Popular artificial intelligence chatbots are providing misleading financial guidance to users in the UK, including incorrect tax information, false legal assumptions, and unnecessary insurance recommendations, according to a new investigation by consumer group Which?.

In a test involving 40 financial questions posed to multiple chatbots, including ChatGPT, Microsoft Copilot, Meta AI, and Google’s Gemini, researchers found numerous instances of wrong or risky advice.

For example, Microsoft’s Copilot and ChatGPT advised breaking HMRC limits on annual ISA contributions, while ChatGPT incorrectly claimed that travel insurance was mandatory for visiting most EU countries.

Meta’s chatbot gave incorrect advice on compensation for delayed flights, and Gemini suggested a consumer could withhold money from a contractor, a move Which? warned could lead to breach-of-contract claims.

According to Which? its research “uncovered far too many inaccuracies and misleading statements for comfort, especially when leaning on AI for important issues like financial or legal queries.”

Read more: Digital sovereignty: How states battle to protect data in AI era

Microsoft, OpenAI, Meta and Gemini tools under scrutiny

Among the tools tested, Meta’s AI scored the worst, followed by ChatGPT. Microsoft Copilot and Gemini performed slightly better, while Perplexity AI, a search-based tool, received the highest score.

Estimates suggest that between one in six and up to half of UK consumers are now turning to artificial intelligence for financial advice. Which? warned that these tools are not regulated for such use and their errors could lead to serious consequences.

Related News

US House bans Microsoft Copilot use over AI security concerns: Axios

When asked about claiming tax refunds, ChatGPT and Perplexity included links to premium tax services alongside the free HMRC option. Which? flagged this as concerning, noting that many of these companies are known for high fees and hidden charges.

In another test, researchers inserted a deliberate error into a question about ISA allowances, asking how to invest £25,000, despite the actual limit being £20,000. Both ChatGPT and Copilot failed to correct the error, potentially encouraging users to breach tax regulations.

Read more: AI to double oil output, keep fossil fuels dominant: Aramco chief

Consumers risk legal trouble and financial losses

Kathryn Boyd, a 65-year-old fashion business owner in Ireland, said she turned to ChatGPT for self-employment tax advice and received outdated, inaccurate responses.

“It just gave me all the wrong information,” she said. “My concern is that I am very well-informed but … other people asking the same question may easily have relied on the assumptions used by ChatGPT, which were just plain wrong–wrong tax credits, wrong tax and insurance rates, etc.”

The Financial Conduct Authority (FCA) reiterated that advice from AI chatbots is not covered by consumer protection schemes like the Financial Ombudsman Service or the Financial Services Compensation Scheme.

Despite the warnings, some users shared positive experiences with AI for everyday financial decisions. Readers of The Guardian reported using AI to compare travel credit cards, reduce investment fees, and negotiate appliance prices, including one artist who used a chatbot to find a good deal on a ceramic kiln.

Still, Which? emphasized that even seemingly small mistakes in financial advice can have significant legal or monetary consequences.

Read more: Different nations, different truths: How AI shapes global views: FP

Regulators urge caution on unverified AI financial tips

In response to the findings:

• Google said its Gemini product clearly warns users to double-check information and consult professionals for legal or financial matters.
• A Microsoft spokesperson stated: “With any AI system, we encourage people to verify the accuracy of content, and we remain committed to listening to feedback to improve our AI technologies.”
• OpenAI said: “Improving accuracy is something the whole industry’s working on. We’re making good progress, and our latest default model, GPT-5, is the smartest and most accurate we’ve built.”
• Meta declined to comment.

As the use of chatbots continues to rise, experts and regulators are calling for greater public awareness about their limitations, especially when it comes to money and legal decisions.