Seeking comment: AI-generated financial advice

The findings ring true. The danger with these tools is that they sound most confident exactly where they're most likely to be wrong, and a consumer has no way to tell a good answer from a fluent, authoritative one that happens to be incorrect. That 12% on complex questions is the alarming part, because in personal finance almost everything becomes complex once it's about your own situation. Pensions, tax and mortgages all turn on individual circumstances the model doesn't know and won't think to ask about. A mortgage answer that's perfect for one person is actively wrong for the next. I use AI heavily in our own business, so I'm no sceptic. We have the expertise to catch it when it's confidently wrong; a consumer using it as their adviser does not. So I wouldn't read this as never use AI. Use it to understand the basics and ask sharper questions, then take the decision that's actually yours to a human. It's a brilliant starting point and a dangerous place to finish.

The 43% is the good news in this study. Nobody asks AI the easy questions. They ask the long-form ones about tax, pensions and their own situation, and that's where it was right just 12% of the time. What makes it so dangerous is that it answers wrong with exactly the same confidence as it answers right, and I've had clients arrive certain of something ChatGPT told them that was flatly wrong. Is the research fair? Mostly yes. AI is fine for learning what a question means. When it gets your money wrong, there's no complaint to make and nobody accountable, and that's what regulated advice is for.

Anything involving data processing is ruled by GIGO, garbage in, garbage out. AI learns from the internet, so it produces garbage faster. In FX, the risk is magnified by real-time rate volatility, shifting interest rate parities and multi-jurisdictional tax rules. Generic AI models lack live market feeds and stumble over cross-border regulation, a recipe for spot-forward spread errors and tax penalties. Across personal finance, the study's poor accuracy reflects a basic flaw: probabilistic language models are the wrong tool for precise planning. In pensions and tax, AI conflates regional rules and misses personal nuance leading to irreversible fines and flawed retirement strategies. On mortgages and debt, it explains the basics but can't judge lender underwriting or distressed debt. The danger is the "confidence gap", outdated or hallucinated advice delivered with the same authority as the right answer. AI is a research tool but no substitute for an experienced qualified adviser.

AI can be a useful starting point for explaining mortgage jargon or helping someone understand the questions they should be asking, but it is not a substitute for regulated advice. Mortgages are very case-specific: income structure, credit history, deposit, property type and individual lender criteria can all change the answer. The danger is that AI often sounds confident even when it is wrong or working from outdated information. I would use it to become better informed, but for a decision that could affect your finances for decades, verify everything with a qualified adviser.

I'm not surprised that AI makes errors, but being wrong over half the time (57%) is very concerning. This is compounded by its conviction and confidence. When it comes across as this confident, it's easy for experts using AI to question their own knowledge. So, for the average consumer, they have little chance of assessing the validity of any AI generated financial advice. The rising use of AI as supposedly an alternative to professional advice is almost certainly going to create issues for users down the line. And when things go wrong, users would have any recourse or compensation, other than a simply 'Sorry, I was incorrect' response from their AI agent.

It’s extremely concerning how much misleading advice AI chatbots are giving, affecting millions of people who use them for quick help.

The biggest danger is that AI can produce a wrong answer that sounds confident and authoritative. Regulations around topics such as taxes, mortgages, or pensions change regularly, so simple searches can return outdated information. These small errors could cost someone thousands of pounds.

Utilising AI for initial insight can be useful for gaining an understanding of a financial concept, but consumers need to be cautious and should never treat AI-generated answers as financial advice. Answers to a major, specific personal matter are where you are most likely to be led astray, as it doesn’t take into account an individual's circumstances, which is where professionals are much more reliable.

Don’t substitute it for regulated financial advice, as wrong decisions financially will have a significant impact on your life, and AI won’t be held accountable.