Half of Americans are now asking AI for financial advice, and a new MIT Sloan study says that advice is surprisingly solid. According to Hacker News, researchers led by finance professor Taha Choukkhomane measured the quality of guidance from large language models and found it lines up well with what academics consider good financial planning. The catch: how you ask the question changes how much money you end up with.
“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,” Choukkhomane said. His team set out to answer that.
What the researchers did
The team built a model of how incomes, jobs, investments, and taxes typically evolve over a person’s life. That gave them a benchmark for what “good” decisions look like. Then they ran a real-world test:
- They asked 1,000 adults to write their own prompts to GPT-5.2, GPT-5.6, or Gemini 3 Flash.
- They simulated what would happen if people aged 22 to 89 followed that advice over decades.
- They repeated the exercise with polished “academic” prompts that included full financial details and clear economic assumptions.
An academic prompt spells everything out: age, job status, income, savings, life expectancy, retirement age, and assumptions like current tax and Social Security rules holding steady. Most people don’t write like that, and that gap turns out to matter.
What they found
The headline result is that AI advice, followed over time, builds sizable savings buffers for nearly everyone over 30. The models consistently pushed people to:
- Save during working years and draw down in retirement.
- Invest heavily in diversified stock funds.
- Reduce stock exposure after age 45, matching age-appropriate risk.
“We were somewhat surprised by how good the advice was,” Choukkhomane 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.”
The weak spots are just as clear. The models leaned on simple rules of thumb and didn’t adjust well when life changed. After a job loss, the AI told people to slash spending too hard, even when they had savings to cushion the blow. It also let portfolios drift instead of actively rebalancing them. Better, more structured prompts improved the advice, but the rebalancing problem stuck around.
The part that should worry you
Here’s what stands out: the advice changed based on who was asking, and those differences compound into real money.
- Prompts written by men and by financially literate users got higher equity allocations. Over a lifetime, that gap meant roughly $50,000 (4%) less wealth at age 60 for women and less financially literate users.
- People who had never used AI for financial advice got lower recommended saving rates, leaving them with almost $100,000 (6%) less wealth at 60 than users with prior experience.
Same tools, different outcomes, driven partly by how each group phrased their questions and what topics they raised. That’s a bias problem hiding inside a helpful product.
What you can do with this
The practical takeaway is that the prompt is the product. If you use AI for money decisions:
- Give it the full picture: age, income, job stability, savings, and debts.
- State your assumptions: retirement age, expected expenses, and that tax rules stay put.
- Ask it to plan for shocks, not just steady state. Tell it you have an emergency fund before it tells you to gut your spending.
- Push for active rebalancing. The models won’t volunteer it, so make it part of the ask.
- Sanity-check the risk level. If the allocation feels off for your age, say so and ask why.
The researchers are careful to note the limits. This is a simulation, not a track record of real portfolios, and the models still fumble on rebalancing and sudden life changes. Their broader point holds, though: LLMs can be an affordable, accessible alternative to human advisors who carry their own costs, biases, and conflicts of interest.
The gap between a casual question and a well-structured one is worth tens of thousands of dollars. That’s the real finding, and it’s one you can act on today. Full details are available at the original source.