leverages Microsoft 365 Copilot’s ability to ground responses in organizational data via the Microsoft Graph. When a user makes a request, the agent performs several actions in sequence:

  • Template Retrieval: Using natural language understanding, it identifies the type of record (emails, contracts, bodycam footage, etc.) and pulls a pre-approved template from a SharePoint document library customized by the legal team.
  • Context Injection: It queries a secure SharePoint list that contains details about ongoing projects—subjects, key dates, involved parties—and inserts that context into the draft so reporters don’t have to re-type recurring information.
  • Jurisdictional Tuning: A custom model built with Copilot Studio evaluates the target agency and applies specific rules. For example, if the request is for California’s Public Records Act, it inserts a standard three-day notification clause. If it’s a federal FOIA, it adds a request for a fee waiver justification.
  • Draft Generation: Microsoft 365 Copilot generates the final text in a Word document, with placeholders for the sender’s contact information pulled from Azure Active Directory.
  • Routing and Tracking: Once approved, the agent can send the request directly via Outlook, attaching the Word file, and simultaneously update a Power Automate-driven tracking log so editors can see the status of all open requests in a Power BI dashboard.

USA TODAY’s IT team built the agent using a combination of low-code and pro-code tools. Copilot Studio handled the conversational logic and natural language triggers, while Power Automate facilitated the multi-step workflows. Custom connectors to internal databases were created with Azure Functions, and the whole system is governed by role-based access controls to ensure only authorized staff can view sensitive investigation data.

From Days to Minutes: Measurable Impact

Preliminary data from USA TODAY’s pilot, which ran across three newsrooms in the Southwest, showed a 72% reduction in the time spent drafting routine requests. Reporters who previously spent 90 minutes on a single city-level request were finishing them in under 20 minutes, including the final review. More complex multi-agency requests saw even larger gains, with the agent able to generate simultaneous filings in parallel.

The tracking dashboard, built in Power BI, gave editors real-time visibility into pending requests, overdue responses, and success rates by agency—data that previously took weeks to compile manually. “We’re now able to spot patterns, like which police departments consistently blow past statutory deadlines, and report on that as a story in itself,” Lin added.

Built on Microsoft’s AI Ecosystem

The agent is a prime example of Microsoft’s strategy to embed AI deep into vertical workflows. Since the release of Microsoft 365 Copilot in 2023, the company has steadily opened its platform to custom agents, first with plugins, then with the general availability of Copilot Studio in 2024. By 2026, the “agent” metaphor—where AI accomplishes multi-step tasks autonomously—has become a staple of the Microsoft 365 suite, with tools for HR, sales, finance, and now journalism.

Microsoft positions Copilot as the UI for AI in the enterprise. “This isn’t a chatbot that just answers questions—it’s an agent that takes action on your behalf, securely and within the compliance boundaries you set,” said Jared Spataro, Corporate Vice President of AI at Work at Microsoft, in the announcement. “What USA TODAY has done is a perfect illustration of how domain-specific agents can turn hours of grunt work into minutes of oversight.”

Crucially, all data processing remains within the Microsoft 365 compliance boundary. The agent does not train on customer data, and USA TODAY’s legal team can audit every action through the Purview compliance portal.

A Blueprint for Newsroom AI

Other major news organizations have experimented with generative AI for tasks like summarization and headline testing, but USA TODAY’s move marks one of the most ambitious operational deployments yet. By targeting a high-friction, rule-based process that benefits from both templating and adaptive logic, the agent avoids many of the ethical quagmires that have made editors wary of AI in editorial content.

“We’re not using AI to write stories or replace journalistic judgment,” Lin emphasized. “We’re using it to eliminate the administrative overhead that keeps journalists from doing their best work. The final wording, the decision to file, the pursuit of the story—that’s all human.”

The agent’s design also bakes in transparency. The draft email includes a note saying it was “generated with AI assistance,” and every suggested template must be approved by a legal reviewer before it becomes available to reporters. This approach aligns with ethical guidelines from groups like the Society of Professional Journalists, which stress disclosure and accountability.

Ethical Boundaries and Transparency

While the tool has been well received internally, outside observers have raised questions. “AI-generated legal documents, even in a journalism context, require rigorous oversight,” said Dr. Amelia Voss, a media ethics professor at Columbia University, who was shown a demo of the agent. “A misapplied statute in a request could lead to a denial or, worse, a court challenge that costs the newsroom time and money. The human-in-the-loop element is non-negotiable here.”

USA TODAY maintains that every request generated by the agent must be reviewed and explicitly approved by a journalist before sending. The system also logs every prompt, revision, and final output to maintain a chain of custody. That audit trail could prove invaluable if a newsroom’s request method is ever challenged in court.

Another area of caution is the handling of sensitive source information. USA TODAY built the agent with strict data separation: investigation details that could reveal whistleblowers or confidential informants are never ingested into the model. The agent deals only with the metadata of what records are being requested, not the underlying evidence or tips that prompted the request.

What’s Next for Copilot Agents?

Microsoft indicated that the USA TODAY deployment will be featured as a case study in its upcoming “AI at Work” summit, with the goal of inspiring similar agents in legal departments, procurement teams, and government agencies themselves—ironically, the same entities that field public records requests. The company also teased tighter integration with Microsoft Loop and Planner to turn requests into trackable project tasks.

For newsrooms, the success of this agent could pave the way for additional tools: an agent that automatically transcribes and indexes interview footage stored in OneDrive, or one that drafts FOIA appeals based on common denial patterns. The building blocks are already in place within the Microsoft 365 ecosystem.

As AI agents become more autonomous, the line between “assistant” and “coworker” will blur. USA TODAY’s experience suggests that when the technology is applied to defined, rules-heavy workflows, the gains are immediate and relatively low-risk. But that safety relies on continuous oversight, clear boundaries, and a culture that sees AI as a tool to amplify human skill—not replace it.

The Copilot agent for public records requests isn’t just a time-saver; it’s a signal that AI can make investigative journalism more efficient at a moment when newsroom resources are strained. By handling the paperwork, it frees up reporters to do what they do best: connect dots, question authority, and tell stories that matter.