Most AI rollouts are planned as a software project. They land as a people project, and nobody warned the people.
Here's the pattern, as plainly as we can put it. A newsroom picks an AI tool, or a handful of them. Someone writes a policy. There's a training session, usually one, often on a call or over lunch. An email goes out saying the tools are available and everyone is encouraged to explore. Then very little happens, or the wrong things happen, and leadership wonders why.
The answer is almost never the tool. The rollout was planned as a technology project, and what it actually does is change how people do their jobs. That's the change-management problem hiding inside every AI rollout: the software is the easy part, and the people are the part nobody budgets for.
What I mean by change management
It sounds like consultant jargon, so let me be specific. Change management is the work of getting people from how they do their job today to how they'll do it next, including the awkward stretch in the middle where they're slower, unsure and a bit worried. In an AI rollout, that stretch is where most of the real outcomes get decided, and it's the stretch most plans skip.
Where it shows up
It tends to show up in the same four places.
Nobody says what's changing about the job. “AI will help you work faster” is easily heard as “AI will do some of what you do.” People start wondering which parts, and whether they're next. If leadership doesn't answer that question, the newsroom answers it for itself, and the version people come up with is rarely the generous one.
Learning time isn't anyone's time. Getting good with a tool takes hours of fumbling around. In a newsroom with no slack, those hours come out of lunch breaks and evenings. So the people who enjoy tinkering get good, and everyone else learns one trick, or avoids the tools altogether. That gap then gets mistaken for a difference in attitude.
Craft gets mistaken for resistance. Journalists take pride in how they write, edit and check things. A tool that produces a passable draft in seconds can feel like a quiet comment on all of that. When a sceptical editor pushes back, it gets filed under “resisting change.” Often it's a fair question about standards, such as who is accountable when the output is wrong. Those are the people whose instincts you want in the room when the rules get written.
Managers are asked to lead something they haven't tried. If the editor has never used the tool on real work, they can't answer a reporter's questions, can't tell good use from bad use, and can't show what sensible looks like. People take their cues from what their managers actually do, not from the policy document.
And when the official route is vague, people find their own
This is the connection to something we've written about before. When the sanctioned route is unclear, slow, or feels like being watched, people quietly use whatever works. That's the story behind shadow IT in the newsroom, and it's the same problem seen from the other end. It's also why a good AI governance policy can sit untouched in a shared drive. A policy tells people what the rules are. It doesn't help them feel able to follow them.
Why it gets missed
Because the visible parts are easy to tick off. Tool purchased. Policy written. Training delivered. Each of those has an owner and a deadline. The change work has neither. Nobody's job title says “make sure the desk feels okay about this,” so it ends up belonging to everyone, which usually means no one.
There's a measurement trap too. Adoption numbers are easy to report, and they can look healthy while people are using the tools badly, or grudgingly, or only because they were told to. A rising usage chart tells you people are logging in. It doesn't tell you whether they trust what they're doing.
What tends to help
In our experience the things that move the needle are fairly human:
Say out loud what AI is for in your newsroom, what it isn't for, and what isn't changing. Give people protected time to learn on a real task, not a demo. Bring the sceptics into the room when the rules are written, because they will find the holes before your readers do. Have leaders use the tools themselves, imperfectly and in the open, before asking anyone else to. And treat the first few months as a listening exercise, asking what's actually getting in the way.
None of that is a secret. Knowing it is easy. Doing it in a particular newsroom, with its own history, its own politics and its own deadlines, is where it gets hard, and it's why the same rollout can work well in one building and fall flat in the next.
The short version
If your AI rollout feels stuck, resist the urge to look for a better tool. Look at what the rollout has asked of people, what it hasn't told them, and who has been given the time and permission to work it out. That's usually where the real problem is.
This follows on from our pieces on Shadow IT in the Newsroom and AI Governance for Newsrooms, and to the conversations on The Signal. If your rollout has stalled and you're not sure why, get in touch.
This article has been AI-assisted, which means the core substance of the work comes from us, and AI has been used to improve, refine, or support it.