AI governance for newsrooms: A practical starter guide

Every newsroom already has an AI policy. Most just haven't written it down.

Right now, somewhere in your building, a reporter is pasting a leaked document into a chatbot to summarise it. A social editor is using AI to draft ten headline variants before lunch. An intern is "just checking" a quote against a transcription tool nobody signed off on. None of this is malicious. All of it is happening without anyone in the building agreeing it should.

That's not an adoption problem. That's a governance problem and it's the one keeping media leaders up at night this year.

Why this is suddenly a boardroom conversation

AI governance used to be the kind of phrase that lived in tech company compliance decks, several steps removed from the newsroom floor. Not anymore. Boards are asking editors what their AI policy is. Funders are asking it as a condition of grants. Audiences, increasingly sceptical of what's real, are asking it without even knowing the term.

The trouble is, most of what's written about "AI governance" is built for banks and insurers, not newsrooms. It talks about model risk registers and enterprise procurement committees. It doesn't talk about what happens when a trusted reporter wants to use a tool nobody's vetted, or how you keep byline integrity intact when half your desk is quietly experimenting with the same three apps.

Editors don't need a governance framework. They need a newsroom one.

Oversight doesn't mean a committee that meets quarterly

The instinct, when "governance" gets mentioned, is to reach for structure, a committee, a policy document, a sign-off chain. And some of that is necessary. But the newsrooms actually managing this well aren't the ones with the thickest policy binder. They're the ones who've worked out where editorial judgment needs to sit in the process, and made sure it can't quietly get skipped.

That's a different design question to "who's on the AI committee." It's about which decisions can be delegated to a tool, which absolutely cannot, and who notices when the line gets crossed because it will, usually by someone with good intentions and a deadline.

Tool approval is where most policies quietly fall apart

Here's the pattern we see again and again: a newsroom writes a sensible-sounding AI policy, then a journalist finds a new tool that solves a real problem faster than anything on the approved list. The policy says "get sign-off first." The deadline says otherwise. Guess which one wins.

A workable approval process has to survive contact with a Tuesday afternoon deadline, which means the real design challenge isn't the approval criteria themselves, it's how fast and low-friction the path to "yes" can be, without becoming a rubber stamp. Get that balance wrong in either direction, and you either end up with shadow AI use everywhere, or an editorial team that resents the policy enough to route around it entirely.

Your editorial team is the policy, not the audience for it

The most common mistake we see isn't a weak framework, it's treating governance as something written for the newsroom rather than with it. Policies imposed top-down without the people actually using these tools every day tend to have a short shelf life: technically in force, practically ignored.

The newsrooms getting this right are running it more like an editorial conversation than a compliance rollout, surfacing what people are already doing (sanctioned or not), being honest about where the real risks sit, and building enough trust that when something new and uncertain comes up, someone flags it instead of quietly working around it.

Where to actually start

If you're staring at a blank page trying to write "the AI policy," you're starting in the wrong place. The newsrooms that get this right don't start with a document. They start by mapping what's actually happening across their desks right now, working out which of those uses carry real editorial risk, and building oversight around that, not around a hypothetical worst case borrowed from a tech company's compliance template.

It's less a single document and more an operating rhythm: a way of deciding, reviewing and adapting as the tools and the risks keep changing. Which is precisely why it needs to be built for your newsroom, not lifted from someone else's.

This is a natural companion to The Signal, Episode 7, where we dig further into how newsrooms are navigating AI governance in practice. If you're trying to work out where your newsroom actually stands, get in touch, this is exactly the kind of thing we help editorial teams figure out.