Stanford research finds AI financial advice quality varies by user

Research from the Stanford Graduate School of Business indicates that while artificial intelligence chatbots can encourage sounder financial decisions, such as increased diversification and larger savings buffers, the quality of advice is highly dependent on user input.

A simulation involving 1,000 virtual individuals following advice from general-purpose large language models (LLMs) revealed significant disparities based on financial literacy and gender. Prompts simulating low financial literacy resulted in advice that left subjects nearly $50,000 poorer by age 60.

Gender-based differences in prompting also impacted outcomes. Prompts from women tended to focus on terms like “family,” “grocery,” “credit,” and “loan,” whereas men’s prompts focused on “portfolio,” “equity,” “strategy,” and “crypto.” This discrepancy resulted in advice that left women nearly $60,000 poorer by retirement.

Financial advisors report observing these trends in real time, noting that clients often lack the knowledge to provide the specific, relevant facts necessary for an AI to produce a complete and accurate financial analysis.