Utahn uses AI to read documents, assemble data, and surface the questions worth asking. Utahn does not use it to write stories, aggregate other outlets' reporting, or produce additional versions of work we've already published. Every piece is reported, written, and edited by a Utahn journalist.
The question for journalism is no longer whether to use AI. It's too powerful and useful to ignore. The most pressing questions now are when to use it, the best methods for disclosure, and under what circumstances it should or shouldn't be used. The industry, like every other industry, is grappling with how best to integrate AI into its workflow. That's how we talk nowadays. Workflows, circling back, taking a holistic approach. My word.
As the technology improves daily and becomes an increasingly effective tool, these are not simple questions to answer. Trust and credibility should be the first thing anyone weighing them thinks about. It's not clear the media industry cares about such things. Journalists are wildly distrusted, been that way for a long time, having rightly earned almost no credibility with the public. AI hasn't earned it either, but its verdict isn't in yet. (I recently wrote about the AI side of things at length.)
Only 28% of Americans trust the media to report the news fully, accurately, and fairly, according to Gallup's most recent poll. That's the lowest number on record in more than 50 years of asking. To be honest, I'm surprised that number is as high as it is.
As I've written in the past, trust in media has collapsed, and much of that collapse is deserved. Too many institutions abandoned impartiality while still claiming the mantle of objectivity. They slant coverage. They minimize inconvenient facts. Editorial boards tell readers what to conclude (some even call for the state to mobilize the National Guard against its own citizens), as if readers are unable to think for themselves.
AI's problem is different. While 51% of Americans now research topics with it, up from 37% in April 2025, only 21% trust what it produces most of the time, according to Quinnipiac. Asked whether AI development is being led by people who represent their interests, 47% said no and 46% said they don't know enough to judge. The public has given up on the media. It hasn't decided about AI yet.
What Utahn AI agents actually do
Utah publishes an enormous amount of material that is technically available and functionally unread. Contract and expenditure records sit on Transparent Utah, where roughly a thousand state and local entities have posted more than 250 million records. Cities post council minutes as scanned PDFs. The Legislature runs one short general session each winter. This year it ran 45 days, took up a record 1,015 bills, and passed 541. Most of the ones that moved were amended or substituted along the way.
We run dozens of agents against that material.
One watches the Legislature's bill pages and flags every substitute to anything we are tracking, with the changed language pulled out so an editor can see exactly what moved. Another pulls municipal contracts by vendor across every portal in the state that publishes them, which is how you find the same firm collecting from six cities that have never compared notes. A third reads council minutes and marks the agenda items that involve money.
None of that is writing. It’s reading and research, at a volume previously impossible without this technology. Newsrooms that refuse AI for this kind of work are choosing to leave public records unread and readers underserved.
The fight the industry is actually having
Some readers have seen the disclosure line and assumed Utahn is an AI news aggregator. We're not. AI is not much use for that anyway. Aggregation is retrieval of material somebody else already judged, and running it through a second machine adds nothing. Google Alerts is free and has been around since 2003.
That confusion is understandable, because aggregation of a sort is what most of the current disputes over AI in newsrooms are actually about.
McClatchy built a tool that takes a finished story and generates rewritten versions for different audiences. Reporters at the Sacramento Bee, the Miami Herald and the Idaho Statesman began withholding their bylines from the output, and unions at several papers filed grievances arguing the rollout violated notice provisions in their contracts. In June, the Centre Daily Times in State College, Pennsylvania, voted to unionize over it, the first newsroom under The NewsGuild-CWA to organize with AI adoption as a primary reason. Every eligible editorial employee signed a card. The Guild's own account notes that reporters and editors were spending their time fixing errors the tool introduced. A time-saver that costs time.
At POLITICO, an arbitrator ruled in November 2025 that the company had violated its union contract by deploying two AI products without the required notice or human oversight: live summaries published during the 2024 convention and vice presidential debate, and a tool that generated branded policy reports for Pro subscribers with no editor involved. In May, the company agreed to shut both down. The arbitrator's finding was blunt: on accuracy and accountability, the output did not measure up to the human work it was standing in for.
In each case, a newsroom used the machine to make more copy out of copy it already had, and put a reporter's name on it.
Retrieval at scale, aimed at primary documents, is a different proposition. It is most of what Utahn has built. Nobody files a grievance over an agent reading a bill.
AI policies in other newsrooms
The Associated Press landed in nearly the same place as Utahn in July, when it updated its newsroom standards to permit AI for early-stage research, document summarization, transcription, translation, headlines, story summaries and shotlists, and grammar and search optimization. Every output is reviewed and edited by an AP journalist, and the guidance states plainly that AI does not replace reporting, sourcing, editorial judgment or verification.
Eileen Drage O'Reilly, who runs standards and AI practices at Axios, described a stricter version to the Columbia Journalism Review this summer: no published text created with generative AI at all, though it may help build some data visualizations and suggest headlines that reporters then edit. Let's see if that's still their policy in a year. I doubt it. Utahn sits closer to the AP.
AP has revised its guidance repeatedly since 2023. We've revised ours twice since launch. Anyone publishing an AI policy right now and calling it finished is telling you they haven't thought about it much.
What a journalist’s job actually is
Done right, AI should make the job of a journalist bigger, not smaller. The best use of a reporter's time is on the ground, talking to sources, and understanding what a community is actually worried about. The best newsrooms will use these tools to get reporters out from behind a desk.
An agent can hand a journalist a fact sheet, a stack of verified documents, and a list of questions worth asking. It cannot ask those questions of a county commissioner, work a source deciding whether to divulge information, or explain why a story matters to the people who live here.
One thing we do not claim: that AI removes bias. A model carries its own, trained on the same corpus that produced the coverage everyone is already suspicious of. What guards against bias here is structural, not technological. Utahn has no editorial board and makes no endorsements, and every factual claim is tied to a primary source the reader can click.
It's not a journalist's job to tell a reader what to conclude. An unfortunate amount of what's published in the news section of most legacy institutions is opinion dressed up as reporting. The outcome is determined before a single word is written, with the aim of pushing a particular worldview or preaching to a subscriber base that likes hearing the same sermon on repeat. The widespread belief in those newsrooms that this is their top priority is the best explanation for where their credibility went.
Oh, and slop predates AI by a long stretch. The tool did not invent the problem. The journalists most against AI act as though they're Tolstoy-caliber writers, when in fact they've been publishing slop for years. AI can replace those types of journalists right now, but they should have been replaced or changed their approach to actually serve their reader a long time ago.
What journalism needs most is to reorient itself away from telling people what to think, toward better understanding what the communities they serve think and how they actually feel, and toward holding those in power, who make decisions that affect their lives, to account.
Newsrooms should use every tool available, including AI, to go back to doing that work. That’s the job. It always was.
