Book reviews for books that don't exist? More proof why journalists must be careful when using AI

Toronto Star

23-05-2025

Chicago-based freelance journalist Marco Buscaglia had a week from hell that began with some nasty emails he received on his cellphone early Monday.
In a nutshell, the blowback he received from readers that morning at 6 a.m. pertained to an article he wrote earlier this year that was published last Sunday, about the top books to read this summer. That article contained major errors — several of the books, 10 of 15 that he listed, don't exist at all. Readers immediately spotted the problems.
It's the kind of calamity that gives many of us in the media shivers.
As it turns out, Buscaglia relied on content generated by artificial intelligence (AI) to write his article. The AI he used, including a tool named Claude, contained bad information.
Buscaglia's erroneous article was featured in a special 'advertorial' section called the Heat Index guide to the best of summer, syndicated by a third party and picked up by two widely read U.S. newspapers, the Chicago Sun-Times and the Philadelphia Inquirer.
It was disastrous for Buscaglia, 56, a veteran in journalism for 33 years. He told me in a telephone interview this week from his base in Chicago that he takes full responsibility for what went wrong here, admitting that he didn't do his due diligence by fact-checking the information he gleaned from AI.
'The fact that I completely dropped the ball on this, (not) checking up on it, makes me feel awful and incredibly embarrassed,' he said.
He later added: 'I didn't do the leg work to follow up and make sure all this stuff was legit.'
While he sat in bed reading those awful emails Monday and pieced together what had happened, he felt like a 'cartoon character' blasted through the stomach by a cannon ball, walking around with a gigantic hole in his stomach for the rest of the day.
'I was devastated,' he said. 'It's been a couple of really bad days here, but I can't say I don't deserve it.'
ARTICLE CONTINUES BELOW
ARTICLE CONTINUES BELOW
Paramount in his thoughts was the notion that what he did fell well below the standards of the Inquirer and Sun-Times.
While AI can be a powerful tool that can assist journalists in some scenarios, as public editor I would state that this unfortunate case demonstrates that AI can also be quite fraught. AI can contain flaws and must be handled with caution by journalists.
When things go wrong, like the books fiasco, it can undermine the media's credibility in a climate where public trust is already shaky.
The special Heat Index section was produced and licensed by a U.S. operation, King Features, which is owned by the large magazine outlet, Hearst. A spokesperson for the company that owns the Sun-Times said in a statement that the content was provided by the third party and not reviewed by the Sun-Times, but these oversight steps will be looked into more carefully for the future and a new AI policy is also being worked on for the Sun-Times.
Buscaglia said he had used AI before writing his book summaries and was familiar with AI 'from a layman's' perspective: he assumed it was akin to a 'glorified search engine.'
It was only after his mishap this week that he delved deeper into how this technology works. He told me that's when he felt 'incredibly naïve' and that he should have known more about AI while using it.
ARTICLE CONTINUES BELOW
ARTICLE CONTINUES BELOW
Generative artificial intelligence relies on large language models (LLMs) to create content, such as images, text and graphics. These LLMs are trained by massive amounts of digital data 'scraped' from the internet.
Flaws with AI come when incidents, sometimes called 'hallucinations' occur. That's where AI simply invents facts. This has even caused significant problems in court here in Canada where, in one example, a lawyer relied on legal cases 'invented' by AI tool ChatGPT.
Full disclosure: the Star uses AI for processes such as tracking traffic to our website, thestar.com. But we have a strict AI policy — internal and in our publicly accessible Torstar Journalistic Standards Guide.
Among the rules stated: human verification of any AI-generated information or content is always required in our newsrooms. In addition, all original journalism must originate and be authored by a human. AI 'must not be used as a primary source for facts or information.'
Since stepping into her role last summer, Nicole MacIntyre, the Star's editor-in-chief has spoken publicly about her concerns around AI and its impact on journalism.
'I said then — and still believe — that we must harness the benefits of this technology cautiously, with public trust always at the forefront.
ARTICLE CONTINUES BELOW
ARTICLE CONTINUES BELOW
'Since then, I've immersed myself in the topic, watching closely as newsrooms around the world experiment with AI. I've seen the risks, including some very public missteps that have shaken reader confidence. But I've also seen what's possible when this technology is used responsibly and with purpose,' MacIntyre told me.
She went on to say the Star's AI guidelines protect our commitment to people-powered journalism.
'With the right guard rails, I'm excited about the possibilities,' she added.