If you ask an AI chatbot for advice on how to revive your wilted tomato plant or for a movie recommendation, even a bad answer has pretty low stakes.

Not true when your money is on the line.

Still, some Americans are giving it a try. The data analytics company JD Power asked 4,000 people about their financial health and found that 40% of them had turned to AI in the prior three months to help manage their finances. More than a third of those using AI said they found the advice helpful in making smarter financial decisions, comparable to the share of those who found their bank's advice helpful.

David Kendrick, a 53-year-old IT manager in Dayton, Ohio, uses ChatGPT so often for advice that he has started referring to it as "Chatty." He has asked it about everything from what to do with his home equity line of credit to how to handle his recent salary bump. Should he put more of that income toward paying off debt or direct it to his Roth IRA?

Chatty suggested the Roth. Kendrick listened.

He does see a human adviser once a year, but Kendrick said having access to advice anytime, as well as being reassured that his financial plans look OK, calms his financial anxiety.

"I've always kind of had that, because my parents struggled," Kendrick said. "And this very much helped."

Overall, Kendrick said, his finances are solid. But that's not necessarily true for the biggest group of people using AI for financial advice, according to the JD Power survey. The data firm labeled those people as "overextended," meaning they might be a bit over their budgets and have some debt.

Those users reported asking chatbots questions about things like how to stretch their dollars, and they got back advice along the lines of skipping the name-brand cereal for the store version.

Taha Choukhmane, an associate professor at the MIT Sloan School of Management, recently co-authored a paper — not yet published — that found that AI does well at giving fairly broad economic advice. "It tends to push people toward saving more, participating more in the stock market, de-risking as they get older," Choukhmane said. "It gets a lot of things right."

In that study, the researchers got 1,000 adults to write prompts asking an AI model for advice. The team then simulated the lifetime effects of what would have happened if people had followed that advice, compared with what would have happened if they hadn't. Overall, they found, following the advice would have helped them end up with more savings.

But the AI didn't always give advice that lined up with what human financial experts would suggest — especially when dealing with more complicated requests, like how to handle a job loss. In that case, the AI suggested spending cuts that the researchers considered too harsh, and it didn't advise the user to dip into savings meant for tough financial moments.

The researchers also felt it didn't give good advice on how to rebalance a portfolio. And they concluded that the AI suggested riskier financial moves for men than for women.

Without enough information, AI can be confidently wrong

Overall, finance experts think AI advice seems to work best for people whose questions fall on the two ends of the sophistication spectrum: for people asking very basic Finance 101 questions, and for power users who are prepared to hand over lots of data about their situation and craft detailed prompts.

But, of course, people ask a lot of questions that are somewhere in between very basic and very detailed — and in those cases, AI can make mistakes.

Danielle Harrison, founder of Harrison Financial Planning in Columbia, Mo., decided to test an AI model. She told it about her husband joining her firm and asked how they should structure their business.

"It gave an answer that it was sure about. It was like, 'You need to be an S corporation,'" she recalled. That's a type of business with some tax advantages.

But Harrison kept prodding the AI and giving it more info. "It completely changed tunes by the end of it," she said, instead telling her that she needed to form an LLC, another type of company. "If I had not had that background knowledge, it would have given me the wrong information," Harrison said.

AI models can make up sources — a phenomenon called "hallucination" — or make incorrect assumptions about someone's situation.

Sharon Bloodworth, the CEO of White Oaks Wealth Advisors, which has offices in Minneapolis and Sarasota, Fla., said that in her experience, AI is wrong more than it's right.

But she also expects AI to only get better in the future and said that it presents an opportunity to offer financial planning advice to people who don't have access to human advisers. "Ignoring it would be almost like saying, 'Don't pick up a calculator' or 'Don't get into a car, and just still ride a horse,'" she said.

For all his enthusiasm for AI-driven financial advice, Kendrick, the IT manager in Ohio, is still careful about his interactions with Chatty. He doesn't give the AI model direct access to his financial accounts, just copies. And he knows not to take its advice as gospel, because AI models can be sycophantic, telling users what they want to hear.

Kendrick said whenever the AI starts calling all his ideas great, he sets it on track. "I'm like, 'Hey, quit that. You got to be real with me,'" Kendrick said.

Tagged: financial technology

AI chatbots are offering financial advice. Should you trust them?