Study reveals AI chatbots offer inaccurate financial advice

New research testing the most popular artificial intelligence (AI) chatbots by the consumer organisation Which? found that they frequently provide inaccurate and misleading advice on important financial and legal matters for British consumers. The tests, which involved 40 questions posed to five rival AI tools, revealed numerous inaccuracies.

For example, Microsoft's Copilot and ChatGPT incorrectly advised users to breach HMRC investment limits on ISAs. ChatGPT also incorrectly claimed that travel insurance is mandatory for visiting most EU countries, while Meta's AI provided incorrect information about claiming compensation for delayed flights. Google's Gemini suggested withholding payment from a builder for a poor job, a move Which? warned could lead to a breach-of-contract claim.

Meta's AI received the lowest score in the testing, followed by ChatGPT, while Perplexity, an AI specialising in search, received the highest score. The findings are significant given that estimates suggest as many as half of UK adults are using AI for financial advice. 

Training Announcement: Freevacy offers a range of short one-day courses on a range of data-related subjects, including data protection topics such as conducting DPIAs and privacy-by-design, as well as how to use AI tools responsibly, cybersecurity best practices, and information access. The interactive sessions cover basic concepts through to advanced examinations of specific areas. Find out more.