AI Financial Advice: Good, But Better With Right Prompts

The Rise of the AI Financial Advisor

Nearly half of Americans are now turning to artificial intelligence for financial guidance, according to recent surveys. This shift is reshaping the landscape of personal finance, but a critical question remains: is the advice actually any good?

A comprehensive new study from the MIT Sloan School of Management provides the first rigorous answer. The research, led by finance professor Taha Choukhmane, suggests that while LLMs like GPT-5.2 and Gemini 3 Flash can offer surprisingly robust financial counsel, their effectiveness is highly dependent on how users ask their questions—and who is doing the asking.

The findings come at a time when wealth management firms are grappling with clients who are increasingly consulting AI chatbots like Claude and ChatGPT for portfolio recommendations and tax advice. As one industry CEO told CNBC, "ChatGPT is the single largest investment advisor in the world right now."

How the MIT Study Worked

To assess the quality of AI-generated financial advice, the researchers built a sophisticated economic model simulating how income, investments, and taxes typically evolve over a lifetime. This provided a benchmark for what constitutes "good" financial decision-making.

They then asked 1,000 adults to write their own prompts seeking spending and investing advice from GPT-5.2, GPT-5.6, or Gemini 3 Flash. The team simulated what would happen if individuals aged 22 to 89 followed that advice over time. Finally, they repeated the exercise using meticulously crafted "academic prompts" that included full financial profiles and explicit economic assumptions.

The results were clear: the quality of advice varied dramatically. Prompts written by everyday users often led to generic rules of thumb, while academic prompts produced advice that aligned closely with established financial theory.

Strengths: Saving More and Diversifying

The study found that LLM advice was surprisingly good at encouraging core financial principles. Regardless of prompt quality, the AI consistently steered users toward higher savings rates, increased stock market participation, and well-diversified portfolios.

"We were somewhat surprised by how good the advice was," Choukhmane told MIT Sloan. The models recommended saving during working years, drawing down savings in retirement, and reducing stock exposure after age 45—all hallmarks of sound life-cycle investing.

For individuals over 30, following AI recommendations could result in sizable savings buffers, the research found. This is particularly valuable for those who cannot afford traditional human financial advisors, who often charge fees that can eat into investment returns.

Weaknesses: Missing Nuance and Failing to Rebalance

Despite these strengths, the AI advice fell short on several critical dimensions. The models struggled to adjust to shocks like unemployment, often advising users to cut spending too sharply even when they had adequate savings.

Perhaps more concerning, the LLMs allowed portfolios to "drift" rather than actively rebalancing them. This means the AI gave advice that, over time, led to risk profiles that deviated from the user's optimal allocation without correction.

"Regular people are not writing their prompts the way a finance professor is," Choukhmane noted. This gap between user behavior and optimal prompt engineering is a central challenge for the industry.

The Wealth Gap: How Prompt Variation Hurts Some Users

Perhaps the most troubling finding is that AI advice varies depending on who is asking. The study found that prompts written by men, more financially literate users, or those with prior AI experience generated about 5% more wealth by retirement.

The disparities are stark: women and less financially literate users received advice that left them with roughly $50,000 (4%) less wealth at age 60. Users without prior AI experience were left with almost $100,000 (6%) less than their AI-savvy counterparts.

This gap stems from two sources. First, different users ask different questions—women were more likely to mention "family" and "grocery," while men focused on "strategy" and "growth." Second, the model sometimes changes its advice even for identical questions, depending on whether it perceives the user as male or female.

Real-World Risks: When AI Gets It Wrong

Wealth management professionals caution that AI advice is far from infallible. A CNBC report highlighted cases where LLMs made simple but costly errors, such as ChatGPT telling a client that two ETFs were identical when they tracked the same index but used different weighting methods.

Wealth manager Luciana told CNBC that when she asked an LLM for examples of how clients could save on capital gains taxes, "it got the math completely wrong." Her assessment: "More often than not, especially on the advice side, it's wrong."

These risks extend beyond financial advice. Consumer Reports warns that people tend to over-trust AI-generated medical advice, and a recent study found many respondents couldn't tell the difference between an AI-generated answer and one from a real doctor. The same cognitive biases apply to financial advice.

Prompt Engineering Is the Key

The MIT research suggests that better prompts can significantly improve outcomes. Academic prompts that grounded the AI in life-cycle planning, portfolio theory, and real-world assumptions produced advice that was far more nuanced and accurate.

Experts recommend using multiple AI models to cross-reference advice. As one Forbes contributor noted, "If they disagree, I don't immediately decide which one is right. I usually dig a little deeper to figure out why they came to different conclusions."

For consumers, the takeaway is clear: use AI as a starting point, not a final authority. Ask follow-up questions, request sources, and always verify critical information with a human professional.

What This Means for the Industry

For wealth management firms, the rise of AI advice presents both a threat and an opportunity. Some advisors report that clients who consult AI chatbots come to meetings with more informed questions, leading to deeper conversations.

However, the industry is also grappling with the reality that AI may change how financial products are discovered. The MIT study found that LLMs often recommended specific providers—Vanguard appeared in 6% of responses, iShares in 3.4%—even when users never mentioned them.

As one insurance industry leader put it, "No AI can nurture relationships with your clients. That's your job." The human touch remains crucial, especially in times of crisis or when navigating complex personal circumstances.

For now, the best approach may be hybrid: use AI for research and second opinions, but rely on human advisors for implementation and personalized guidance. As Choukhmane suggests, AI can serve as "a good complement" to working with a financial advisor you meet with twice a year.