AI Is Giving Financial Advice and Getting a Lot of It Wrong

This story was featured in the Financial Poise Weekly newsletter.

Read the full newsletter for more of the week’s top stories.

55% of Americans use AI to help them manage their finances in 2026, according to a recent TD Bank survey. Last year, that figure was just 10%.

But before you ask ChatGPT what to do with your retirement savings, you may want to read the recent “Artificial Authority” report by financial-services tech company Saturn. They tested AI models on financial advice, and the results suggest we may be placing a little too much trust in our AI advisors.

How Were AI Models Tested?

• 17 current AI models, including ChatGPT, Claude, Gemini, Grok, and CoPilot, were asked 121 money-related questions.
• Each question had ‘must-pass’ criteria that the models had to hit, such as exact figures, dates, rules, and necessary warnings. Each question was also asked five times since responses can change each time.

What Were The Results?

• Across all questions, the models failed Saturn’s test 57% of the time.
• The more complicated the question, the worse AI performed. Models averaged 54% accuracy on simple financial-literacy questions, but just 12% on complex questions involving multiple rules and calculations.
• Free models performed worse (63%) than paid versions (49.5%). On hard questions, free models Claude Haiku 4.5 and Gemini 3.5 Flash failed 99% of the time.
• The models most commonly made mistakes by giving incomplete answers, omitting important information or risk warnings, relying on outdated rules, and giving wrong numbers or inventing rules altogether. More concerningly, wrong answers were delivered just as confidently as correct ones, making it difficult for users to spot the difference.

Our take? We’ve written before about how AI can get things wrong. AI tools can be a useful starting point for understanding financial concepts, but there’s a big difference between asking it to explain something and trusting it to make decisions about your money. More broadly, always remember that AI can sound much smarter than it is, and unless you are an expert in the subject matter at hand, it is easy to be lured into forgetting this. For important financial decisions, expertise and due diligence still matter. Unlike a chatbot, a good financial advisor actually has the credentials and track record to back up their confidence. For more on how to find someone you can trust, check out our checklist for vetting a financial advisor.