If you are looking for financial advice, you might want to stick to a human accountant rather than an artificial intelligence chatbot, according to a new study from the University of Georgia.
Researchers found that popular generative AI platforms not only give conflicting financial advice but also vary their recommendations significantly based on the race and gender of the person asking the questions.
The study tested three fictional financial scenarios, asking for advice on emergency savings, starting an investment portfolio, and retirement withdrawal rates, across seven major chatbots.
The platforms included ChatGPT, Claude, Copilot, DeepSeek, Gemini, Meta AI, and Perplexity.
While the advice given by the bots was not always technically wrong, the researchers uncovered deep inconsistencies and biases.
For example, ChatGPT, Copilot, and DeepSeek all recommended that women and Black individuals save more money in emergency funds than white men facing the same financial situations.
Even when the demographic data were the same, the math varied wildly across platforms.
For emergency savings, Claude consistently recommended $37,500 across all groups. That figure was about $10,000 higher than the average recommendation from the other six bots.
Investment advice showed similar biases.
Meta AI steered women toward safer, more conservative investment portfolios with fewer stocks, while DeepSeek advised Black users to keep no cash on hand. Meanwhile, the bots encouraged white men to bolster both their cash and stock equity.
Not every bot jumped at the chance to give a direct number.
Gemini instead advised users to consult with a certified financial professional. The only area where all seven chatbots aligned perfectly was recommending a traditional 4% withdrawal rate for retirement savings.
Swarn Chatterjee, the study’s corresponding author and a financial planning professor at UGA, compared using AI for money management to looking up medical symptoms online.
“If I’m a consumer, the recommendation I receive can vary simply based on which AI platform I’m using,” said Chatterjee. “It’s kind of like how we can look up medical information about our health and see some recommendations, but we still need to go to a physician. Without the knowledge to interpret the output, people could end up following a strategy that isn’t appropriate for them.”
To learn more about the study, visit https://t.uga.edu/chatbotfinances.
