AI Financial Advice: MIT Study Finds LLMs Effective but Prompt-Dependent

LLMs provide high-quality baseline financial guidance

Research from the MIT Sloan School of Management indicates that following AI-generated financial recommendations can result in sizable saving buffers for nearly all individuals over the age of 30. Large Language Models (LLMs) consistently steer users toward fundamental financial best practices, including increasing savings during working years, investing in diversified stock funds, and reducing equity exposure as they approach retirement (specifically after age 45).

According to Taha Choukhmane, assistant professor of finance at MIT Sloan, LLMs offer an affordable and accessible alternative to traditional human financial advisors, helping users bypass the high costs and potential conflicts of interest associated with professional human guidance.

The "Prompt Gap": Expert vs. Novice Queries

The quality of AI financial advice is heavily dependent on the structure of the prompt. The study found a significant performance gap between "regular" user prompts and "academic" prompts:

  • Novice Prompts: Typical users often ask simple, narrow questions (e.g., "Where should I invest starting with $50?"). These prompts tend to elicit "rule-of-thumb" responses that lack nuance.
  • Academic Prompts: When researchers used structured prompts containing full financial data—including age, job status, income, savings balances, and specific economic assumptions (such as tax laws and Social Security rules)—the LLMs provided significantly better, more tailored advice.

Despite these improvements, LLMs still struggle with active portfolio rebalancing and adjusting to sudden financial shocks. For instance, the models often advised users who lost their jobs to cut spending too aggressively, even when the users had sufficient savings to weather the transition.

Demographic Disparities and Wealth Gaps

The study highlights a concerning trend: AI advice can inadvertently widen wealth gaps because the quality of the output depends on the user's existing financial literacy and demographic profile.

Wealth Outcomes by User Profile

Users who are men, more financially literate, or experienced with AI generated approximately 5% more wealth by retirement. Specifically:

  • Gender and Literacy: LLMs recommended higher equity allocations to men and financially literate users. This compounded into roughly $50,000 (4%) lower wealth at age 60 for women and less literate users.
  • AI Experience: Users unfamiliar with AI received lower recommended saving rates, resulting in nearly $100,000 (6%) less wealth at age 60 compared to experienced AI users.

Sources of Bias

These disparities stem from two primary sources. First, users ask different questions; the study noted that women more frequently mentioned "family" and "grocery," while men used terms like "strategy" and "growth." Second, the models themselves may exhibit bias; approximately one-third of the gender wealth gap was attributed to the model changing its advice when the same prompt was labeled as coming from a woman rather than a man.

Implications for the Financial Industry

For Consumers

To maximize the utility of AI financial tools, users should move beyond simple questions and adopt life-cycle planning and portfolio theory in their prompts. Experts suggest using AI as a tool for building financial understanding rather than blindly following its output. AI is best positioned as a complement to human advisors—helping users implement professional advice in real-time between semi-annual human consultations.

For Businesses

LLMs are shifting how consumers discover financial products. The study found that AI often recommends specific providers (e.g., Vanguard or iShares) even when the user did not mention them. This suggests that for financial firms, visibility within LLM recommendation patterns may become more important than traditional search engine optimization (SEO) or direct marketing.

Community Perspectives and Counterpoints

Discussion among technical users and practitioners provides additional context to the MIT findings:

  • The "Normie" Baseline: Some users argue that AI is simply distilling "common wisdom" or "financial platitudes." As one user noted, "You could replace the AI with a piece of paper that says 'set aside 10% of your income and invest it in an ETF' and it would outperform the financial 'advice' that people receive on a daily basis."
  • The Human Element: Critics point out that financial planning is often more about psychology than mathematics.

"The hard part is behavioural/emotional/psychological rather than technical... That’s where a real advisor earns their keep. Understanding the client and instilling confidence/comfort."

  • Practical Utility: Some users report high success using AI for complex tasks like detecting tax overpayments or analyzing CSV exports from budgeting software (e.g., YNAB) to identify spending patterns that human accountants missed.
  • Risk of Monetization: There are concerns that as AI becomes the primary interface for financial advice, the responses may be compromised by advertising, with models potentially steering users toward high-risk products in exchange for payment from financial firms.

Sources