What AI Tells People Seeking Low-Cost Financial Advice

Would you trust a large language model to help plan your financial future? Many Americans already do. In a 2025 survey, more than half said they’d asked AI for financial advice. By comparison, about 40% have worked with a human financial advisor, suggesting that when generative AI talks money, people listen.

Tim de Silva, an assistant professor of finance at Stanford Graduate School of Business, has been struck by how quickly Americans have embraced generative artificial intelligence as an investing tool. “There have been lots of innovations in ways in which people could get financial advice over the last couple of decades,” he says, noting robo-advisors as an example. “But AI is the first new tool people really seem to be taking up rapidly and in waves.”

AI holds the promise of making guidance more accessible for people who can’t afford professional financial advice. Yet what kind of financial advice is it giving?

In a new paper, de Silva, along with Taha Choukhmane, Matthew Akuzawa, and Weidong Lin of the MIT Sloan School of Management, measures and analyzes the quality of AI-generated financial guidance. The team finds that while LLMs generally give good long-term advice, the quality of their guidance varies substantially based on the prompts users write.

“AI seems to be nudging people in the right direction,” says de Silva, a faculty affiliate at the Stanford Institute for Human-Centered AI. “It’s not perfect, but it’s better than the way many people make decisions, such as talking to friends and family or doing simple internet searches. That’s not something that should be taken for granted: It’s not at all obvious LLMs would provide good financial advice, because the way they are trained has nothing to do with that objective.”

De Silva and his colleagues asked 1,000 people to write three prompts for how they would ask an LLM for financial advice: one describing their situation, one asking for advice on spending and saving, and another seeking advice on investing in different asset classes. About half of the participants said they had recently used AI for financial advice or information.

Real Advice for Virtual People

These human-generated prompts were then used to run a simulation featuring “virtual people” who experienced changing labor and asset market conditions, including job losses, stock market slumps, and death. Prompting ChatGPT 5.2 and Gemini 3 Flash with human-written queries from similar subjects, the researchers evaluated the impact of following the LLM’s financial advice over the course of a lifetime.

The authors find that LLMs nudged virtual people toward the kind of smart financial habits economists have long prescribed but that few real people actually practice: spending less during their working years, building up a “buffer stock” of savings to weather hard times, investing in the stock market, and gradually shifting their savings from stocks to safer investments later in life. The results were strikingly different from individuals’ observed behavior. For example, most of the simulated users over age 30 accumulated meaningful savings, and many retired with more than $1 million. In contrast, around 40% of people who wrote prompts had less than $10,000 in savings.

In many cases, AI gave users more than they asked for. Liquidity was mentioned in 83% of responses, even though only 6% of users brought it up. Even though only 20% of people asked about saving, the AI-generated advice consistently pointed to safer options, such as high-yield savings accounts and government bonds.

Overall, these results make a case for AI as a low-cost alternative to traditional financial advice. “It does very well on general principles,” de Silva says, noting its bias toward saving. “That’s maybe not surprising, given that a lot of the texts the model will have read are describing how most people don’t save enough.”

Still, there was plenty of room for improvement. LLMs had a hard time adjusting users’ spending after they experienced an income shock, such as job loss, passively shifting their portfolios instead of actively rebalancing them. The models also recommended drawing down retirement savings too slowly, often following heuristics such as the 3–4% “safe withdrawal” rule recommended by many financial advisors.

A key finding of the paper is that the models’ advice was only as good as the questions people asked. People with low financial literacy wrote prompts that conveyed less information about their financial situation, resulting in less personalized advice. Following this advice left simulated users nearly $50,000 poorer by age 60. Inexperience with AI also widened the gap: Users new to LLM-based financial guidance ended up nearly $100,000 behind those with more experience.

Asking the Right Questions

De Silva’s team also discovered a gender gap in how men and women prompted AI and the financial advice that resulted. Women used words such as “family,” “grocery,” “credit,” and “loan” more often. This resulted in advice that emphasized keeping more money in liquid or safer assets and building an emergency buffer for household expenses. Men were more likely to mention words like “portfolio,” “equity,” “strategy,” and “crypto,” leading AI to respond with more aggressive investment advice. Overall, women received advice that would leave them nearly $60,000 poorer than men by the time they retire.

In general, the AI models did not know users’ genders. Yet even when men’s and women’s prompts were identical, the AI recommended less stock exposure to women. The researchers say it is hard to know whether that reflects biases built into AI training data or reasonable assumptions about women’s longer lifespans or greater risk aversion.

“The way you write the questions matters a ton,” says de Silva. “The models have gotten better, and even if you ask the wrong questions, they can still push you in the right direction. But they don’t do so entirely.”

The AI models often recommended specific financial products, such as Vanguard or Fidelity index funds, even when users did not request them. Those are reasonable recommendations, de Silva says, but he flags a potential conflict of interest. “The designers of these models are going to know that a lot of people are using these things for financial advice,” he says. “There may be an incentive to get people to buy certain products.”

Overall, the results leave de Silva feeling more comfortable about the rapid adoption of generative AI as a financial advising tool. LLMs are not about to fully replace human financial advisors, especially for the wealthy. But they can help people who can’t afford more formal financial advice.

De Silva is interested in how people ultimately put the advice they get from AI into practice. Yet first, he’d like to see people develop a baseline knowledge of financial literacy so they can ask LLMs better questions. “Because then you can use this tool in a very powerful way,” he says.