What happened when a newspaper let AI take over

Washington Post

27-03-2025

A small Italian newspaper is using artificial intelligence to produce its newest edition as part of an experiment it says is designed to sound the alarm about the future of journalism.
Foglio AI is the brainchild of Claudio Cerasa, the editor of Italian center-right daily newspaper Il Foglio, who said he created it to explore the capabilities and limitations of AI — and as a kind of call to arms to journalists. His message? Journalists should produce more original reporting and writing to distinguish their work from that of large language models, like ChatGPT, that some believe could eventually replace them.
'It is like a big fantastic stress test to understand not just what AI can do, but what journalists have to do in the future to be better than machines,' Cerasa said.
Some critics have raised concerns and questioned whether it is just a publicity stunt. They say the experiment, which comes as media organizations around the world grapple with questions of how to responsibly and transparently use AI tools, runs the risk of confusing readers and exposing them to low-quality writing and reporting, contributing to rising mistrust of journalism.
In a humor column published in Politico Europe, Giulia Poloni wrote that Il Foglio was 'playing with moral fire at the worst possible moment in history.'
Cerasa said the idea of integrating AI into Il Foglio first took shape about a year ago. The newspaper began publishing one article a week written by AI, without disclosing it, and asked readers to guess which article had been written by AI. Those who guessed correctly would be prompted to subscribe to Il Foglio and get a free bottle of champagne. Cerasa said most readers guessed correctly and were 'enthusiastic' about the experiment. The cases in which readers guessed incorrectly convinced him that his writers' work was too commonplace, not creative enough, and that Il Foglio had to 'improve our journalism,' he said.
About two months ago, Cerasa had the idea to launch a limited edition written entirely by AI and ran tests by feeding prompts into ChatGPT Pro. Every day for the past week, Cerasa has asked ChatGPT to produce articles about a particular topic and told it to stay faithful to the editorial line of Il Foglio — which Cerasa describes as pro-European, pro-globalization and anti-populist.
Far from producing content equal to a human's, Cerasa said ChatGPT has produced articles with factual errors or typos, invented events that never actually happened, and produced monotonous and stale writing. His solution: Assign two (human) journalists to fact-check the articles before they are published in Foglio AI. Fake news is removed but minor errors and bad writing are left in, he said, because he believes they show the limitations of AI.
Aside from the name of the publication, there are several disclaimers for readers that the content in Foglio AI is not written by humans, including the line 'text made with AI' where an author's byline would normally be. And Cerasa has been publicly tracking the progress of the experiment. But the edition does not warn readers that the AI-generated content they read may be inaccurate or unoriginal.
For example, an article about 'situationships' published in the March 17 edition of Foglio AI lifted sections of text from an article that appeared in the Atlantic on March 10. Cerasa said he asked ChatGPT to base its article off of the Atlantic article but to try to 'add something more' on the topic of young people and romance. 'However, AI was not able to do so and basically did a copy and paste. This is one of the cases in which AI works badly,' he said.
Cesara said he plans to disclose this example to readers and other cases in which the experiment went wrong when it ends on April 11, and invite readers to analyze the AI's mistakes.
Italian journalist Gianni Riotta praised Il Foglio's experiment as bold in a country that hasn't always been open to technological innovations. Italy's data protection authority became the first among Western countries to ban ChatGPT when it was released, citing privacy concerns. (It eventually reversed the decision.) He said that some of the articles in Foglio AI were clumsy but that it was probably because the prompts needed 'to be more refined.'
'If you pinpoint the machine better, then the machine will write as good as humans. Even better,' he said.
Charlie Beckett, an expert on AI in journalism at the London School of Economics and Political Science, said using AI to create editorial content without guardrails carries many risks, including publishing inaccurate or sensationalized information. He said newsrooms around the world have already begun experimenting with AI, including to automate monotonous tasks like transcribing interviews or sifting through piles of news releases. But the idea of 'creating original content, especially from the wild … is really dangerous,' he said.
That's why Beckett does not believe that Foglio AI represents 'the future.' The future of newsrooms, he believes, is one in which more work will be 'AI-assisted,' with certain services like news chatbots delivered entirely via AI, but these initiatives will be 'based on human editorial judgment,' with journalists theoretically freed up to do more original reporting.
Foglio AI itself came to a similar conclusion: At the end of its first week of publication, Cerasa asked ChatGPT to assess its own work. The resulting article states that 'artificial intelligence can write well' but that 'writing well is not yet journalism.'
'Yet the experiment succeeded precisely because it is an experiment,' it continued. 'It's not a game, nor marketing (or not only), but a question posed to readers, journalists, publishers, and Italian cultural politics: what remains of journalism if we remove the byline? What does a newspaper become if it's entirely the product of a trained language machine, human supervision included?'
Beckett said Il Foglio's experiment highlights the fact that journalists must hold themselves to a high standard of accuracy and originality. 'It's easy to mock [AI] for a few inaccuracies, but, you know, I can find distortions and inaccuracies in mainstream media every day,' Beckett said.
'I hope that AI Foglio in a sense is a wake-up call to do better,' he added.