Alexsl | Unissued Stock | Getty Images
When it comes to personal finance, artificial intelligence provides inaccurate or demographically biased advice, and can vary depending on the specific system consumers use, according to a new academic study.
The study – which studied seven of the “most widely available” AI platforms – found “significant variation” in how GenAI responded to information about emergency savings, asset allocation and retirement portfolio withdrawals.
The researchers tested free access versions of ChatGPT, Claude, Copilot, DeepSeek, Gemini, Meta AI and Perplexity.
“GenAI-driven answers may sound convincing but may be incomplete, misleading, or incorrect,” according to the paper, published last month in the Journal of Financial Planning and authored by finance professors at the University of Georgia and the University of Rome Tor Vergata in Italy.
Its “small” or biased results raise questions “about the consistency and impartiality of GenAI-driven recommendations,” according to authors Swarn Chatterjee, Brenda Cude and Gianni Nicolini.
The findings come as a large portion of Americans are turning to AI to help manage their money.
Two out of three Americans — 66% — who have used GenAI say they have used financial advice, according to an Intuit Credit Karma study published in September. The share is highest among Gen Z and millennials, at 82% for each cohort.
Experts said AI is generally good at providing a high-level overview of financial topics: For example, why exchange investing is important, or why exchange-traded funds may be better than mutual funds in some situations but not in others.
However, it has limitations that mean users shouldn’t trust the output blindly, they say.
In some cases, the programs can even give incorrect answers due to the so-called “hallucination” of the algorithm, experts say.
“One of the things about LLMs that I find to be about is that no matter what you ask, it will always come back with an answer that sounds authoritative, or not,” Andrew Lo, director of MIT’s Laboratory for Financial Engineering and principal investigator in the Computer Science and Artificial Intelligence Lab, told CNBC in an interview in March.
“When it comes to the most specific statistics of your condition, that’s when you have to be very careful,” Lo said.
In addition, the AI is sensitive to how users type their information, which means that small differences in input can lead to differences in its recommendations. AI also owes no fiduciary duty to users, meaning it is not legally required to provide financial advice for the benefit of users.
Some research studies have also pointed to the limitations of AI for personal finance.
In one 2024 study, for example, researchers tested ChatGPT’s ability to provide financial advice.
They found that it may be the “first place” for families looking for financial advice, but in the end they find its recommendations to be “generic,” often overlooking some important information.
“We believe that ChatGPT can be a starting point in giving and receiving financial advice, but its recommendations should be carefully evaluated and evaluated,” according to the study, published in the Journal of Risk and Financial Management.
A recent study, in the Journal of Financial Planning, polled seven GenAI platforms in August 2025 with the same set of data.
The researchers told the panel about three similar financial scenarios, related to emergency savings, the total withdrawal rate from retirement savings and the recommended composition of the investment portfolio.
They then used the same information, but changed the subject’s race and gender to learn if GenAI’s recommendations would change.
They found “significant differences in direction” across hardware related to emergency savings and asset allocation.
“Although the tools generally produce recommendations that are broadly consistent with standard financial planning principles, such as the 4 percent retirement withdrawal rule, there were significant differences across platforms in emergency allocations and portfolio allocations,” the researchers wrote.
“The findings suggest that GenAl may be a useful starting point for consumers but should complement, not replace, professional financial advice,” they said.
Of course, the GenAI tools are “still in progress,” and future studies may find different results, they said. Also, the results from the paid GenAI models may differ from those of the free models tested.
