IN FOCUS: Is ChatGPT better than a financial adviser? I put it to the test
People are turning to AI chatbots for anything from writing emails to planning holidays. But can they be trusted with our finances?
(Illustration: CNA/Rafa Estrada)
This audio is generated by an AI tool.
SINGAPORE: A few months ago, a fellow journalist asked me: “Do you invest? Do you think you will be able to retire with a million dollars?”
Thinking back on the mild heart-stopping fluctuations I'd seen in my portfolio amid tariff uncertainties, I replied that I wasn’t sure.
He suggested I get some feedback from ChatGPT. Unfazed by the skeptical look I gave, he persisted. “Try feeding (it) your portfolio and see what it says.”
I was intrigued, and later found out he wasn’t alone. A family member had asked the artificial intelligence chatbot for ideas to recalibrate her investments and save for her first home. And a colleague described stock picks that he got from ChatGPT as “quite solid”.
A quick online search then threw up a long list of articles, blogs and forum threads on people testing the generative AI assistant for investment bets, and their tips on writing better prompts for better responses.
We already know ChatGPT and other chatbots can have human-like conversations and put together fancy presentation decks and holiday plans within seconds. But can we really rely on it for sound financial advice?
KNOWLEDGE VS ASSURANCE
Generative AI is a branch of AI that enables machines to generate text, images, music and other content based on the data they were trained on.
ChatGPT, launched by US-based OpenAI nearly three years ago, is the best known example and market leader of generative AI chatbots, amid the emergence of others such as Google’s Gemini and China’s DeepSeek.
One of its strengths lies in sifting through vast amounts of data and calculations to quickly churn out research. Added to that is the ability to convey findings in a human-like manner, including asking follow-up questions.
This can make the process feel like it's what a user needs, regardless of its accuracy, said Ms Chuin Ting Weber, chief executive officer of financial advisory MoneyOwl.
Sales executive Clarin Florentyna used ChatGPT when she picked up investing about two years ago.
The bot guided her through the basics, with accurate definitions of key terms such as the intrinsic value of a stock.
When asked for investment advice, ChatGPT suggested setting aside an emergency fund of three to six months, as well as how much to put into bonds, equities and other assets, based on Ms Florentyna’s then monthly income and target returns.
The presentation of such a breakdown made the advice feel more "personalised” than what’s available on finance websites, said the 28-year-old.
The willingness to turn to an AI chatbot for something as personal as money could be linked to the increasing normalisation of their everyday use, as well as confidence in the technology at large.
For one, “robo-advisers”, which tap machine-learning algorithms to propose investment portfolios for their users and automate rebalancing, have been around for years.
Communications professional Tan Jiunn Ngee was fresh from signing up with a robo-adviser last year, when he thought of using ChatGPT to reassess his investment-linked policy (ILP).
The policy, which he signed up for in 2019, had been in the red since the COVID-19 pandemic.
“I was inspired after exploring robo-investment platforms that offered automated rebalancing. That got me thinking: Why not replicate a similar process for my own ILP?”
Being able to challenge and talk through assumptions and scenarios with ChatGPT was productive and helped to clarify his thought process, said Mr Tan, a communications professional.
Its suggestions, which included increasing exposure to growth sectors like technology, and diversifying geographically, also seemed reasonable. “Even my financial adviser said it made sense, to some extent,” said the 40-year-old with a chuckle.
But Mr Tan ultimately did not take up ChatGPT's recommendations wholesale; and was more comfortable letting his financial adviser make the final call. His policy was eventually tweaked to better suit his preference for long-term safe returns.
“I would recommend using ChatGPT as a supportive tool, rather than a decision-maker ... because financial planning is just too highly personal,” he said.
Ms Florentyna meanwhile decided that ChatGPT would remain as a learning assistant for her, after spotting errors in its responses such as outdated stock prices and misquoted data from both the web and reports she had uploaded.
“It’s a great tool to bounce ideas. Even now, I still ask ChatGPT what I can do now that I’m older and I am planning for other life goals,” she said.
“AI can give me knowledge, but it doesn’t give me assurance yet.”
Experts whom CNA spoke to echoed the need to be vigilant, citing accuracy as a key concern given how ChatGPT is primarily trained on massive amounts of past data. This can potentially result in "hallucinations", where false information is produced.
MoneyOwl's Mrs Weber also flagged the risk of confirmation bias, noting how generative AI models' reliance on publicly available information can lead to them reinforcing popular or dominant views.
She cited media reports which posited a link between ChatGPT amplifying "meme stocks" popular on the Reddit forum and the latest rally for such viral stocks.
The quality and nature of ChatGPT's responses are also heavily influenced by user prompts, said experts.
“If they are not accurate or are flawed in some way, the output will be suboptimal, which could then lead to losses,” said Mr David Gerald, founder-CEO of the Securities Investors Association (Singapore) or SIAS, who urged investors to not solely rely on AI, and to always check with reliable sources before investing.
Mr Poon King Wang, chief strategy and design AI officer at the Singapore University of Technology and Design (SUTD), said it was important to provide the chatbot with clear instructions and contextual information, instead of simply using it like a search engine.
If the advice requires in-depth analysis, users should be using the chatbot's more advanced reasoning models, instead of the standard models which are better suited for straightforward tasks.
Latest data from OpenAI showed that only 7 per cent of free users were using the reasoning models.
"This suggests that a substantial number of people have yet to fully appreciate how to match the right model of Gen AI to the task to get the right performance," said Mr Poon.
He noted that for those already using such chatbots like search engines, turning to them for financial-related matters can feel like a “continuation of a habit”.
As money matters are often information-intensive and complicated, chatbots may provide users with “the relief from having something to navigate through it all”, he added.
“What some users forget is that just as not every stock analyst or financial planner or pundit's analysis is correct, not every response from Gen AI is and will be correct.”
Echoing that, Assistant professor of finance Aurobindo Ghosh from the Singapore Management of University (SMU) noted that chatbots' ability to offer scenario planning could lead investors to “start believing ChatGPT is a qualified financial advisor”.
“In reality, those results are all based on different analyses of past data or more likely what is available on the web,” he added.
“As we don’t take advice on our finances purely based on websites which are freely available, we should be careful taking advice from ChatGPT or other forms of generative AI.”
Daily Cuts - Would you trust AI to make investments for you?
AN EXPERIMENT
In any case, I decided to take up my colleague's suggestion and see if an AI chatbot held any answers as to whether my portfolio could afford me a comfortable retirement.
I only tested ChatGPT, as it was the default large language model for those I spoke to.
Mindful that the bot’s answers would only be as good as its prompts, I tried to be as detailed as I could in a breakdown of my holdings; while adding some context such as the amount of emergency funds I had.
ChatGPT noted that my portfolio was tilted towards cyclicals, which are industries or firms whose performance is directly tied to the broader economic cycle. These include banks and tech firms.
Ideas to improve my portfolio included adding more non-cyclical stocks like healthcare and consumer staples, as well as exchange-traded funds (ETFs) to smoothen volatility. The latter refers to baskets of various assets that trade like an individual stock.
I could also consider small allocations to markets I currently have no exposure to, such as Europe and emerging markets.
On hitting a million-dollar portfolio, ChatGPT asked me to increase my savings rate, stay invested through downturns and to reinvest my dividends.
I asked further about specific allocations, and when asked for its references, it said it was citing “established portfolio frameworks like core–satellite allocation”.
This is an investment strategy where broad-market ETFs form the core of one’s portfolio for consistent returns, and the rest is made up of stocks for growth or dividends.
Not too bad so far, I thought.
There was no fear of asking "stupid" questions, nor pressure of coming up with follow-up queries before time ran out - both of which are my worries at times when meeting financial advisers.
I asked more questions when I wanted to. Even if I ran out of questions, ChatGPT had prompts at the end of each response, such as whether I would like to see how different allocations would affect outcomes.
For me, it felt almost addictive to reply “yes” and watch tables filled with numbers appear on my screen within seconds.
At this initial stage, it was quite thrilling to feel like I had access to an endless trove of financial information, at least if I put aside the urge to fact check every line.
For the purpose of this article, I also consulted a human - MoneyOwl’s Mrs Weber - with the same set of questions.
She began by asking me a series of questions. These included the purpose of my investment portfolio, my risk appetite and my investment habits.
While Mrs Weber wasn’t able to generate spreadsheets of numbers instantly, her questions reminded me of the need to be disciplined and to keep in mind what my portfolio was for - retirement.
She then made the same observation as ChatGPT - that it could do with some tweaking to move away from cyclicals.
She suggested growing my allocation in broad-based ETFs such as one that tracks the MSCI World global stock market index, and keeping more speculative investments to about 20 per cent of my portfolio. This mirrored the core-satellite strategy suggested by ChatGPT.
If I really wanted to stick with cyclicals, she suggested I double down on an ETF I had and refrain from further stock-picking - that is, trying to outperform the market - to spare myself the agony of market volatility.
“Everyone wants to know how to beat the market. It's not impossible, but even professionals find it very hard to do so, much less ordinary investors,” Mrs Weber said.
Instead of being fixated on an arbitrary retirement target for my stock investments, she proposed I also grow the funds in my Central Provident Fund (CPF) savings account, to qualify for higher payouts when I retire.
Factoring in such other aspects gives a more comprehensive view of retirement financial health and shapes a more realistic gameplan, she said.
I liked how Mrs Weber’s suggestions took me back to the basics of investment and financial planning, while considering the local context.
The latter was something I had to purposefully ask ChatGPT about. It also did not factor in my age, income and other aspects such as risk appetite. Without such key information, the bot ran ahead on assumptions, until I stepped in to clarify.
It might be difficult for someone just starting out in investments or financial planning to recognise this gap.
I also asked ChatGPT for stock picks. Here, it was upfront about its limitations, such as how it did not have access to real-time market data and as such best used as a “research assistant”.
But if you persist, the bot complies and its replies were, in my view, a mixed bag.
For example, when asked for top bargain stocks in Singapore, it generated a list based on a local blog, which I hesitated to take at face value. I refined my prompt for ChatGPT to reference analyst reports and credible news sites instead, but not all the reports it subsequently cited were up to date.
By this time, what started out as fun began to feel like an information overload, which wasn't necessarily helping with the decision-making process. Analysis paralysis kicked in after awhile, and fact-checking made things seem even more onerous.
ChatGPT’s responses remained instantaneous, but it was beginning to feel like a conversation that had gone on for a little too long.
There is a risk of it churning out answers without more precise information, and you might not always get more nuanced advice unless you ask.
I brought this up with SMU's Asst Prof Ghosh, who said that to use ChatGPT well, one must first build up the necessary knowledge - a process that can be aided, but not replaced by AI.
Citing an example of asking ChatGPT for a list of under-valued stocks, he noted that a user would need to first be aware of various indicators that determine such a stock, and also be able to make sense of what is subsequently presented.
Planning and deciding one’s financial future is highly personal, Asst Prof Ghosh stressed.
“The answers to these questions, unfortunately, cannot be found in ChatGPT,” he said. “That can only be determined by looking at your own situation first, your needs and your goals.”
“You have to think it through, or go along that journey to discover yourself. Once you discover yourself, you are better off making better decisions.”
PERSONAL RESPONSIBILITY
Meanwhile, away from investors using ChatGPT on their own, at least two brokerages - Tiger Brokers and PhillipCapital - have rolled out chatbots interwoven with generative AI.
Integrated into their respective trading ecosystems with access to in-house financial data, these chatbots can answer user questions, break down financial concepts, as well as provide stock market research and advice in the blink of an eye.
Both firms said their tools were attempts to meet demand for instant insights, and that user take-up rates have been strong.
On how its offering differs from ChatGPT, PhillipCapital managing director Luke Lim said its Poems GPT was designed to incorporate some of its in-house data, among other things.
It “contextualises data, surfaces relevant trends, and highlights key considerations investors should weigh before acting”. That said, it does not provide direct buy or sell recommendations.
“This empowers investors to engage their trading representatives with sharper questions and make decisions more confidently, which is a step forward in democratising financial intelligence,” said Mr Lim.
At Tiger Brokers, its TigerAI tool started out in 2023 using OpenAI’s generative AI models before being integrated with the DeepSeek-R1 model last year. Users can retrieve stock insights and trading summaries, among others.
“What sets TigerAI apart is its deep interoperability with Tiger Brokers’ internal systems,” said Tiger Brokers Singapore’s CEO Ian Leong. “TigerAI draws directly from live data feeds, trading infrastructure and internal financial databases.”
That said, both platforms carry warnings of possibly inaccurate responses from generative AI. They also urge users to always verify the data provided.
Asked about the risk of hallucination, Mr Leong said: “Investors should use these insights as reference points alongside their own due diligence, to ensure a well-rounded and informed approach to their trading strategies.”
SMU's Asst Prof Ghosh said the popularity of generative AI models made it clear that businesses had to jump on the bandwagon. But such disclaimers suggest that even on industry platforms, investors still need to be discerning and exercise caution.
Personally, I think it's great to task ChatGPT with summarising analyst reports and to bounce ideas on investment options. It did provide some sensible feedback, just as it did with the two individuals I spoke to.
But I also won't be ignoring the usual caveats, along with what I gathered from my mini-investigation - including to not be carried away by slickly presented data and advice.
When I consulted ChatGPT on whether to sell a stock that was experiencing some weakness after a near 40 per cent rally, it presented me with two scenarios: Take profit given recent declines and analyst downgrades, or hold on if I believed in the company’s potential.
It was nothing I didn't already know, or that anyone else couldn't come up with.
I then decided to fall back on a rule I read about years ago, which calls for investors to sell a stock after a 20 to 25 per cent gain to lock in profits and avoid emotional trading.
I sold, and averted a further tumble in the stock after an earnings miss. And this was guided by consulting myself - or a decades-old investment maxim, if you will - as well as a huge dose of luck.
So yes, while I will still be using ChatGPT or the next big generative AI chatbot, I won’t be counting on machines alone to help me retire with a million bucks.
