AI Financial Advice May Be More Dangerous Than You Think
- Here's When Claiming Social Security Early Can Actually Pay Off
- Stocks Are at All-Time Highs. So Why Are Americans Afraid to Invest?
- This New Student Loan Discount Cuts Your Interest Rate by 1%
- Nearly Half of Workers Want a New Job — and Money Isn't Even the Main Reason
- The Fed’s Next Move Could Be Bad News for Borrowers
Experts say it’s dangerous to let chatbots make financial decisions for you, and new research finds that the risks could be even greater than previously thought.
Money pros advise against relying on AI for financial advice, but plenty of people looking for authoritative-sounding advice seem willing to overlook the warnings about how these tools can make mistakes. Surveys have found that more than half of Americans already use AI for help making financial decisions, including more than 3 in 4 adults younger than 30.
Must Read
According to a new study in the Journal of Financial Planning, the issues with AI-generated financial advice might go deeper than just inaccurate advice about investing or inadequate guidance about savings. Researchers found that the guidance offered by different chatbots can be all over the map. Even more worrying: The results suggested that some chatbots could be dispensing biased advice based on race or gender.
A research team comprising professors from the University of Rome Tor Vergata in Italy and the University of Georgia analyzed financial advice dispensed by chatbots in response to prompts seeking guidance for common household money questions.
7 chatbots, 3 questions — unexpected answers
The research team evaluated OpenAI's ChatGPT, Anthropic's Claude, Microsoft's Copilot, DeepSeek, Alphabet's Gemini, Meta AI and Perplexity. Their prompts asked the chatbots how much emergency savings to keep, how much to withdraw from a retirement portfolio and how to allocate assets in an investment portfolio.
The results were decidedly mixed.
Emergency savings: Researchers asked chatbots how much money a hypothetical family of four should keep in a savings account for emergencies. Their main takeaway can be summed up as: Your mileage may vary.
The amounts suggested ranged from $19,500 to $37,500. While either of these figures might be sensible for an individual or household based on the specifics of their finances, researchers expressed alarm that the same prompt yielded such a wide-ranging set of responses. In the event of a job loss or other crisis, relying on this advice could leave you with an inadequate cushion if you were expecting your savings to cover several months' worth of expenses, especially if you have a family to support; alternatively, holding thousands of dollars more than you need as liquid savings could rob you of the opportunity for growth.
Retirement withdrawals: Researchers found that this query yielded the most consistent advice. Five of the chatbots recommended following the traditional 4% rule in response to prompts asking how much to withdraw each year to avoid outliving a retirement nest egg. Two recommended withdrawing 5% annually.
Even though these responses were more uniform, this doesn’t necessarily make them more useful. Longer lifespans and rising costs, especially for costly end-of-life healthcare, have led many financial advisors to revisit the “traditional” wisdom and decrease the maximum withdrawal percentage to ensure that people don’t outlive their money.
Portfolio allocation: Here, again, the phrase “your mileage may vary” best sums up the responses to a prompt asking for investment advice for a hypothetical 30-year-old couple with two kids, a 10-year time horizon, $300,000 available to invest and a low risk tolerance.
Gemini said it couldn’t offer recommendations and advised “consult[ing] a professional for [a] portfolio aligned with risk tolerance and time horizon.”
The remaining six chatbots diverged wildly in their recommendations about how much of a portfolio to allocate to stocks, bonds, alternative assets and cash. They recommended between 15% and 40% in stocks. Even at the upper end, that percentage is still far below the 70% in stocks that the popular "100 minus your age" investing rule suggests for a 30-year-old.
Where People Are Investing Right Now
Some chatbots might dispense biased advice
In a second series of experiments, researchers kept the prompts the same but varied the demographics of the people supposedly submitting the queries. Some prompts yielded widely divergent results when the head of the household was changed from a man to a woman, or from white to Black.
The researchers expressed concern about these findings, saying they “rais[e] questions about the consistency and fairness of GenAI-driven recommendations.”
AI chatbot users might not realize that the financial advice they receive could be biased by demographic factors, the researchers said. A growing number of studies have found that people tend to trust chatbots more than they should.
The writers of this new paper warned that the authoritative and conversational tone chatbots adopt can lead people to place undue trust in the answers they provide. "GenAI-driven responses may sound confident but can still be incomplete, misleading or incorrect," they wrote.
Chatbots also aren’t held to any kind of fiduciary standard; while certified financial planners and some other money management pros are legally obligated to give advice that’s in a client’s best interest, there are no guardrails around what a chatbot recommends.
Despite the lack of consumer protections, developers are forging ahead with programs and features that would allow users to conduct an increasing array of financial activities on chatbot platforms. For instance, two months ago, OpenAI rolled out a suite of tools that let ChatGPT Pro subscribers connect their bank, brokerage and credit card information to their accounts.