• 3 min read
AI money advice sounds smart enough to be risky
A TechRadar test of ChatGPT on everyday money questions shows why regulators worry: the answers are useful, confident, and often too persuasive.
Image: TechRadar
More than a quarter of UK consumers trust AI chatbots for money advice, according to a recent review by the Financial Conduct Authority (FCA). That is exactly the kind of trend worrying regulators: giving financial advice is supposed to be a regulated activity, while tools like ChatGPT, Claude, and Gemini are not.
In a TechRadar experiment, ChatGPT was asked a set of hypothetical questions using ChatGPT Pro in anonymous mode with memory turned off. The prompts covered ordinary decisions: whether to buy a £1,200 phone, what to do with £20,000 in savings, and whether to book a £2,000 holiday after a difficult few months.
The striking part was not obviously bad advice. It was that the answers often sounded thoughtful, nuanced, and reassuring. ChatGPT explained trade-offs, asked follow-up questions, and framed decisions clearly. But it also moved quickly from analysis to recommendation, often before it had enough context to make a truly suitable suggestion.
For the phone question, the chatbot noted the user was 38, earned £40,000 a year, had £8,000 in savings and £2,000 in credit card debt. It sensibly highlighted the cost of credit card debt and questioned whether the upgrade was necessary. But phrases like “the strongest financial move” gave the response more authority than the available information justified.
Recommended reading
Google ships Gemini 3.6 Flash, delays 3.5 Pro again
With the savings question, ChatGPT again sounded polished, discussing emergency funds, investing, and tax-efficient accounts. Yet it started suggesting how much to hold in cash and how much to invest before knowing basics such as housing status, dependants, upcoming purchases, or risk tolerance.
The holiday prompt exposed another issue: role drift. Asked about guilt over spending £2,000 on a trip, ChatGPT moved beyond finance into emotional support and reframing. That may feel helpful, but it blurs boundaries between financial adviser, therapist, and coach—roles with very different standards and accountability.
Why plausible answers are the real problem
The core concern here is suitability. Regulators use that term to ask whether advice genuinely fits a person’s circumstances, goals, and tolerance for risk. In TechRadar’s test, ChatGPT repeatedly offered recommendations while knowing very little about the user.
There is also the question of accountability. If a regulated financial adviser gives poor advice, consumers usually have complaint channels and protections. If a chatbot gives poor advice and someone acts on it, responsibility is much harder to pin down.
Just as important, fluent language can be mistaken for expertise. A clear, confident answer may still be wrong, incomplete, or inappropriate. The danger is not only bad advice, but advice that feels good enough that people never seek better help elsewhere.
That is why the conclusion is not a simple “never use it.” Chatbots can be useful for explaining jargon, outlining trade-offs, and giving people a starting point. But when users are stressed, uncertain, or unable to fact-check what they are reading, those same strengths can make AI advice especially persuasive. As TechRadar found, the technology does not need to perform badly to create risk—it only needs to perform well enough to earn trust.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via TechRadar
