AI Chatbots Give Wrong Financial Answers Most of the Time, UK Research Finds

UK research covered by the FT this week found AI chatbots answer financial questions incorrectly up to 57% of the time, renewing warnings on AI money advice.

A wave of coverage on both sides of the Atlantic has put a hard number on a worry that regulators have been voicing for two years: when people ask AI chatbots questions about their money, the answers are wrong more often than they are right. According to the Financial Times, a recent UK study found AI chatbots give wrong answers to financial queries "most of the time."

For the latest breaking AI news and analysis, visit AI Buzz Wire.

What the Studies Found

The headline figure comes from trade publication Financial Reporter, which reported earlier in September that AI models give incorrect financial advice 57% of the time, citing research by Saturn. IBS Intelligence reported the same 57% error rate. Yahoo Finance's coverage described the research in connection with UK retirement platform PensionBee, while BeInCrypto reported that the test spanned roughly 10,000 answers. The differing attributions in initial coverage suggest the research involved multiple parties; the consistent thread across outlets is the scale of the failure rate.

The errors are not abstract. Mortgage Professional America highlighted one example from the coverage: a chatbot told a mortgage borrower that taking a payment holiday would not affect their credit score. It can — and for a household making a major financial decision on that answer, the distinction is expensive.

By this week, the story had crossed into the mainstream. InvestmentNews reported under the headline "AI chatbots give wrong financial answers most of the time, study finds," and Tom's Guide warned readers that AI could be costing them money based on the study's findings.

A Pattern, Not an Outlier

The UK findings land on top of a growing stack of research with similar conclusions. A previous analysis widely covered in March found ChatGPT gave wrong answers to money questions about 35% of the time. Stanford and MIT researchers were reported in August to have found that AI delivered worse investment advice to women than to men. The Financial Times itself reported in August that nearly half of young Britons wrongly believe AI financial advice is regulated — a perception gap that makes inaccurate answers more dangerous, not less.

And usage keeps climbing regardless. NerdWallet's mid-year data reporting found Americans are already using chatbots for financial advice in significant numbers, which is precisely why the accuracy findings are drawing attention from consumer advocates rather than being filed away as academic trivia.

Why Financial Questions Break Chatbots

Financial advice sits at the intersection of everything large language models find hardest. Correct answers depend on current, jurisdiction-specific rules — tax thresholds, credit-scoring mechanics, pension regulations — that change and vary by country. The models also tend to hedge with confident-sounding generalities, and users often don't supply the personal context (income, existing debts, goals) that determines whether generic guidance applies to them.

There is also an asymmetry problem unique to money: a wrong answer about a poem is embarrassing, but a wrong answer about a payment holiday's credit impact can compound into real financial damage discovered months later.

What Regulators and Users Should Take From It

UK regulators have warned consumers about unregulated AI financial guidance since at least late 2025, when The Guardian reported official warnings over chatbots giving inaccurate financial advice. The new research gives those warnings a concrete failure rate. The sensible reading is not that AI is useless for personal finance — models can explain concepts well — but that the line between education and advice remains blurry for users and crisp only in regulators' rulebooks.

Until chatbot accuracy on domain-specific financial questions improves sharply, the practical guidance from the study's coverage is straightforward: treat AI money answers as a starting point for research, never as the decision itself, and verify anything consequential — credit implications, tax treatment, pension rules — against regulated sources before acting.