You will be able to have an assistant plan what you need to find out about an investment, so the decision to buy stays with you.
Farhan lives in Singapore and has savings sitting idle. For weeks, a colleague has been talking up a REIT listed on SGX. The colleague says it pays out regularly and that its yield is higher than a fixed deposit's.
Farhan types "Is this REIT a good investment?" into an AI assistant and gets back four confident paragraphs. They describe strong properties, a reliable payout history and a yield of a little over six percent (an invented example, and possibly a number the assistant invented too). They close with the line "could suit income-seeking investors, but do your own research." Farhan has no way to tell the true statements apart from the ones that only held two years ago, or from the ones with no source at all. He also still has no idea what his own research is supposed to involve.
A verdict question is any question about whether an investment is good or whether you should buy it. Questions of this kind open the door to the three failures described in Lesson 1.3, "Where it fails": invented numbers, outdated facts and the wrong country.
To reach a verdict, the assistant has to supply facts: the yield, how much debt there is, how the price has moved, and what the properties actually are. About any single REIT or fund, it has read fairly little, and a lot of what it has read is old. Where its material runs thin, it puts in numbers that sound right but may have no basis. It then presents them inside an opinion, and the opinion makes them look more dependable than they are.
The judgement has a second problem. Whether an investment is good depends on your situation: what the money is for, when you'll need it, how large a drop you can live with, and what else you already hold. The assistant knows none of these things and has no obligation to ask. A verdict question therefore fails on two counts. The facts are unreliable, and the judgement isn't about the person asking. Lesson 7.3, "AI output versus licensed financial advice", comes back to why AI can't decide what is good for you.
The change is in the question. Replace "should I buy this?" with "what should I find out before deciding?" When you ask it this way, the assistant works as a planner of questions and no longer acts as a source of answers. Planning questions is the fourth of the jobs in Lesson 1.2, "Four money jobs AI does well".
A prompt that works does five things. It says you live in Singapore. It names the kind of investment and the exchange or market it's listed on. It says outright that you aren't asking whether to buy. It asks for the questions grouped by topic. It also instructs the assistant to leave out any figures or opinions about this particular investment.
Farhan tries again with a prompt that covers all five. He says he lives in Singapore, is looking at an SGX-listed REIT, and is not asking whether he should buy it. He asks for the questions he ought to answer before he decides, sorted by topic, with no figures or opinions about this REIT. The reply contains twenty-two questions. Some had never occurred to him, such as how much of the REIT's income depends on its largest tenants and when its loans need refinancing. Others are about things that don't matter to him. Lists from an AI assistant usually come out like this: helpful questions you wouldn't have reached alone, mixed with ones that don't fit your situation.
The same method works for a unit trust, an ETF, a single stock or a bond, as long as the prompt names which one it is. Each type calls for different questions. With a fund, cost and the index it tracks come first. With a REIT, the priorities are debt and rental income. With a bond, the list opens with who is borrowing and what happens if they can't repay.
Before you remove anything, make sure the list touches all five of these areas:
What it holds: what your money really owns. That could be a fund's holdings and the index it follows, a REIT's buildings and tenants, or the business behind a share. What it costs: each fee to buy, hold and sell, and any ongoing charge taken out of the fund itself. Fund costs are covered in depth in "Unit trusts, robo-advisors and managed money compared". What can go wrong: the main risks, such as prices falling, payouts being cut, interest rates rising, a borrower failing to pay, or the currency moving against you. How you get your money out: whether you can sell on any trading day, how long the money takes to reach you, and any penalty or loss for leaving early. Who regulates it and where you'd complain: whether the provider holds a licence in Singapore, and where the product is listed or registered.
If one area has no questions, send the assistant a follow-up asking for questions on that area alone. Add them to the list before you start cutting.
A list of around twenty questions is more than you can answer properly. Bring it down to ten. For each question, picture getting an answer that surprises you. If that surprise would make you invest differently, keep the question. If you'd go ahead the same way regardless, cross it out. Go through every question like this until ten are left.
Store the ten somewhere you'll find them later. Answers shift every quarter as prices, fees and holdings change, but questions change far more slowly. A REIT question list you write today will still be the right one next year, and for the next REIT you consider. Reuse it for later decisions and for investments of the same type. Lesson 4.2, "Where the real answers live", matches each question to the document that answers it.
Choose one investment you're curious about right now. Write a prompt asking for questions rather than a verdict, check the reply against the five areas, and cut the list down to the ten questions that matter to you.
Pick one investment you are curious about and ask an assistant for a list of research questions. Cut the list down to the ten that matter most to you.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).