The Machine Can't Go Ask

The line making the rounds this week was engineered for a newsroom panic:

Readers preferred an AI-written data story over a journalist's story three to one.

That sounds like the beginning of the "journalism is over" argument, which is why it traveled. So I went and found the actual study.

The headline number is real. A seven-agent system out of Oxford and Stanford. Fifty-three reviewers. Thirty-nine preferred the AI version.

But the reason matters. The AI story didn't win because it out-reported the journalists. It won largely because it was easier to verify. The figures were clickable. The claims traced back to the data. Ninety-three percent of the AI story's claims had a source trail, compared with twenty-five percent for the human stories.

That's important. It's also not the same thing as journalism.

The machine was very good at showing its work, and very good at what was already in the dataset. It wasn't good at the part that lived outside the dataset. The study's own example makes the point: give the system data from a network of repair cafés and it can tell you which products break most often. It can't tell you whether the manufacturer designed them that way.

That answer isn't sitting neatly in the data. It's in a room, with a person, after somebody went and asked the question.

That's the part I think people keep missing.

The machine is very good at what is already there. The story is usually somewhere else.

The debate is in the wrong room

Most of the AI-and-journalism debate is still happening in the newsroom. Can the machine write the story? Will it replace the reporter? Is the byline safe?

I understand why that's the argument. But I've spent my career on the other side of the building, and from there the picture looks different. The newsroom may be the most sensitive place this shows up. It isn't the first place local media will feel the full operational impact.

The biggest and fastest changes are going to hit the work around the journalism and around the revenue. The ad desk. Ad operations. Finance. Agency services. Reporting. Planning. The back office. The parts of the business where talented people spend too much of their day building artifacts instead of making decisions.

That's not a small distinction.

I spent more than twelve years running publisher-owned ad agencies. I know exactly how much of that work is judgment, and exactly how much of it is rote: building the first version of a campaign, drafting copy, turning a seller's notes into a media plan, producing the monthly report, writing the recap, building the slide, checking the pacing, creating the first pass at a website, a social calendar, a landing page or a creative concept.

A lot of that work used to take a team and a week. Now a lot of it takes one person and an afternoon. Sometimes less.

That changes the economics of the business. Not in some vague "AI transformation" way — in the very practical way local media actually needs. More work through the same team. More clients served well. Less time rebuilding the same campaign, report, deck, forecast or recommendation from scratch.

That's real margin. And in this business, margin isn't an abstract thing. Margin is whether you can keep the person, cover the meeting, send the reporter, serve the advertiser, and still have enough left to do it again next month.

So yes, take the efficiency. Take it everywhere it's real. Use the machine for the cheap half. But don't confuse the cheap half with the whole job.

The same line runs through every room

The lesson from that study isn't that the machine can replace the journalist. The lesson is that the machine can produce a better-looking artifact when the work is bounded, the data is available, and the output is judged on clarity, completeness and traceability.

That matters. A lot. It also describes an enormous amount of work inside a local media company: the campaign build, the order summary, the first draft of the creative, the reporting narrative, the recap email, the forecast, the list of anomalies, the first pass at the renewal plan, the comparison table, the research brief, the internal memo, the client-ready deck.

For years, we treated too much of that work as if the value lived in the artifact itself. It usually doesn't. The value is in knowing what the artifact should say, whether it's right, whether it fits the client, whether it fits the market, and whether anyone should act on it.

That's the line.

The machine does the production. The person owns the judgment.

The salesperson still has to know what the owner is really worried about. The strategist still has to know whether the campaign fits the business. The finance lead still has to know what the number means. The operator still has to know when the process is producing something clean but wrong.

Because the machine will build a polished campaign for the wrong customer and never know the difference. It will write a sourced explanation from a bad premise. It will produce a beautiful report that answers the question nobody should have asked.

That's where this gets dangerous. Not because the output is bad — because often it will be good enough to stop the next question from being asked.

Slop with footnotes

The buried warning in that ninety-three percent source-trail number is this:

Traceable is not the same thing as true.

A claim can point cleanly back to a source that measures the wrong thing. A forecast can be built from clean data and still miss what's actually happening in the market. A campaign recommendation can be perfectly formatted and completely wrong for the advertiser. A performance report can be accurate and still not tell the client what they need to know.

This is the flood coming for every part of the business, and it isn't going to look like obvious garbage. It's going to be fluent. Formatted well. Full of links and citations and dashboards and audit trails. It will look more credible than the work it replaces.

Slop with footnotes.

That's more dangerous than the obvious kind, because it gives everyone just enough confidence to stop thinking. And thinking is the job. Is this the right source? The right number? The right campaign? Is this what the client actually needs? Is this what happened in the room? Is it true in the way that matters, or just traceable?

The machine can't answer that on its own. It doesn't know your market. It doesn't know the advertiser. It doesn't know which city council fight has been going on for five years. It doesn't know that the restaurant owner isn't really buying ads — she's trying to figure out whether she can afford to open on Mondays again.

That knowledge is the business. And local media still has more of it than almost anyone else in town.

The opportunity is real

That's why the "AI replaces journalism" framing is too small. It misses both the opportunity and the risk.

The opportunity is that local media can finally strip a massive amount of waste out of its operating model. Not just in the newsroom. Across the business. There's work all over these companies that shouldn't be as slow, manual or expensive as it still is — not because the people doing it aren't good. Usually the opposite. Good people have been buried under production work that should have been automated years ago.

AI changes that. It can draft the thing, assemble it, check it, summarize it, compare it, format it, find the gap, show the pattern and build the first version. That's not nothing. For local media, it may be the best efficiency news the business has had in twenty years.

But the benefit only matters if the time comes back to the right place. Back to the client, the market, the reporting, the relationship. Back to the work that requires someone to go ask.

Because the risk is just as real. Owners can take the savings, cut the people, and let the machine put the name of a trusted local institution on work nobody actually stood behind.

That's the fork.

Done right, AI gives local media capacity it badly needs. It makes the cheap half cheaper, faster and more scalable. It lets a smaller team serve more local businesses. It lets a newsroom get more routine work out of the way. It lets operators see what's happening sooner. It gives the business a chance to put scarce human time back where it matters: in the room, on the call, in the market, asking the question.

Done wrong, it becomes another round of extraction. Fewer people. More output. More dashboards. More "content." More work that sounds right and stands behind nothing.

That version isn't a strategy. It's a faster way to burn down the only thing local media still has that platforms can't copy.

Trust.

The machine still can't go ask

I keep coming back to this because I don't think local media gets many more chances to make the right turn. The tools are here. The economics are real. The pressure is real. The temptation to use this only as labor takeout is also real.

But the better use is capacity. Use the machine to handle the production work — draft, assemble, reconcile, summarize, build, check, accelerate. Then hold the line where the line has always been: a person stays accountable for what goes out under your name, and the time you free up goes back into the part of the business the machine can't reach.

That means the reporter who drives over and asks. The seller who knows the account. The strategist who understands the market. The operator who can tell when the workflow is clean but the answer is wrong.

After the last several years in Minnesota, I have a hard time picturing what this business becomes if we lose those people. Not the abstract idea of them. The actual people. The ones who show up, ask, listen, decide, and put their name on the work.

The machine can rank what breaks. It can't find out why.

It can build the campaign. It can't know whether it's the right one.

It can produce the artifact. It can't go ask.

That is the moat. Go staff it.