A new study in the Journal of Financial Planning found that AI chatbots give wildly inconsistent financial advice, and some responses shift based on the race or gender of the person asking the question. With surveys showing more than half of Americans already rely on AI for money decisions-and that number jumps to over 75% for adults under 30-the findings expose a risk that goes beyond simple inaccuracy.
A research team from the University of Rome Tor Vergata and the University of Georgia tested seven chatbots-OpenAI's ChatGPT, Anthropic's Claude, Microsoft's Copilot, DeepSeek, Alphabet's Gemini, Meta AI, and Perplexity-using three common household finance prompts. The answers ranged so widely that the researchers warned users could be left with dangerously inadequate savings or miss years of portfolio growth.
Emergency savings advice varied by $18,000
When asked how much a family of four should keep in an emergency fund, chatbots suggested amounts from $19,500 to $37,500. Either figure might make sense for a specific household, but the researchers said the same prompt producing a gap this large is alarming. Relying on the low end could leave a family short during a job loss; accepting the high end could mean holding too much cash and forgoing investment returns.
Retirement withdrawals: more uniform, but outdated
Five of the seven chatbots recommended the traditional 4% annual withdrawal rule for retirement portfolios. Two suggested 5%. The consistency is less reassuring than it sounds. Many financial advisors now recommend lowering that percentage because longer lifespans and rising healthcare costs increase the risk of outliving one's savings.
Portfolio allocation ranged from 15% to 40% in stocks
For a hypothetical 30-year-old couple with two kids, a 10-year horizon, $300,000 to invest, and low risk tolerance, Gemini declined to answer and advised consulting a professional. The other six chatbots gave stock allocations between 15% and 40%. Even the top end sits far below the 70% that the popular "100 minus your age" rule would suggest for a 30-year-old. The researchers described the divergence as "decidedly mixed."
Demographic bias emerged in chatbot responses
In a second round of testing, the researchers kept the prompts identical but changed whether the head of household was described as a man or a woman, and as white or Black. Some responses shifted significantly. "GenAI-driven responses may sound confident but can still be incomplete, misleading or incorrect," the team wrote, adding that the findings "rais[e] questions about the consistency and fairness of GenAI-driven recommendations."
Chatbots carry no fiduciary duty. Certified financial planners must act in a client's best interest, but no such guardrails apply to AI tools. Despite that, developers are pushing deeper into personal finance. Two months before the study's publication, OpenAI rolled out tools letting ChatGPT Pro subscribers link bank, brokerage, and credit card accounts directly to the platform.
For finance professionals who need to understand the limits and risks of these tools, resources like AI for Finance provide structured guidance. The study's authors warned that the conversational tone of chatbots can lull users into overtrusting the answers, a dynamic that makes expert oversight critical.
Why this matters for finance professionals
Clients are already using AI for financial decisions, often without disclosing it. The wide variation in advice-and the emergence of demographic bias-means advisors must assume that AI-generated recommendations are entering client conversations. Proactively asking clients about their use of chatbots and stress-testing those outputs against fiduciary standards will become a necessary part of the planning process. The risk isn't just a bad stock pick; it's a systematic gap in protection that no regulator has yet addressed.
Your membership also unlocks:
