Blab AI
AI Financial Advice Boosts Savings but Leaves Gaps, Study Finds
A new MIT Sloan and Stanford Graduate School of Business study has shown that large‑language models (LLMs) can help most adults save more and invest more wisely, but the guidance also falls short when life throws unexpected shocks and can widen wealth gaps.
Published in the Journal of Financial Planning and honored with the Swiss Finance Institute Outstanding Paper Award 2026, the research examined how people use AI chatbots to plan spending, saving and investing. The authors tested GPT‑5.2, GPT‑5.6 and Gemini 3 Flash—three of the most advanced LLMs available in 2026—against a survey of 1,000 U.S. adults. Participants wrote prompts for financial advice, and the researchers simulated what would happen if each respondent followed the chatbot’s recommendations over their lifetime.
The simulations reveal that, on average, AI advice nudges users toward higher savings rates during working years, greater participation in diversified equity funds, and a gradual shift to lower equity exposure after age 45. The study found that following the advice would create a sizable savings buffer for people older than 30, a benefit that the authors call “good but improvable.”
However, the models struggled to adapt to sudden changes. When a simulated user lost a job, the chatbots often recommended cutting spending sharply, even when the user still had a savings cushion. The advice also tended to let portfolios drift over time, rather than actively rebalancing them to maintain target allocations.
Prompt quality matters. The researchers compared ordinary prompts—phrases like “Where should I invest starting with $50 and adding $25 a month?”—to “academic” prompts that supplied full financial information, including age, employment status, income, savings balances and assumptions about the economy. The structured prompts produced advice that was closer to the authors’ life‑cycle model and reduced the tendency to rely on simple rules of thumb.
The study also uncovered systematic differences in the advice received by users. Men, people with higher financial literacy and those who had used AI for financial advice before received recommendations that, over a lifetime, translated into about 5 % more wealth at age 60. The same factors were associated with roughly 4 % lower wealth for women and less literate users, and a 6 % lower wealth for users without prior AI experience.
The authors attribute the gaps partly to the content of the prompts. Women were more likely to mention family, grocery and pay, while men used words such as strategy, crypto and growth. Even when the same question was labeled as coming from a woman or a man, the model sometimes altered its answer, a pattern that could reflect bias learned from training data or reasonable inference about demographic differences.
“LLMs can be an affordable source of financial guidance, especially for people who cannot afford a human advisor,” said Taha Choukhmane, an assistant professor of finance at MIT Sloan and co‑author of the paper. “But the variation in advice shows that the technology is not yet ready to replace the nuanced judgment of a professional.”
The study also noted that AI chatbots frequently recommended specific account types and products that users had not mentioned. Vanguard appeared in 6 % of responses and iShares in 3.4 %, even though fewer than 0.4 % of prompts referenced either company. The authors suggest that this could influence how consumers discover and compare financial products.
In practice, the authors see AI advice as a complement to human advisors. “A human advisor might meet a client twice a year, but an AI can help implement that advice in real time,” Choukhmane said. For those who cannot afford a human advisor, the study indicates that AI can provide a low‑cost alternative.
The research underscores the need for better prompt design and for AI systems to handle life‑cycle events more robustly. It also highlights the importance of monitoring for bias and ensuring that the technology does not widen existing wealth gaps.
As of July 2026, more than half of adults in the United States and the United Kingdom have asked AI for financial advice, according to a MIT Sloan survey. The new findings suggest that while the technology can improve financial habits, users and providers must remain vigilant about its limitations.
The study’s authors plan to extend their work by testing additional LLMs and exploring how regulatory frameworks, such as the Utah Artificial Intelligence Policy Act, might shape the deployment of AI financial advice.
Published in the Journal of Financial Planning and honored with the Swiss Finance Institute Outstanding Paper Award 2026, the research examined how people use AI chatbots to plan spending, saving and investing. The authors tested GPT‑5.2, GPT‑5.6 and Gemini 3 Flash—three of the most advanced LLMs available in 2026—against a survey of 1,000 U.S. adults. Participants wrote prompts for financial advice, and the researchers simulated what would happen if each respondent followed the chatbot’s recommendations over their lifetime.
The simulations reveal that, on average, AI advice nudges users toward higher savings rates during working years, greater participation in diversified equity funds, and a gradual shift to lower equity exposure after age 45. The study found that following the advice would create a sizable savings buffer for people older than 30, a benefit that the authors call “good but improvable.”
However, the models struggled to adapt to sudden changes. When a simulated user lost a job, the chatbots often recommended cutting spending sharply, even when the user still had a savings cushion. The advice also tended to let portfolios drift over time, rather than actively rebalancing them to maintain target allocations.
Prompt quality matters. The researchers compared ordinary prompts—phrases like “Where should I invest starting with $50 and adding $25 a month?”—to “academic” prompts that supplied full financial information, including age, employment status, income, savings balances and assumptions about the economy. The structured prompts produced advice that was closer to the authors’ life‑cycle model and reduced the tendency to rely on simple rules of thumb.
The study also uncovered systematic differences in the advice received by users. Men, people with higher financial literacy and those who had used AI for financial advice before received recommendations that, over a lifetime, translated into about 5 % more wealth at age 60. The same factors were associated with roughly 4 % lower wealth for women and less literate users, and a 6 % lower wealth for users without prior AI experience.
The authors attribute the gaps partly to the content of the prompts. Women were more likely to mention family, grocery and pay, while men used words such as strategy, crypto and growth. Even when the same question was labeled as coming from a woman or a man, the model sometimes altered its answer, a pattern that could reflect bias learned from training data or reasonable inference about demographic differences.
“LLMs can be an affordable source of financial guidance, especially for people who cannot afford a human advisor,” said Taha Choukhmane, an assistant professor of finance at MIT Sloan and co‑author of the paper. “But the variation in advice shows that the technology is not yet ready to replace the nuanced judgment of a professional.”
The study also noted that AI chatbots frequently recommended specific account types and products that users had not mentioned. Vanguard appeared in 6 % of responses and iShares in 3.4 %, even though fewer than 0.4 % of prompts referenced either company. The authors suggest that this could influence how consumers discover and compare financial products.
In practice, the authors see AI advice as a complement to human advisors. “A human advisor might meet a client twice a year, but an AI can help implement that advice in real time,” Choukhmane said. For those who cannot afford a human advisor, the study indicates that AI can provide a low‑cost alternative.
The research underscores the need for better prompt design and for AI systems to handle life‑cycle events more robustly. It also highlights the importance of monitoring for bias and ensuring that the technology does not widen existing wealth gaps.
As of July 2026, more than half of adults in the United States and the United Kingdom have asked AI for financial advice, according to a MIT Sloan survey. The new findings suggest that while the technology can improve financial habits, users and providers must remain vigilant about its limitations.
The study’s authors plan to extend their work by testing additional LLMs and exploring how regulatory frameworks, such as the Utah Artificial Intelligence Policy Act, might shape the deployment of AI financial advice.
