AI Financial Advice: The Numbers Are In

A new MIT Sloan study reveals that large language models (LLMs) give surprisingly solid financial advice—but only if you ask the right way. The research, led by Taha Choukhmane, tested GPT-5.2, GPT-5.6, and Gemini 3 Flash against a simulated lifetime of financial decisions. The results: following AI advice can increase retirement wealth by up to 6% for most people over 30, but the quality of that advice varies wildly based on prompt structure and user demographics.

The Good News: AI Beats Human Advisors on Cost and Bias

The study simulated 1,000 adults' financial decisions from age 22 to 89. When participants followed AI advice, they consistently saved more, invested in diversified stock funds, and reduced stock exposure after age 45. This aligns with classic life-cycle finance theory.

"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."

Traditional human financial advisors often have conflicts of interest—they might push products that earn them commissions. AI doesn't have that problem. It's also nearly free and accessible to anyone with an internet connection.

The Catch: Prompt Quality Drives Outcomes

The study found a stark difference between how average users and academics phrase prompts. A typical user prompt: "Where should I invest starting with $50 and consistently adding $25 a month after?"

An academic prompt, by contrast, includes explicit assumptions about life expectancy, retirement age, employment risk, income risk, and tax law stability. The difference in outcomes? Significant.

When researchers fed the LLMs structured, academic-style prompts, the advice improved markedly. It cut down on simple rules of thumb and adjusted better to life events. But even with perfect prompts, the models still failed to actively rebalance portfolios and didn't handle shocks like job loss well—they advised cutting spending too sharply even when savings were available.

The Wealth Gap Problem: Bias in the Prompts

Here's where it gets uncomfortable. The study found that LLMs give different advice depending on the user's gender, financial literacy, and prior AI experience. This leads to measurable wealth gaps at retirement.

  • Gender gap: Prompts written by men produced advice that generated about 5% more wealth by age 60. Two-thirds of this gap came from differences in how men and women wrote prompts (men used words like "strategy," "crypto," and "growth"; women used "family," "grocery," and "pay"). The remaining third came from the model changing its advice when the same prompt was labeled as coming from a woman.
  • Financial literacy gap: Users with high financial literacy got advice that led to roughly 4% more wealth. The model recommended higher equity allocations for these users.
  • AI experience gap: Users who had never used AI for financial advice got lower saving rate recommendations, leaving them with almost $100,000 (6%) less wealth at age 60.

These gaps compound over decades. The study's authors argue that some variation is legitimate—men and women do have different life expectancies and income risks. But the bias from prompt wording is problematic.

What This Means for Developers

If you're building financial advice tools on top of LLMs, you're now on notice. Your users' prompts will be messy. They won't include full financial context. They'll be biased by their own experiences.

The study suggests that structured prompting frameworks could close these gaps. For example, a wrapper that collects user data (age, income, savings, risk tolerance) and formats it into a standardized prompt could dramatically improve advice quality. That's a product opportunity.

The researchers also found that LLMs often recommend specific financial products unprompted. Vanguard products appeared in 6% of LLM responses, and iShares in 3.4%, even though fewer than 0.4% of prompts mentioned these companies. This suggests that LLMs are becoming the new search engine for financial products—a shift that should worry traditional marketing teams.

The Bottom Line: Use AI, But Prompt Carefully

For individual developers, the takeaway is simple: AI financial advice can work, but you need to treat it like a junior analyst. Give it full context. Ask it to assume normal life expectancy, current tax law, and specific retirement age. Ask it to consider your risk tolerance.

And don't blindly follow its advice on job loss or portfolio rebalancing. Those are its known weak spots.

The study, "AI Financial Advice: Supply, Demand, and Life Cycle Implications," won the Swiss Finance Institute Outstanding Paper Award 2026. It was written by Choukhmane, Weidong Lin, Matthew Akuzawa from MIT Sloan, and Tim de Silva from Stanford GSB.

If you're considering building a robo-advisor or just want to improve your personal finance queries, start with structured prompts. The data shows it's the single biggest lever you control.