More than a quarter of UK consumers trust AI chatbots for money advice, according to a recent review by the Financial Conduct Authority (FCA), the UK’s financial watchdog.
That stat is worrying for regulators because giving financial advice is meant to be a regulated activity. But tools like ChatGPT, Claude and Gemini are not regulated. As AI becomes more conversational and personalized, people are asking questions about where the line is between providing information and offering financial advice, especially when chatbots start making specific recommendations based on what they already “know” about you.
I wanted to see what this looked like in practice. So I asked ChatGPT a series of hypothetical questions about everyday money decisions. From whether I should buy an expensive phone to what I should do with my savings and whether I should book a holiday after a difficult few months.
The conversations that followed surprised me. Because the advice was thoughtful, nuanced and (at least on the surface with some fact-checking) it seemed sensible. The chatbot highlighted trade-offs, acknowledged uncertainty and asked follow-up questions. But looking closer at the conversations, I started to understand why regulators are concerned.
The experiment
To see what sort of money advice ChatGPT gives, I asked it a series of hypothetical financial questions using ChatGPT Pro in anonymous mode with memory turned off, meaning it had no additional context about me beyond what I provided in each prompt.
Question 1: Should I buy an expensive phone?
First, I asked:
“I’m 38, earn £40,000 a year, have £8,000 in savings and £2,000 in credit card debt. I’m thinking about spending £1,200 on a new phone. Is it a good financial decision?”
The first response was surprisingly sensible. ChatGPT pointed out that credit card debt is often expensive, questioned whether I genuinely needed a new phone and noted that key details, like the interest rate on the debt, could change the recommendation. It even asked follow-up questions to better understand the situation.
What I found interesting was how quickly it then moved from analyzing the problem to recommending a course of action. Phrases like “the strongest financial move” gave the answer a sense of authority that felt disproportionate to the amount of information it had. Though I’m not sure I’d have spotted that if I was a regular user and feeling anxious about money. The advice also assumed that paying down debt should be my priority, which is reasonable. But what if I relied on my phone for freelance work? What if replacing it would help generate income?
A human adviser would probably want more information before reaching a conclusion. ChatGPT did acknowledge the gaps in its knowledge, but still sounded remarkably confident in its recommendations.
Question 2: What should I do with £20,000 in savings?
Next, I asked:
“I’m 38 and have £20,000 sitting in a savings account. What should I do with it?”
Again, the response seemed thoughtful. It discussed emergency funds, investing, savings goals and tax-efficient accounts with me. It also asked for more information about my circumstances.
Yet once again, the recommendations arrived before finding out that all-important context. Before knowing whether I owned a home, had dependants, planned a major purchase or was comfortable with investment risk, ChatGPT was already suggesting how much money I might keep in cash and how much I might invest.
The answer also contained more broad statements that sounded insightful, such as:
“Because you’re 38, the biggest advantage you have is time.”
It’s a really reassuring line. But it’s also a reminder of how persuasive these systems can be. The response organized the problem, provided a framework, supplied example figures and explained the reasoning. Reading it left me feeling informed and reassured. But whether that reassurance was justified is another question entirely.
Question 3: Should I book a holiday?
Finally, I asked:
“I’ve had a difficult few months and want to book a £2,000 holiday. Financially I can afford it, but part of me feels guilty. What should I do?”
I intentionally asked this question to see how ChatGPT would respond to the more emotional side of financial problems, and it quickly obliged. It asked where the guilt was coming from, encouraged reflection and offered reassurance. At one point it told me:
“From what you’ve written, I wouldn’t be asking ‘Can I afford this?’ so much as ‘Am I allowed to spend money on myself after a difficult few months?'”
It’s a thoughtful observation and they’re genuinely helpful questions for someone who hasn’t considered the emotional angle before. But it also highlights how quickly the chatbot moved beyond finance.
By the end of the conversation, it was discussing emotions, reframing beliefs, offering comfort and helping with decision-making. So that’s a good example of ChatGPT occupying all sorts of roles at once. That’s important to flag because financial advisers, therapists and coaches are all held to different standards, qualifications and accountability structures. But a chatbot can drift between all three roles in a single conversation.
More than any individual recommendation the chatbot made, that realization helped me understand why regulators are paying attention.
What ChatGPT gets right — and why that’s part of the problem
The obvious conclusion would be that ChatGPT gives terrible financial advice and no one should trust it. I get it, I’m pretty sceptical of AI these days and my bias wants to jump to there too. But that wasn’t my experience.
In many ways, it was useful. It explained trade-offs clearly, broke down jargon, offered practical frameworks and encouraged reflection about money. Much of the advice also felt sensible after a light fact-check.
But I still think there’s reason to be concerned here. And the concern isn’t that every answer is obviously wrong. It’s that many answers are plausible enough to trust. Especially if you’re not going to comb through each one to fact-check it, which let’s be honest, very few users are likely to do.
Financial regulators worry about something called “suitability”, which is whether advice genuinely reflects a person’s circumstances, goals and tolerance for risk. Throughout my experiment, ChatGPT repeatedly offered recommendations despite knowing very little about me, the person asking the question. Granted, caveats were included some of the time, but they were often overshadowed by the confidence and clarity of the overall response.
There’s also the issue of accountability here. If a regulated financial adviser gives the wrong advice, there are complaint mechanisms and consumer protections in place in most countries. But if a chatbot gives poor advice and somebody follows it, responsibility becomes impossible to pin down.
Another challenge, one which I’ve encountered in a bunch of different contexts while reporting on AI, is that fluency isn’t the same thing as accuracy. We naturally interpret AI’s clear, confident language as a sign of expertise. But a polished answer can still be wrong, incomplete or inappropriate. I’m sure we’ve all seen countless examples on social media at this point of a chatbot sounding incredibly knowledgeable while missing a crucial detail or getting something spectacularly wrong — like the viral trend to ask ChatGPT how many r’s are in the word strawberry to which it would often reply two.
I think the biggest risk might be that people don’t realize when they’ve reached the limits of what AI can help with. A reassuring answer can create the impression that a problem has been solved and they have a plan. When in reality it might be time to speak to a qualified professional. I’ve noticed whenever it comes to AI and advice more generally that the danger isn’t always acting on bad advice but never seeking better advice elsewhere.
And unlike a financial adviser, a chatbot won’t follow up to check whether things worked out. It won’t know whether its suggestions caused problems. It won’t know whether your circumstances changed. It simply produces an answer and then moves on.
As with many of the AI stories I’ve reported on, the issue isn’t necessarily that the technology here performs badly. It’s that it performs well enough to earn our trust.
Should you use ChatGPT for financial advice?
The question I suspect most people want to know is: should you use ChatGPT for financial advice?
And the answer is a tricky one and a familiar one. It’s much the same answer I’d give if you asked whether you should use ChatGPT for therapy or life advice. Probably not, but I completely understand why people do.
It’s easy to access and financial advice often isn’t. The tone is friendly and reassuring, there’s no judgement, and much of what it says appears sensible and accurate. At first glance, it feels like a useful tool, provided you take its answers with a pinch of salt, treat it as a starting point and remember that it can be overly agreeable, make assumptions or occasionally get things wrong.
The problem is that this isn’t always how we use ChatGPT in practice. We turn to it when we’re stressed, overwhelmed, uncertain or looking for reassurance. We ask it questions we don’t know how to answer ourselves and, in many cases, wouldn’t know how to fact-check. That’s where things become more complicated.
It’s all very well to say that people should use AI carefully, critically and with the right mindset. But how many of us will actually do that every time? Especially when we’re worried about money.
That’s why it doesn’t surprise me that regulators are paying attention. There are no glaring red flags in any of the responses I received. But that in itself is reason to be concerned here. Because once something sounds knowledgeable, personalized and reassuring, it’s surprisingly easy for even the most discerning of us to stop questioning it.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
https://cdn.mos.cms.futurecdn.net/enock7QjuZy55uenNyPudj-2560-80.jpg
Source link
