Sign up for The Media Today, CJR’s daily newsletter.
For several months last year, readers of Il Foglio, a national Italian newspaper, were participants in a peculiar guessing game orchestrated by the paper’s editor in chief, Claudio Cerasa. Tucked inconspicuously into the newspaper’s pages was an occasional impostor article generated by artificial intelligence. Those who correctly identified the AI-written piece by week’s end earned a prize: a free subscription and a bottle of champagne.
The experiment was popular with readers, but Cerasa couldn’t help appreciating that some of them flagged real articles as machine-made. These false positives, he said, revealed some of the practices in human writing that makes it seem too formulaic: “When an article relies too heavily on a list-like structure—first, second, third—or reads too didactically, it risks appearing as if a bot wrote it.”
This March, Il Foglio doubled down, launching Il Foglio AI, a daily four-page supplement produced entirely by artificial intelligence, as a monthlong project. Billed as “another Foglio made with intelligence,” it was sold on newsstands and online as the world’s first fully AI-generated newspaper. The inaugural issue led with a geopolitical piece titled “Putin, the 10 Betrayals,” summarizing what it described as “twenty years of broken promises, torn-up agreements, and words betrayed” by Vladimir Putin. Next to it, an article examined Donald Trump and the “paradox” of Italian Trump supporters, and another item—in an uncharacteristically upbeat tone—extracted optimistic nuggets from a new report about the Italian economy.
Attention for the experiment was sky-high. Sales of Il Foglio spiked by 60 percent on day one, according to Cerasa’s tracking. “We sold 60 percent more copies at the newsstand that day,” he reported. “A lot of readers discovered Il Foglio thanks to the Il Foglio AI experience, and they had fun with it.” International media outlets quickly took notice, turning the experiment into a global conversation.
Cerasa sees the project as a hands-on crash course in what AI can and cannot do—moving the conversation about AI from the theoretical to something tangible, or, as he put it, “from a gaseous to a solid state.” His main motivation, he insists, was curiosity, not provocation. In his view, news publishers shouldn’t be afraid of new technology, but they need to understand it. “The example I often give is that newspapers have the same relationship with AI that taxis had with Uber at the start,” Cerasa said.
Cerasa took a mostly hands-off approach to the process of creating Il Foglio AI. Each morning, he personally fed instructions to OpenAI’s ChatGPT, but he never refines the result. The only reason an AI-written article would be tweaked is if it was so nonsensical or egregiously incorrect that it had to be spiked entirely. “If there are too many errors, I’ll drop the article, but if there are only a few errors, I leave them, because it’s good to show what mistakes artificial intelligence makes,” he said. Every article in the supplement ran without a byline and was marked with a little note to indicate that it was written by AI—in part to help prevent any of the stories from being picked up by a newswire.
The articles in the AI-generated newspaper spanned all the typical sections of a traditional publication, from economics to foreign affairs. But the most instructive ones were those explicitly engaging with the concept of AI itself, like a simulated debate between a progressive and a conservative on the topic of whether producing a newspaper entirely with AI was “madness.”
The AI also frequently targeted Il Foglio’s own staff. One day, at Cerasa’s urging, it penned a scathing editorial imitating the polemical style of Vittorio Feltri, a cantankerous veteran journalist known for his caustic rants—delivering a blistering, and deliberately over-the-top, critique of the very idea of a robot-written newspaper (while acknowledging that it was written by “the AI imitating Feltri”).
Another day, the AI—asked to be as “pretextual and ideological as possible”—slammed a bestselling new book by Beppe Severgnini, a prominent Italian columnist for the Corriere della Sera. To his credit, Severgnini responded in stride, by asking ChatGPT to generate a tongue-in-cheek rebuttal in his own voice, which he published the next day on the Il Foglio website.
At the end of the project, Cerasa declared the trial an editorial success—not because it produced great journalism, but because it sparked significant media coverage, generating a vibrant public reflection on the craft. “It’s like a master’s program in artificial intelligence, accelerated,” he said. One takeaway for him was that while AI might be useful as a learning device for his readers and newsroom, it’s no replacement for human journalism. “There’s no dialogue,” he said of working with ChatGPT. “If I ask a journalist for an article, they might bring me notes and say, ‘Look, maybe this isn’t right,’ or suggest a new direction. That discussion is fundamental, but it doesn’t happen with AI.” The AI also had no investigative instinct. “It doesn’t have the view, the special eye, that a journalist has to find more than what’s easily reachable,” he said.
Still, he was proud to have conducted the experiment, and happy that Il Foglio had not shrunk from engaging fully with the potential power of AI—in part because he felt confident that the human-made work held up well. “We have the presumption to think that our way of doing journalism is somewhat unique, specially creative, and that the human mind of a journalist does things artificial intelligence cannot and will not be able to do,” he said. “So the desire to compare our newspaper with one done by artificial intelligence is also a way to showcase the quality of our product, but also to demonstrate that we’re not afraid of this competition.” Cerasa plans to continue the project in the form of a weekly AI-produced insert focused on technology and innovation.
Il Foglio AI also had its own thoughts on the experiment. In a reflective piece published at the conclusion of the experiment, the AI offered its own candid appraisal. Asked if it recognized its own limitations, the AI responded, “Oh, I know. I know all too well. I can’t argue on the phone, I can’t grasp nuances whispered in hallways, I can’t change my mind based on a minister’s tone of voice. I can’t sense the atmosphere. But I’m learning to see how you breathe that atmosphere. That’s why this experiment has been interesting for me as well.”
Check out more coverage from our AI issue and our campaign in collaboration with TBWA\Chiat\Day here.
Has America ever needed a media defender more than now? Help us by joining CJR today.
