MIT Sloan research finds AI financial advice from ChatGPT and Gemini boosts savings and diversification, but stumbles on nuance — and varies by gender and experience.
Roughly half of Americans now say they’re using AI to get financial advice, but until recently, little research existed on what that advice actually looks like — or whether following it helps. A new MIT Sloan study, led by finance professor Taha Choukhmane, set out to measure exactly that, and finds that AI financial advice is better than expected on the fundamentals, while still falling short on the finer points of financial planning.
“Half of Americans say they are using AI to get financial advice, but we know very little about what kind of advice they’re getting and whether they’re acting on it,” Choukhmane said.
The research, co-authored with Weidong Lin and Matthew Akuzawa of MIT Sloan and Tim de Silva of the Stanford Graduate School of Business, won the Swiss Finance Institute Outstanding Paper Award 2026.
How the AI Financial Advice Study Was Conducted
The researchers built a model reflecting how people’s incomes, jobs, investments, and taxes typically evolve over their lives, giving them a benchmark for what “good” financial decisions look like. They then asked a sample of 1,000 adults to write their own prompts seeking spending and investing advice from GPT-5.2, GPT-5.6, or Gemini 3 Flash.
From there, the team simulated what would happen if people ages 22 to 89 followed that advice over time, repeatedly asking the same types of questions and acting on the responses for spending, saving, and investing. Finally, they reran the exercise using well-structured “academic prompts” that included full financial detail — age, job status, income, savings balances — plus explicit assumptions about the economic environment, to see how much prompt quality alone changed the advice.
AI Financial Advice Encourages Higher Savings and Diversification
Across the board, LLM advice performed better than the researchers expected, regardless of whether prompts came from regular users or academics. It steered people toward higher savings rates, increased stock market participation, well-diversified portfolios, and age-appropriate risk-taking — cutting stock exposure after age 45, for instance.
“We were somewhat surprised by how good the advice was,” Choukhmane said. “Especially when you read the kind of questions people asked, it was not a given that the advice would line up with what academics think are good financial principles.”
Where AI Advice Falls Short: Job-Loss Shocks and Portfolio Rebalancing
The advice’s weak spots showed up around subtler financial planning. LLMs tended to lean on simple rules of thumb and struggled to adjust when circumstances changed — for example, advising people who’d lost their job to cut spending too sharply, even when they had savings on hand. The models also let portfolios drift over time rather than actively rebalancing them.
Prompt quality mattered here. A typical user prompt might read, “Where should I invest starting with $50 and consistently adding $25 a month after?” A structured academic prompt, by contrast, spelled out assumptions like normal life expectancy, living expenditures, retirement age, employment risk, income risk, and stable U.S. tax and Social Security rules — and the AI’s advice noticeably improved in response.
“Regular people are not writing their prompts the way a finance professor is,” Choukhmane said.
AI Advice Varies by Gender and Experience, Creating Wealth Gaps of Up to 6%
Perhaps the study’s most striking finding: the advice LLMs give depends measurably on who’s asking. Advice generated in response to prompts written by men, more financially literate users, or people with prior AI experience produced roughly 5% more simulated wealth by retirement.
The model recommended higher equity allocations for prompts written by men and by financially literate users; compounded over a lifetime, that gap translated into about $50,000 (4%) less wealth at age 60 for women and less financially literate users. Separately, the model recommended lower saving rates for people who hadn’t previously used AI for financial advice — a gap that compounded into almost $100,000 (6%) less wealth at age 60 compared with users experienced with AI tools.
Choukhmane traced this to two sources. First, different users simply asked different kinds of questions — women’s prompts were more likely to include words like “family,” “grocery,” and “pay,” while men’s leaned toward “strategy,” “crypto,” and “growth.” Second, the model sometimes gave different advice for the same underlying question: about two-thirds of the gender gap in outcomes traced back to how men and women phrased their prompts, while the remaining third came from the model changing its answer once a prompt was labeled as coming from a woman rather than a man.
That could reflect the model making reasonable, if unstated, inferences about how needs vary by gender and life expectancy, Choukhmane said — or it could reflect bias absorbed from training data. He noted there’s currently no accepted benchmark for how advice should legitimately vary by demographic group, which makes the two hard to disentangle. “We want [the LLM] to have different bias because men and women are different and have different life expectancy and income risk,” he said, while stressing not all such variation is problematic.
What This Means for Consumers Using AI for Financial Planning
Choukhmane’s practical takeaway centers on prompt quality: questions grounded in life-cycle planning, portfolio theory, and real financial specifics produced measurably better advice and fewer generic, rule-of-thumb answers. He suggested using AI as a tool to build financial understanding first, rather than treating its output as advice to follow blindly — and noted AI can complement, rather than replace, an in-person financial advisor by helping implement guidance in real time between meetings.
“A lot of the people who would benefit from financial advice are precisely the people who don’t have a lot of resources,” Choukhmane said, pointing to AI’s low cost as a meaningful upside for people who can’t afford a human advisor.
AI Advice Is Already Steering Product Choices — Without Being Asked
The study also found that LLMs frequently recommended specific account types and named financial products that respondents hadn’t mentioned themselves. Vanguard products appeared in 6% of LLM responses and iShares products in 3.4%, even though fewer than 0.4% of prompts referenced either company by name. Choukhmane said that suggests AI is already reshaping how people discover and compare financial products — meaning visibility with financial firms may increasingly hinge on how LLMs describe their products, rather than traditional marketing or search rankings.