Your Client Asked ChatGPT About Their Money. Here’s How to Respond.

When a client brings you financial advice from ChatGPT, welcome it. Ask to see what the AI said, sort the general facts from the personal recommendations, check the facts together, and fill in what the AI couldn’t know about them.

You are not there to beat the chatbot. You add the judgment, context and accountability it can’t.

More clients are asking AI first

In a survey for MDRT, run by Opinium in April 2026 with 2,000 adults in Singapore, more than a third (35%) of people who manage their money online said they use AI tools for financial purposes. 29% said they use AI chat tools such as ChatGPT and Gemini.

The top uses were understanding ideas like mortgages, investing and taxes (54% of AI users) and comparing products such as loans, credit cards and insurance (51%). Nearly one in five AI users (19%) said they rely mainly on AI-generated advice when they make financial decisions.

The reasons are easy to understand. Convenience and cost topped the list. An AI tool answers at 11pm, charges nothing, and never makes anyone feel silly for asking a basic question.

So the question is no longer whether your clients will ask AI about money. Many already do. The question is what happens when they bring that answer to you.

What an AI chat tool is, in plain words

A generative AI chat tool, such as ChatGPT or Gemini, writes an answer by predicting the most likely next words, based on patterns in the huge amount of text it learned from.

That makes it good at explaining an idea in simple words, summarising a long document, or listing questions to think about. It also has three blind spots that matter for money questions:

• It only knows what your client typed. If they left out their family, their existing cover or their cash flow, the answer is built on guesses.
• It may not know this year’s rules or figures. Some tools search the web and some don’t, and either way they can mix up one country’s rules with another’s.
• It can be wrong and sound completely sure. People call this a hallucination: an answer that reads well but isn’t true.

When the UK consumer group Which? tested six AI chatbots on money and consumer questions in November 2025, it found errors, incomplete answers and an overconfident tone, and noted that the tools rarely suggested speaking to a professional.

For more on how these tools work, see How AI “thinks”.

Five steps when a client brings you an AI answer

1. Thank them, and mean it

A client who did some homework is an engaged client. If you sound defensive, they’ll stop telling you what they read, and keep reading it anyway. “Good that you looked into it. Can you show me what it said?” keeps the door open.

2. Ask to see the question, not just the answer

An AI answer is only as good as the question. Ask what they typed. Most of the time the question was short and the important details were missing, and that alone explains most of the gaps in the answer.

3. Sort the answer into three piles

Go through it with them and sort each point:

• General facts: how something works, or what a term means. Often fine, but still worth checking.
• Assumptions: anything it guessed about your client, such as their age, income, family or goals.
• Recommendations: anything that tells them what to do. This is where the missing context matters most.

4. Check the facts together

For anything that depends on Singapore rules or current numbers, go to the source with your client: the government agency’s own website, or the product’s own documents. Doing it together teaches them a habit they can use the next time they ask AI something.

5. Fill in what the AI couldn’t know

This is your part: their goals, what they already have, their cash flow, the people who depend on them, how they’d feel if things went badly, and what has changed since you last met. Then take it through your usual fact-find and recommendation process, and record it the way you normally would.

An example

Here’s how it might go. The client and every detail are invented.

Mei Ling, 34, sends her adviser a screenshot. She had asked an AI tool, “I’m 34. How much insurance do I need?” The answer was tidy: a list of factors, a rule of thumb, and a line suggesting she “consider speaking to a professional”.

Her adviser thanks her and asks what else she told the AI. Nothing else, it turns out. The AI didn’t know that she has a two-year-old, a home loan shared with her husband, and cover through her employer that would end if she changed jobs, which she is thinking of doing next year.

None of that made the AI’s answer wrong as general information. It made it incomplete for her. The adviser walks her through those three facts, explains what each one changes, and books a proper review. Mei Ling leaves knowing why the generic answer wasn’t enough, and with a better question to ask next time.

What not to say

• “Don’t trust ChatGPT.” It sounds as if you feel threatened, and they’ll keep using it anyway.
• “AI is always wrong.” It isn’t, and they can test that for themselves.
• “Yes, that looks right,” before you’ve checked. Once you agree with it, you own it.

Where you add what AI can’t

Judgment. AI can list trade-offs. It can’t weigh them against what this client values, or notice the worry they didn’t say out loud.

Context. You know their history, their family and how their plans have changed over the years. An AI tool starts from zero each time unless they tell it everything, and most people don’t.

Accountability. You are an appointed representative, named on MAS’s public register, and you answer for your advice. A general AI chat tool is none of those things. MAS’s consumer site MoneySense makes the wider point: when you deal with someone outside MAS regulation, you give up the protection its rules provide.

Clients sense this. In the same MDRT survey, the top reason people gave for not acting on AI advice was the lack of human oversight (45%). And half of those working with an adviser said they would rather meet face to face when markets are volatile.

Two prompts worth saving

For your client. Suggest they add this to the end of any money question they ask an AI tool:

Before you answer, list the assumptions you are making about my situation, the facts you would need to give me a personal answer, and which parts of your answer I should check with a licensed adviser or an official source.

It won’t turn the AI into an adviser, but it makes the gaps visible, and it often sends you a better-prepared client.

For you. When a client sends you an AI answer, remove their name and any personal details first, use only a tool your firm allows, then paste the answer in with this:

Here is an answer an AI tool gave someone about a money question. List each factual claim it makes, mark the ones that depend on Singapore rules or figures that may have changed, and list what it assumed about the person.

You still check every claim yourself. The prompt just gives you a faster starting list. For what to remove before you paste anything into AI, see Is my client data safe with AI?

Frequently asked questions

Related reading

• Forget the hype: practical ideas AI can help (not replace) Singapore financial advisers
• Think of AI like a super-smart assistant, not a robot overlord
• AI for FA: your essential guide

Sources

• MDRT, Singaporeans increasingly use AI to manage their finances and make financial decisions, July 2026. Online survey by Opinium of 2,000 adults in Singapore, 6 to 9 April 2026.
• MoneySense, Dealing with unregulated persons.
• UKTN, Consumer group warns against trusting AI financial advice, 18 November 2025.