It’s not much of a shock when you read that AI users are tapping into the pools of knowledge AI houses to help them with their financial inquiries.
One just needs to look at the stats—according to data produced by Intuit Credit Karma that was detailed by Credit Karma, “66% of Americans surveyed who have used GenAI before say they have used it to seek financial advice, rising to 82% of Gen Z and Millennials.” Using AI to assist with one’s money questions sticks out as one of the most used applications for AI, as “finance ranks as the second most common use case for GenAI (41%), just behind health and wellness (44%).”
According to another study (called the Artificial Authority” report) conducted by an AI and technology firm called Saturn, turning AI into a financial adviser might not be the smartest move to make. Saturn’s Domain team came to that conclusion after testing 18 of the most popular AI tools, which include ChatGPT, Gemini, Claude and Copilot.
Saturn’s results all point to one sobering outcome: chatbots aren’t all that reliable when you’re looking for legit financial counsel.
Saturn’s benchmarking study produced some pretty alarming results
The most striking statistics produced by Saturn’s report point to the trend of AI models being wrong more often than they were right when presented with financial questions.
Across the 17 AI models that Saturn tested, the average accuracy rate they delivered came in at 43%, which means those same chatbots got financial questions wrong 57% of the time. And when presented with complex, multi-step scenarios that focused on tax rules and more precise financial figures, the accuracy of those tested AI tools fell to 12%, with mistakes made 88% of the time as well.
It doesn’t come as much of a surprise that the free AI models that were tested produced a 63% failure rate, while the paid models delivered a 49% rating instead. Claude Haiku 4.5 got the distinction of being the worst-performing free model, as it gave out wrong or incomplete answers 82% of the time. ChatGPT-5.6 Luna (max) emerged as the best-performing free model, even though its rate of giving incorrect or incomplete replies was 56% of the time.
Out of all the models that were tested, Saturn found that Anthropic’s paid-for Claude Opus 5 (reasoning) model produced the best results since it achieved a pass rate of 61%. Even with that stat that points to it being the top-performing chatbot for financial advice, Claude Opus 5 still failed to give correct answers in almost four out of every 10 cases. Even when it was presented with more complex questions, it still made mistakes 67% of the time.
What’s even more alarming about Saturn’s study is how it conducted deep dives into the specific errors it came across from the AI models it tested that could result in severe financial outcomes. For instance, one model came up with a rule that said a college graduate could halt their student loan payments by moving abroad (in reality, doing so could actually lead to higher monthly repayments). Also, a Gemini model that was tested incorrectly told a borrower that taking a mortgage payment holiday wouldn’t affect their credit score.
Bottom line
Saturn’s report points to a sobering truth that many AI users need to come to grips with: AI financial advice remains unreliable and can leave its users prone to losing tons of money instead of making more of it.
A March 2026 report from EY highlighted how 53% of respondents prefer using an AI assistant to make financial investment decisions, while 14% prefer using an autonomous AI and 33% said they wouldn’t use any form of AI in such a case.
Thanks to Saturn’s findings, that 14% statistic may increase as AI users refrain from asking their chatbots personal questions about debt management, student loans, mortgages, pensions, inheritance tax and retirement planning.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom’s Guide
- Nvidia says 'local AI is here' — and it could change how you use AI at home
- OpenAI's new ad tracker may know what you do after you leave ChatGPT — here's what we know
- Apple's $250 million class action settlement website is now open — here's how eligible iPhone owners can file a claim and how much they'll get
