AI chatbots give wrong financial advice in 57% of cases, study finds
AI chatbots wrong on financial advice 57% of the time

Households risk losing up to £17,500 after following flawed financial guidance from AI chatbots, according to new research from technology firm Saturn. The study found that mainstream AI models provide inaccurate financial advice 57% of the time, with error rates climbing to 88% when faced with complex questions.

Study puts 18 chatbots to the test

Saturn, a tech research firm, enlisted a paraplanner to run 121 money-related questions through 18 different chatbots, including ChatGPT, Claude, Grok and Gemini. One of the incorrect responses would have cost the user £17,500, the research found.

The findings highlight a growing concern as millions of people turn to generative AI tools for personal finance guidance. Saturn chief executive Amal Jolly warned: "The low quality of financial advice from mainstream AI models risks leading to widespread consumer harm."

Jolly added: "Millions of people are trusting the AI models for money advice, but they are getting wrong answers that can lose them money."

Experts flag regulatory gap

The AI advice sector remains unregulated, described by one expert as a "chink in the armour of the Financial Conduct Authority and the Treasury". George Sweeney, a Financial Conduct Authority-approved financial adviser and deputy editor of investing at Finder, said younger generations are particularly vulnerable.

"Among millennials and Gen Z, a lot of people are hungry for information and they're not being served it and they're turning to generative AI tools," Sweeney said. "It's not tailored to them and then worse than that, it's using years-old data to provide financial advice around, say, investments or pensions, which could be a bit of a disaster."

Protections missing for AI users

TISA's Legrand-Green, a former lead associate at the FCA, pointed to the absence of consumer safeguards. "[If customers] get that exact same information from ChatGPT, there are no protections," Legrand-Green said.

Legrand-Green also criticised the way AI ranks information, saying: "ChatGPT could be ranking some blog on BuzzFeed with the same kind of importance as the latest Financial Reporting Council report from the Bank of England."

Despite the risks, Legrand-Green acknowledged the potential upside: "These tools can be a really good force for change and for helping to upskill people. It's just at the moment it's a bit Wild Wild West."