Study finds AI chatbots often get money and tax questions wrong

In short: A study found that popular AI chatbots often give incorrect answers to personal finance questions, especially on taxes and calculations.

What's going on

Research by tech firm Saturn tested widely used chatbots from providers including ChatGPT, Claude, Copilot, Grok, and Gemini. Across more than 100 different money related questions, the models were wrong 57 percent of the time on average.

When questions were more complex, meaning they required more than one calculation, the results got worse. The models made mistakes in 88 percent of cases on average, and some were wrong up to 99 percent of the time.

Saturn said the errors included simple math mistakes, missing upcoming tax changes, and “hallucinated” rules, which means the chatbot made up a rule that does not exist (like a confident friend giving advice based on something they imagined). In one example, Saturn said a wrong answer about UK pension tax rules could lead to a £17,500 charge from HMRC. In another, it said a chatbot invented a student loan rule and claimed repayments could stop if a graduate moved abroad.

The study also found that paid versions and newer models tended to do better than free and older ones. However, even the best performer in the test still made mistakes in 39 percent of answers.

What to watch

Saturn’s chief executive said many people are already using chatbots for money advice, but AI financial advice is not regulated. That means users do not get the same protections they would get from a qualified human adviser, including possible compensation if advice is wrong. For now, experts quoted in the report said chatbots may be safer for basic tasks like budgeting, and riskier for decisions involving taxes, pensions, debt, or investing.

Source: Financial Times