Analyzing the News: The midterm results

Twelve outlets covered the US midterm results. Fíor read three of them as AI, one of them only just.

Fíor is a natural language classifier that predicts whether a piece of text was more likely written by a person or by AI.

Americans voted in the midterm elections on 3 November. The Associated Press, NBC News and other major news organisations projected that Democrats had won control of both the House of Representatives and the Senate, including Senate races in states that had previously voted Republican. Some races were still being counted as the coverage was written, so the seat totals in different articles reflect different moments.

Coverage pointed to the cost of living as a central issue for voters, alongside the war with Iran and views of President Trump. Democratic leaders said the result would bring more oversight of the administration. Republicans focused on their losses and the challenges ahead. Much of the coverage moved quickly from the result to what a divided government would mean for spending, healthcare, trade and appointments.

The outlets were Reuters, AP, the Guardian, CNN, the BBC, NBC News, ABC News in Australia, NPR, CBS News, Bloomberg, USA Today and Fox News. That is two wire services, American and British broadcasters and newspapers, and ABC News in Australia. NBC News and NPR were scanned as published by local partner stations, under their original bylines. Bloomberg's article sits behind a subscription wall, so the scan covered only its opening and picked up some login text. The Fox News scan included video descriptions and other page text alongside the article.

The Washington Post was left out. Its coverage was a live blog, and in recent stories its scans have picked up an AI-written summary of reader comments. This time it was not confirmed that the scan avoided it.

The result

Nine read as human: Reuters, AP, CNN, the BBC, NBC News, NPR, Bloomberg, USA Today and Fox News.

Three read as AI: ABC News in Australia, CBS News and the Guardian.

The Guardian was a close call. It came down only just on the AI side, Brath marked it not confident, and its other label said it reads like a person wrote it. It should not be read as a firm result either way. ABC News in Australia read as AI, but Brath said it reads like a person wrote it. CBS News read as AI on both labels. None of the pages drew a writing-habit flag.

This is the fifteenth story scanned with Brath, Fíor's new model. CBS News has now read as AI in seven stories running where it ran its own reporting, extending the pattern we looked at in the DoorDash piece.

The ones that read as AI

Two of the three were guides rather than straight news reports. ABC News in Australia explained what Democrats can and cannot do with control of Congress. CBS News went through the results race by race. Guides have read as AI before in this series, in the PHP security flaws story among others.

It does not hold as a rule. The BBC's guide, which promised readers everything they needed to know from the results, read clearly as human.

Why election coverage reads like this

Election-night stories are written at speed, from the same projections and the same statements, and updated as results arrive. The pieces that read as human included both the fast and the considered: AP's results story, CNN's takeaways, NPR's look at divided government and Fox News's analysis of its own voter survey.

The framing differed more than the facts. Several headlines described the result as a rejection of the president. Fox News built its coverage around what voters said in its survey. Reuters and AP led with the results and the reasons voters gave. None of those choices lined up with which articles read as AI.

Some of these pieces may have had AI help, with a first draft, a rewrite or a final polish. Or election coverage has become formulaic enough that a reporter and a model would write it the same way. Fíor cannot tell those apart on its own.

On election night, some races were too close to call. One of Fíor's results this week was too close to call as well. When the margin is that small, the honest answer is to say so.

What Fíor is

Fíor is a machine learning classifier, currently in beta and available as a browser extension. It gives a prediction, not a verdict. It cannot see who sat at the keyboard, and nothing here is a claim about any journalist or publication. Reading as AI and being written by AI are not the same thing.

Treat it as a signal, not proof, and check the pages you read yourself.

Try the browser extension at Fíor.

Analyzing the News is a series where we run one major story through Fíor and see what comes back.