Overview
- Which? tested ChatGPT, Google Gemini, Gemini AIO, Microsoft Copilot, Meta AI and Perplexity with 40 consumer questions and found frequent inaccuracies and unclear guidance.
- Meta AI scored lowest at 55% and ChatGPT scored 64%, while Perplexity ranked highest, with Birmingham Live reporting a 71% score.
- Examples included advice that could breach HMRC ISA limits, a false claim that most EU countries require travel insurance, incorrect flight-compensation steps and risky guidance to withhold builder payments.
- Researchers also found ChatGPT and Perplexity surfaced links to fee-charging tax-refund firms alongside HMRC’s free service, which the watchdog called worrying.
- The FCA said tips from general-purpose AI tools are not covered by the Financial Ombudsman Service or the Financial Services Compensation Scheme, while tech firms urged users to verify outputs and OpenAI touted improvements in its GPT-5 model.