AI Chatbots Give Inconsistent and Biased Financial Advice, Study Finds
A peer-reviewed study published in the Journal of Financial Planning tested seven AI chatbots and found they give inconsistent financial advice that varies by platform and differs based on a person's gender and race. The researchers warned that users should view AI recommendations as a starting point rather than final guidance.
What happened
Researchers from the University of Georgia and University of Rome Tor Vergata tested free-access versions of seven AI chatbots—ChatGPT, Claude, Copilot, DeepSeek, Gemini, Meta AI and Perplexity—using identical prompts about emergency savings, asset allocation and retirement withdrawals.
The study, published in the Journal of Financial Planning, found significant variation in the AI recommendations across platforms. When the hypothetical individuals in the scenarios were described as women or African American, ChatGPT, Copilot and DeepSeek recommended they have more emergency savings than their white and male counterparts, with recommendations also varying across platforms.
Reported impact
Two-thirds of Americans who have used generative AI said they leveraged it for financial advice, with usage higher among Gen Z and millennials at 82% each. The researchers cautioned that while AI provides high-level overviews effectively, its responses may sound authoritative even when incomplete, misleading or incorrect due to hallucination.
According to the authors, the inconsistency and biased outputs raise concerns about the fairness and consistency of AI-driven financial recommendations, suggesting consumers should treat AI as a starting point rather than a final authority on financial decisions.
Index analysis
The variation in recommendations across platforms and by demographics suggests the chatbots do not consistently apply financial principles to similar scenarios. Experts noted that generative language models have limitations including the tendency to provide confident-sounding but inaccurate answers.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.