I’ve spent the last six months talking to editors, reporters, and newsroom leaders about their AI experiments. The hype cycle is calming down, and we’re finally getting honest answers about what works and what doesn’t.

Here’s the thing nobody wants to admit publicly: most newsroom AI projects have been expensive disappointments. But some have genuinely transformed how journalists work. The difference? It’s not the technology. It’s almost never the technology.

The Three Categories of Newsroom AI Use

After dozens of conversations, I’ve noticed AI implementations fall into three buckets.

Transcription and summarization is the clear winner. Every editor I spoke with said their teams now use AI transcription as standard practice. The time savings are real—one sports desk told me they’ve cut their post-match turnaround by nearly 40 minutes because they’re not manually transcribing press conferences anymore.

The BBC has been particularly open about this. Their journalists are using AI to transcribe interviews, summarize lengthy documents, and pull key quotes from council meeting recordings. Nobody’s losing their jobs over it. Journalists are just spending less time on tedious tasks.

Automated content generation is where things get dicey. Yes, the Associated Press has been using AI for earnings reports and sports results for years. But that’s templated, data-driven content with clear right-and-wrong answers. When publishers try to expand this into more nuanced reporting, the results range from embarrassing to lawsuit-worthy.

CNET learned this the hard way. Their AI-generated articles needed so many corrections that the reputational damage likely outweighed any cost savings. I’ve heard similar stories from smaller outlets who tried—and quietly abandoned—similar programs.

Research and discovery tools sit somewhere in the middle. Some newsrooms are using AI to surface connections in large document dumps or identify patterns in datasets. It’s genuinely useful for investigative work. But it requires journalists who understand both the tools and their limitations.

What the Successful Experiments Have in Common

The newsrooms seeing real results share a few characteristics.

They started small. Really small. One reporter. One specific problem. Then they expanded based on actual results, not theoretical possibilities.

They involved journalists from day one. The implementations that crashed and burned? Usually pushed from management without meaningful input from the people who’d actually use the tools. Turns out, journalists have strong opinions about their workflows. Who knew.

They’re brutally honest about quality. The Agence France-Presse approach is instructive—they’ve been testing AI tools quietly, measuring accuracy obsessively, and only rolling things out when the error rate drops below their human baseline. No press releases. No industry panels. Just work.

And critically, they haven’t cut staff. The newsrooms treating AI as a cost-cutting measure are getting worse results than those treating it as a capability enhancement. There’s probably a lesson there about human nature and motivation.

The Real Costs Nobody Talks About

Every vendor will tell you about the productivity gains. Nobody mentions the hidden costs.

Training time is substantial. Even “easy” tools need weeks of practice before journalists use them effectively. That’s weeks of reduced output while people climb the learning curve.

Quality control overhead actually increases initially. You need editors reviewing AI-assisted content more carefully, not less, especially in the early months. One outlet I spoke with estimated their editing time went up 30% for the first quarter.

There’s also the trust problem. Readers don’t fully trust AI content yet, and they shouldn’t. Any outlet that hides AI involvement is taking a reputational risk. Disclosure requirements are coming—better to get ahead of them.

What I’d Tell an Editor Starting Out

If you’re running a newsroom and haven’t started with AI yet, you’re actually in a decent position. You can learn from everyone else’s mistakes.

Start with transcription. It’s the lowest risk, highest reward application. Pick a tool, train your team, measure the time savings. This isn’t revolutionary, but it’s genuinely useful.

Avoid automated content generation for anything requiring judgment. I don’t care what the salesperson told you. Until AI can understand context, nuance, and the specific needs of your audience, keep humans in the loop for anything beyond basic data reporting.

Invest in AI literacy for your entire team. Not everyone needs to be an expert, but everyone should understand what these tools can and can’t do. Some newsrooms bring in specialists—team400.ai or similar firms—for initial training. The journalists who understand AI’s limitations use it more effectively than those who either fear it or trust it blindly.

And for the love of journalism, don’t announce anything until you’ve actually proven it works. The graveyard of hyped AI journalism initiatives is full enough.

Looking Ahead

The next 18 months will be interesting. I’m watching a few developments closely.

Local news might be where AI has the biggest positive impact. Small outlets with tiny staffs could use AI to cover municipal meetings, process public records requests, and handle the grunt work that currently goes undone. It won’t replace reporters, but it might make local journalism economically viable again in some markets.

The regulatory environment is shifting. The EU’s AI Act has implications for media, and Australia is working on its own framework. Smart newsrooms are building compliance into their AI strategies now.

And the tools themselves are improving rapidly. What failed a year ago might work today. The newsrooms that maintain their skills while staying skeptical will be best positioned to take advantage.

I’ll be tracking these developments closely. The media industry’s relationship with AI is still being written, and the next chapter promises to be more interesting than the last.