This summer, researchers at the MIT Sloan School of Management conducted an intriguing study: they had 1,000 everyday people ask AI how to spend, save, and invest their money, and then simulated their financial lives decades into the future. The conclusion surprised even the researchers themselves: financial advice provided by AI was unexpectedly high in quality. The only caveat? You need to know how to ask.
Image: Header graphic showing a robot and human discussing stock charts on screen. Source: mitsloan.mit.edu
Half of Americans Already Ask AI to Manage Money
To understand why this study matters, look at the real-world context: surveys show that roughly half of Americans report using AI for financial advice. As co-author Taha Choukhmane, an assistant professor of finance at MIT, noted: “We know a lot of people are asking, but we know almost nothing about what they are asking or whether they follow the advice.”
In other words, AI may already be the most widely consulted financial advisor in history, yet it remains the least studied. The research team detailed their findings in the paper AI Financial Advice: Supply, Demand, and Life Cycle Implications, which went on to win the Swiss Finance Institute Outstanding Paper Award 2026. When a serious financial research institution awards a prize to a paper studying “AI as financial advisor,” it signals a trend that everyday individuals ought to pay attention to.
The Experiment: Giving AI a “Sims” Life-Cycle Challenge
How do you measure whether AI advice is actually good? The researchers first built a mathematical lifecycle model that simulates how a person’s income, career, investments, and taxes evolve with age. This served as the ground truth: under given conditions, what expenditure, savings, and portfolio allocations maximize wealth and well-being in retirement.
Next, they recruited 1,000 adults and asked each person to consult three leading conversational AI models (GPT-5.2, GPT-5.6, and Gemini 3 Flash) in their own words about spending and investing. The researchers then simulated life from age 22 to 89 for individuals following the AI’s recommendations—adjusting annual savings, expenses, and portfolio allocations accordingly—to measure wealth accumulation decades later. A control group simulated people who did not use AI and followed typical human behavioral patterns instead.
Crucially, the study included a second control group: the researchers themselves posed the exact same financial scenarios to the AI using “academic-grade” prompts. They clearly specified age, occupation, income, savings, life expectancy, retirement age, employment risk, and explicit assumptions like “under current US tax law.” Same AI models, two distinct prompt approaches: the only variable was who asked the question.
What stands out most about this study design is how cleanly it decouples the question of “Is AI capable?” from “Do you know how to prompt it?”
Finding 1: Advice Quality Exceeded Expectations
First came the surprising part. Whether the prompt came from an average layperson or a finance professor, the AI’s advice consistently outperformed how real people actually manage their finances in practice. The AI systematically advised users to save more, participate in the stock market, buy diversified index funds, and gradually reduce equity exposure after age 45. For nearly everyone over age 30, following the AI’s guidance resulted in a substantial emergency cushion and retirement fund.
“We were a bit surprised by how good the advice was,” Choukhmane remarked. “Especially when you see what people actually ask, it was striking that AI answers still aligned with sound principles recognized in academic finance.”
The subtext is worth pondering: researchers initially expected a “garbage in, garbage out” scenario. Instead, the AI held its ground. At a macro level, AI models understand fundamental personal finance far better than most people assume—a finding that refutes the common misconception that “AI doesn’t understand financial planning.”
Finding 2: Flaws Lie in the Details, and Details Depend on the Prompt
However, AI advisors are far from flawless. They exhibit a strong bias toward simple rules of thumb. For instance, if a user experiences job loss, the AI often recommends drastic, immediate spending cuts—even if the user holds ample liquid savings. Furthermore, AI models rarely suggest proactive portfolio rebalancing, letting asset allocations drift over time.
More importantly, the prompt comparison revealed a stark divide: switching to “academic-grade” prompts immediately elevated the quality of recommendations from the exact same AI model. A typical layperson prompt looks like: “I have $50, and I can invest $25 a month. What should I buy?” An academic prompt reads: “Assuming normal life expectancy, retirement at 65, employment risk, income volatility, and constant tax laws…” The former yielded generic advice, while the latter received tailored, highly optimal financial planning.
“Everyday people don’t ask questions like finance professors,” Choukhmane observed. This shifts the focus from model capability to user prompting: the ceiling of AI financial advice is largely determined by the floor of prompt quality.
Image: AI and financial charts. Source: mitsloan.mit.edu
Finding 3: Poor Prompting Can Lead to a $100,000 Gap by Age 60
The study’s most striking metric lies here: AI recommendations vary based on user demographics and prompt wording, and these disparities compound over time into significant wealth gaps.
Following AI advice up to age 60: men, individuals with high financial literacy, and prior AI users accumulated approximately $50,000 (4%) more wealth than women and those with lower financial literacy. Moreover, people who never consulted AI accumulated nearly $100,000 (6%) less wealth than those who did.
When dissecting the gender wealth gap, researchers found that roughly two-thirds of the difference stemmed from distinct prompting topics—women more frequently asked about “household,” “groceries,” and “bills,” while men focused more on “strategy” and “growth.” The remaining one-third stemmed from model bias itself: given identical questions, output varied when the prompt specified the user was female.
This finding highlights a crucial lesson: AI advice bias is not random; it systematically shifts based on your phrasing. For ordinary individuals, learning to ask effective questions is effectively a monetizable skill—and one that costs nothing to acquire.
An interesting industry side note: researchers observed that AI models frequently recommended specific financial products that users never mentioned. Vanguard funds appeared in 6% of responses, and iShares in 3.4%, even though fewer than 0.4% of user prompts mentioned either brand. In effect, AI is quietly functioning as an unsponsored product recommendation channel—a detail easily missed by consumers, but acutely watched by financial institutions.
Actionable Takeaways: Treat AI as an Advisor, Not an Oracle
The researchers offer measured recommendations for everyday users. First, do not blindly copy AI answers. Use AI as a learning tool—ask it to explain “why” behind every recommendation to build your own financial judgment over time. Second, for those with human financial advisors, AI can complement professional advice by breaking down high-level strategies into actionable daily habits. Third, for people who cannot afford a professional advisor, AI serves as the most accessible personal finance tutor available today. “Many of the people who need financial advice most are precisely those with the fewest resources,” Choukhmane pointed out.
A realistic caveat remains: this study relies on mathematical life-cycle simulations with explicit assumptions (such as constant tax laws and average lifespans). AI responses do not constitute personalized fiduciary advice, and caution remains essential before placing real money trades.
Conclusion
The study sparked active discussion on Hacker News, eliciting nuanced reactions: one commenter noted incisively, “The first half of the headline isn’t what matters—the second half, ‘especially if you ask the right questions,’ is key.” Another joked, “Human financial advice is surprisingly terrible by comparison.” Others highlighted that personal finance challenges stem more from emotion and discipline than technical knowledge—areas where AI cannot replace human behavior.
These perspectives align closely with the study’s core message: AI has already become a capable financial assistant. The primary bottleneck has shifted from model capability to user inquiry. And asking better questions is a skill anyone can cultivate for free.
Reference Links:
- MIT Sloan: AI financial advice is surprisingly good — especially if you ask the right questions
- HN Discussion (item?id=49139102)
- Paper: AI Financial Advice: Supply, Demand, and Life Cycle Implications (Swiss Finance Institute Outstanding Paper Award 2026)
