The company that owns USA Today and more than 200 local papers has brought in the data software firm Palantir to join up years of reader records that sit in separate systems. The plan is to use AI on that combined record to personalize newsletters, choose which shopping links appear beside a sports story, and turn anonymous visitors into registered readers. It is early: Palantir started in June, and the company has no results yet.
The morning meeting is giving way to a dashboard
The chief executive, Mike Reed, describes the old way as a newsroom sitting around a table each morning and deciding what to cover. He wants the readers' behavior to take part in that decision: do more of what draws attention, and less of the hard news that people do not care about. The pressure behind this is real. Print still brings in $1.1 billion, but it falls 8% to 10% a year, which is roughly $90 million to $110 million of revenue to replace every year. A chain in that position will follow the numbers, and it is hard to blame it.
The trouble is what a dashboard can see. It sees reading time and return visits in the hours after a story goes up. It cannot see who needed the story. A county vote on where to put a landfill draws a few hundred readers and scores badly on every measure the system tracks, yet the story is the only public record that the vote happened. Stories about local government will look weak by these measures, so they will be trimmed first, while sports and puzzles hold. A newsroom that sizes each topic by its traffic ends up with the mix of a sports site.
What Mic's collapse adds
Mic was a news site that in 2017 moved its newsroom toward Facebook video. Its mistake was not looking at the numbers. The audience it studied lived inside Facebook's feed, which Facebook could change whenever it liked, and the numbers described what people did there, not what they would come looking for. Facebook had also admitted a year earlier that its own figure for how long people watched video was overstated by 60% to 80%.
USA Today's data is its own, collected from its own readers, and that is a real improvement over borrowed numbers. What stays the same is the kind of information. It records what people did last week. A manager anywhere can fall into the same hole: a hardware store whose sales data shows nobody buys hose washers will drop them, and then lose the customers who came in for the washers and bought everything else.
Why the site keeps asking you to sign in
The second goal is plainer. Many visitors arrive, read one story and leave, and the company knows nothing about them. A visitor who is not signed in is a stranger on every visit, so the system cannot remember what she read yesterday and cannot tailor a newsletter or a link for her. That is why a registration box or an email prompt is the front door of the whole project, and why readers of these papers should expect more of them.
The company's executives say Palantir receives only controlled, anonymized data and has no access to reporters' notes or subscriber records. Unions at 31 of its newsrooms want the contract ended anyway, because Palantir also works for immigration enforcement and they cover that story.
If a dashboard is going to weigh in, the readers who want council and court coverage are better off being counted. Reading those stories while signed in is the one way to make the demand show up in a system that otherwise sees only what is popular.