A new MIT Sloan study finds that AI financial advice is “surprisingly good” — boosting savings rates, encouraging diversified portfolios, and guiding age-appropriate risk reduction — but the quality of advice users receive depends heavily on how well they phrase their questions. Researchers tested GPT-5.2, GPT-5.6, and Gemini 3 Flash with 1,000 adults, measuring real-world wealth outcomes over a simulated lifetime. The results reinforce why the emerging field of AI financial advisors is drawing serious attention from both consumers and regulators alike.
What the Research Found
The paper, “AI Financial Advice: Supply, Demand, and Life Cycle Implications,” was co-authored by Taha Choukhmane, Weidong Lin, and Matthew Akuzawa of MIT Sloan and Tim de Silva of Stanford’s Graduate School of Business. It won the Swiss Finance Institute Outstanding Paper Award 2026. According to MIT Sloan Ideas Made to Matter, AI models performed well on core planning tasks — encouraging stock market participation, diversification, and reducing equity exposure after age 45 — but struggled to adjust recommendations during economic shocks such as unemployment, and rarely recommended proactive portfolio rebalancing.
The Prompting Gap Costs Real Money
The most striking finding is how much prompting style affects outcomes. Users who provided structured, detailed prompts received significantly better advice than those who asked casually. Women and less financially literate users accumulated approximately $50,000 (4%) less wealth by age 60 compared to their peers — two-thirds of that gap traced back to differences in how they phrased their questions, not model bias. Users with no prior AI experience ended up with nearly $100,000 (6%) less at retirement. The implication is clear: AI advice tools are powerful, but they amplify existing inequalities in financial literacy and technology fluency unless platforms actively guide users on how to interact with them.
What This Means for AI Advisory Tools
The findings put pressure on AI financial platforms to go beyond raw model capability and invest in prompt scaffolding, onboarding, and guardrails that ensure all users — not just tech-savvy ones — get consistently sound guidance. Half of Americans already use AI for some form of financial advice, per the study’s survey data. As adoption scales, the advice gap created by uneven prompting habits could become as significant a policy concern as access itself.