How to safely use AI chatbots for personal finance questions

In an article for The Conversation, a finance professor describes how to get reliable answers to personal finance questions from AI chatbots. Her starting point is that many households cannot afford the hourly rate of a human financial adviser, so for them a chatbot is their only available source of advice rather than a supplement. According to the author, chatbots are reliable for widely accepted recommendations—building an emergency fund, prioritizing repayment of the debt with the highest interest rate, taking full advantage of the employer contribution to a retirement plan, investing in low-cost index funds—especially if the user supplies their own specific numbers and lets the chatbot only do the calculations.

According to the author, roughly a tenth of decisions are riskier: those that are major, irreversible, tax-sensitive or connected with the sale of a financial product—for example, withdrawals from retirement accounts, handling inherited accounts, applying for social insurance benefits or offers of annuities or life insurance. In these cases, a chatbot may give confident but incorrect answers that could cost users thousands of dollars.

The author recommends specific steps: have the chatbot first ask follow-up questions like a real adviser, ask for the strongest counterargument after each recommendation, ask it to explain the assumptions used in the calculation (e.g. a constant annual rate of return), and verify figures such as tax limits or deadlines directly with institutions such as IRS or Social Security Administration. She also warns against sharing identifying information, such as account numbers or a national identification number. As an intermediate option between a free chatbot and an expensive human adviser, she mentions robo-advisers, automated platforms that manage a diversified portfolio for a fee of roughly 0.25 to 0.50 % of assets under management annually, often with no minimum deposit.

The available version of the text ends midway through a description of a 2022 study by the author and her co-authors on the typical user of a robo-advisory platform. You can find details in the source article.