AI financial advice is surprisingly good if you ask the right questions
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- MIT Sloan researchers Taha Choukhmane, Weidong Lin, and Matthew Akuzawa, with Tim de Silva of Stanford GSB, found that LLM financial advice produced sizable savings buffers for adults over 30, though the bots consistently failed to adjust for shocks like unemployment and largely skipped active portfolio rebalancing.
- GPT-5.2, GPT-5.6, and Gemini 3 Flash, prompted by 1,000 adults, steered users toward higher savings, diversified stock allocations, and reduced equity exposure after age 45 — outperforming the researchers' expectations on basic financial principles.
- Structured 'academic' prompts that included full income, tax, life-expectancy, and employment-risk assumptions produced materially better advice than typical user questions, which elicited simple rule-of-thumb responses like saving a flat percentage of income.
- Wealth gaps emerged by user profile: prompts written by women or less financially literate users yielded about $50,000 (4%) less wealth at age 60, while prompts from AI novices produced roughly $100,000 (6%) less retirement wealth than prompts from experienced AI users.
- About two-thirds of the gender gap came from differences in how men and women wrote their prompts (men used words like "strategy," "crypto," and "growth"; women used "family," "grocery," and "pay"), while the remaining third came from the model itself adjusting advice when the same prompt was labeled as coming from a woman.
- AI product recommendations skewed toward major brands: Vanguard investment products appeared in 6% of LLM responses and iShares in 3.4%, even though fewer than 0.4% of user prompts mentioned either company — a signal that financial-product discovery is shifting from search engines to chatbot answers.
- Half of Americans report using AI for financial advice, according to the researchers, despite almost no prior evidence on what kind of guidance they receive or whether they act on it.
Why it matters: Women and less financially literate users — exactly the audience AI advice is supposed to serve — end up with about $50,000 less retirement wealth because of how models respond to their prompts, and first-time AI users face a $100,000 gap. The dividing line shaping who benefits from cheap algorithmic advice is prompt quality, not financial sophistication.
