USA TODAY brings AI into real newsroom workflows
In journalism, trust isn’t a tagline — it’s the job. And when the work moves at speed, the difference between what’s possible and what gets done often comes down to one thing: time.
Instead of chasing “AI everywhere,” newsroom teams at USA TODAY focused on a practical goal: using AI with intentionality to remove friction from one of investigative reporting’s most essential workflows — public records requests — without compromising editorial integrity.
The bottleneck hiding in plain sight: records requests
Public records requests (FOIA at the federal level, plus state records requests) are a cornerstone of investigative journalism. But they’re also easy to delay when the day is already full.
As Thomas Elia, a newsroom leader at The Palm Beach Post, put it: these requests “take time to compose,” must be “carefully written,” and routed to the right agency.
Drafting a request could mean “spending an hour drafting out a legal letter”. That’s time that could have gone to reporting, interviews, and the human details that make a story stick.
The shift: a “virtual colleague” inside the tools journalists already use
The team didn’t reinvent the newsroom. They built an assistive experience that fits where journalists already work—Teams and Outlook—so there’s no tool-switch tax.
How it works (in plain terms)
At a high level, the agent supports a workflow journalists already know — while compressing the slow parts:
- A journalist starts with the story question that needs proof
- The agent helps shape it into a usable request
- It helps route it correctly
- The journalist reviews, edits, and sends — accountability stays human
That’s the point: AI handles the mechanics. Journalists keep the judgment.
How they built it: familiar on the surface, deliberate underneath
On the USA TODAY Network side, Stephen Harding, Senior Product Manager, highlighted Microsoft 365 Copilot’s advantage: integration with the Microsoft tools teams already rely on — and the ability to point the agent to internal knowledge sources (including sensitive files in SharePoint and OneDrive).
On the Newsquest side, Calum Banister, AI Agent Orchestrator, described how integrations and flexible knowledge sources helped them “break apart a workflow to its base components,” and — coming from a low-code background — pick it up quickly.
The litmus test: front pages — and what happens after the headline
For Jody Doherty-Cove, Head of AI at Newsquest, usefulness is the only metric that matters: a tool is only real if it works. And in local journalism, front pages are a litmus test — an editor’s decision that this is the story the community needs most.
He’s seen “5–6 front page stories” come from requests enabled by the agent.
But the request is just the start. The reporting that follows: the calls, the interviews, the accountability, the human consequence, that’s what turns information into meaning.
The guardrail that matters most: accountability stays human
If you’re building with AI, you don’t just need capability, you need confidence. USA TODAY’s teams were clear on the boundary: “AI is a tool. It’s not in charge… We’re still in control.”
Stephen described the principle behind responsible use: a human-in-the-loop workflow where AI assistance is reviewed — and accountability stays with the human whose name is attached to the work.
What every business can take from this
If you’re wondering what “AI done right” looks like in your world, start here:
- Begin with a human problem. Find the tasks that drain time and dull momentum.
- Build in the flow of work. Familiar tools reduce friction — and friction decides adoption.
- Keep people in charge. AI can speed the process; people own the judgment.
- Make trust visible. Human review, clear guardrails, and accountability that doesn’t blur.
The best AI stories aren’t about models. They’re about momentum: what people can do when friction disappears. For USA TODAY, that means more time reporting in communities — following leads and telling the stories that matter.
Want to take the next step? Start here:
