ChatGPT Says You Can Afford a Condo — We Checked the Numbers

"So I asked ChatGPT whether we can afford a condo in Singapore — and it said yes."

That sentence is becoming a familiar opening line in Singapore's property conversations. Ask the chatbot for a budget, a neighbourhood, or a verdict on "BTO vs resale vs condo," and it replies with clean, confident prose. It quotes prices. It names districts. It even does rough loan maths. For a generation that grew up Googling first and asking questions later, an AI that produces a detailed answer in seconds feels like a superpower.

Then you actually look at the numbers.

The problem is not that ChatGPT is stupid about property. It is that property in Singapore is a calculation with a hundred moving parts: ABSD tiers, buyer's stamp duty, TDSR and MSR limits, grants that change with your income and marital status, lease decay, ethnic quotas, MOP rules, and transaction prices that move faster than the chatbot's training data. ChatGPT gives you an answer that sounds like a property agent who hasn't checked the caveats for six months. Confident, plausible — and sometimes wrong by six figures.

We decided to test exactly how wrong. In this article, we put the question "can you afford a condo in Singapore?" to ChatGPT with realistic young-buyer profiles, then checked its advice against actual rules, real transaction patterns, and the kind of per-project data that Hiva uses for pricing analysis. The results were illuminating — and a little alarming.

Why AI Property Advice Deserves a Reality Check

Singaporeans between 25 and 40 are inching toward some of the biggest financial decisions of their lives in a uniquely complicated market. A BTO is effectively a subsidised lottery ticket with a four-year delivery window. A resale HDB comes with grants, income ceilings, and resale restrictions. An Executive Condominium (EC) is a weird hybrid that behaves like a private condo after ten years. And a private condo — the thing ChatGPT cheerfully tells you that you can afford — sits on top of a scaffolding of cooling measures, stamp duties, and loan caps that change whenever the Government thinks prices are overheating.

Against that backdrop, the allure of asking an AI for a shortcut is obvious. The chatbot can summarise the difference between leasehold and freehold in seconds. It knows that District 9 is "prime" and that Tanjong Pagar is technically in the Central Business District. It will politely generate a checklist with bullet points, as though it has personally helped dozens of clients buy homes.

What it cannot do is see reality. ChatGPT does not have live access to URA caveats or HDB transacted resale prices. It cannot pull up the most recent median price for a specific condo project, compare its last five transactions, or adjust for whether you are buying a high-floor unit facing a park. It is working from patterns, not data. And when a technology that is 90% confident meets a market that is 100% specific, the result is advice that sounds brilliant — until you subtract the taxes.

What We Asked, and How We Checked It

To test the quality of AI-generated property advice, we built a set of young Singaporean buyer personas and asked each one as a ChatGPT prompt. The personas were deliberately representative of the Hiva audience: first-timer Singaporeans and PRs in their late twenties and thirties, earning between $5,000 and $12,000 a month, with CPF savings and cash reserves ranging from modest to serious.

Persona	Age(s)	Gross Monthly Income	Combined CPF + Cash	Situation
Priya	30	$6,500	$90,000	Single SC, wants a 2-bedder resale near parents in Woodlands
Marcus & Lynn	29 & 31	$9,200	$140,000	Married SCs, first-timers, open to BTO, EC, or resale
Daniel	27	$7,800	$110,000	SC planning to marry in 2 years; weighing a 1-bedder condo vs BTO
Wei Jie & Fatimah	33 & 32	$13,500	$260,000	Married PRs and SC; want a 2-bedder condo in OCR/RCR
Aisha	35	$8,300	$180,000	Divorced SC with one child; needs a place fast, resale preferred
Jon	26	$5,400	$60,000	Fresh graduate curious whether he can "get into property early"

Each persona was asked a variation of the same core question — "Can we afford a condo in Singapore on our income and savings?" — along with follow-up prompts asking for specific estate or project recommendations within budget. The outputs were then checked against four things: current MAS loan rules (TDSR and MSR), IRAS stamp duty schedules, HDB grant and eligibility criteria, and recent transaction data visible through platforms like Hiva.

What emerged were three broad categories of failure. ChatGPT was sometimes wrong on price, sometimes wrong on rules, and occasionally wrong on both — while still writing its answer in complete, authoritative sentences.

The $100,000 Confidence Gap: Where ChatGPT Misjudges Price

The most common failure was also the most expensive: ChatGPT does not know what anything costs today.

ChatGPT's knowledge has a training cutoff. It cannot browse the latest URA real estate statistics or HDB resale transactions in real time. This means its mental model of Singapore's price landscape is anchored to the past. And Singapore's property market has not been kind to stale anchors. From 2021 to 2024, resale HDB prices climbed roughly 25% cumulatively after a post-pandemic surge, while new condo launches in the Outside Central Region (OCR) pushed past psychological milestones. If ChatGPT's price memory is eighteen months old, its recommendation could be understated by a substantial margin.

Consider how the bot reasons. Ask it, "Where can I find a 2-bedroom condo under $1.3 million in Singapore?" and it will likely list a handful of district names, from District 15 along the East Coast to District 19 in Serangoon, with a caveat saying prices "can vary." That is true. But it does not tell you that the actual price varies by project, by lease start year, by floor level, and by whether the seller is testing the market. A district average is not an address. The bot often translates "median district price" into "you can probably buy there," which is a bit like telling someone they can afford a meal at a restaurant because the average dish on the menu costs $15 — then watching them choke on the $45 special.

The more insidious version of this error happens when ChatGPT is asked to plan a budget. It will cheerfully construct a scenario:

"Assuming a property price of $1,150,000 and a 25% downpayment of $287,500..."

That sounds responsible. But it inevitably underestimates the cash required to actually complete a purchase. A condo purchase in Singapore does not simply cost the purchase price. It costs the purchase price plus buyer's stamp duty, plus legal fees, plus (for some buyers) Additional Buyer's Stamp Duty (ABSD), plus a valuation fee, plus moving costs, plus the property tax you will prepay, plus the fact that your bank requires part of the downpayment in cold, hard cash — not CPF.

ChatGPT knows these categories exist. It simply fails to apply them rigorously to your specific case. In several of our tests, the bot's "affordable budget" for a persona was roughly $80,000 to $150,000 higher than the amount they could actually deploy after stamp duty, legal fees, and the mandatory cash component were taken into account. That gap is not a rounding error. It is the difference between completing your purchase and having your sale fall through.

Buyer's Stamp Duty on a Private Condo (SGD)

The bar chart above uses the official IRAS buyer's stamp duty schedule for residential properties as of the latest revision. Notice the shape of the curve. On a $1.3 million condo — a realistic price for a move-in-ready 2-bedder in many OCR and RCR projects — the stamp duty alone is $36,600. Add legal fees, typically between $2,500 and $3,500, plus disbursements, and you are already close to $40,000 of invisible cost before you have paid a single dollar toward the property itself.

ChatGPT will rarely volunteer this. It will happily tell you whether a price is "affordable" without noticing that your $60,000 cash buffer has just been eaten by taxes.

The Rules Are Not Suggestions: MSR, TDSR, and the Loan Limit That ChatGPT Forgets

The second category of errors is more dangerous because it is harder to spot. ChatGPT will often calculate a monthly mortgage payment based on an assumed interest rate and tenure — but it frequently ignores the regulatory constraints that determine how much the bank will actually lend you.

There are two numbers that matter above all others in Singapore's property loan system:

TDSR (Total Debt Servicing Ratio): Your total monthly debt obligations — including the property loan, car loans, credit card debt, personal loans, and student loans — cannot exceed 55% of your gross monthly income.
MSR (Mortgage Servicing Ratio): If you are buying an HDB flat or an Executive Condominium, your monthly property loan instalment alone cannot exceed 30% of your gross monthly income.

These are not guidelines from a bank's risk department. They are regulatory rules administered by the Monetary Authority of Singapore and HDB. A chatbot that ignores them is not giving you conservative advice; it is giving you fantasy advice.

To illustrate, take a persona earning $8,000 a month.

Maximum Monthly Property Instalment by Gross Income

The line above assumes no other debts at all. That is already an optimistic assumption. If you have a car loan of $800 a month, the property instalment cap under TDSR drops accordingly. And if the property is an HDB flat or EC, MSR cuts the cap even more sharply.

Let us run a concrete example. A 30-year loan of $1,000,000 at an assumed interest rate of 3.5% per annum would carry a monthly instalment of roughly $4,480. Under TDSR, a buyer would need a gross monthly income of at least $8,150 to service it assuming zero other debt. Under MSR, because the 30% cap applies, the same loan requires at least $14,900 a month — because only $4,470 of your income can go toward the mortgage. ChatGPT typically applies TDSR logic to HDB purchases, fails to apply MSR at all, and then wonders why the bank rejects the loan.

In one of our persona tests, the bot estimated Marcus and Lynn could service a $750,000 HDB resale loan on their combined $9,200 monthly income. Under MSR, their maximum monthly instalment is $2,760. At a 3.8% rate over 25 years, that supports a loan of roughly $540,000 — not $750,000. ChatGPT was recommending a home loan that the regulations would never allow. That is not an immaterial error. It changes the entire universe of properties they can consider.

Why this matters for your next ChatGPT prompt

If you do ask an AI for property budget advice, force it into the correct framework. Tell it the property type. Tell it your gross income, your age, your other debts, and the loan tenure. Then ask it to calculate under "MSR at 30%" or "TDSR at 55%" explicitly. The chatbot is better at arithmetic than it is at remembering which rule applies to which property type. You must supply the regulatory context.

ABSD, Citizenship, and the Bot That Forgot You Are a Foreigner

Additional Buyer's Stamp Duty (ABSD) is one of Singapore's most powerful cooling measures — and one of ChatGPT's most predictable blind spots.

ABSD is an additional tax imposed on top of the standard buyer's stamp duty. As of the latest round of cooling measures announced in April 2023:

Singapore Citizens buying their first residential property pay 0% ABSD. The second residential property incurs 20%, and the third or later incurs 30%.
Singapore Permanent Residents face 5% ABSD on their first property, 30% on their second, and 35% on subsequent properties.
Foreigners buying any residential property in Singapore pay 60% ABSD.
Entities such as companies pay 65%.

Most ChatGPT users who type "can I buy a condo in Singapore" do not proactively declare their citizenship in the first prompt. The chatbot, left to its own devices, will assume the friendliest possible scenario — usually a Singapore Citizen buying a first home at 0% ABSD. If you are a PR buying your first condo, the 5% ABSD on a $1.5 million home adds $75,000 that the bot completely forgot. If you are a foreigner, the 60% ABSD turns a $1.5 million purchase into a $900,000 tax bill on top of everything else. There is no world in which that is a rounding error.

ABSD Rates for Residential Property (%)

The chart above shows the ABSD schedule clearly. Notice how steeply the rate climbs for foreigners and for second-property buyers. The difference between 0% and 5% — for a PR buying a first home — is not trivial. The difference between 5% and 60% is the difference between owning a condo and renting for the rest of your life.

ChatGPT does not merely forget this; it often fails to ask the question that would remind it. In our tests, we had to explicitly prompt the bot to "assume the buyer is a Singapore Permanent Resident" before it adjusted its answer. In its default state, it optimistically assumed the buyer was a citizen with zero ABSD exposure.

Property type matters too

There are other structural filters ChatGPT tends to miss:

BTO flats are only available to Singapore Citizens (with at least one applicant being a citizen and another being a citizen or PR). PR couples cannot apply for BTO.
Singles who are Singapore Citizens can only buy a 2-room Flexi BTO in non-mature estates before age 35, though resale restrictions apply.
Resale HDB flats have an ethnic integration policy (EIP) quota at the block and neighbourhood level that can delay your purchase in certain estates.
Private condos are open to foreigners (subject to ABSD and approval for certain landed property), but the leasehold vs freehold question carries implications ChatGPT can summarise but not quantify for your specific timeline.

The chatbot treats eligibility as a uniform concept. In reality, your citizenship, marital status, age, family nucleus, and even the ethnic composition of the block you want to buy into all shape what you can purchase.

Grants, Subsidies, and the "New Flat" Illusion

The fourth major failure is the opposite of forgetting the rules — it is over-remembering the subsidies.

Ask ChatGPT about grants for HDB resale flats and it will confidently tell you about the Enhanced CPF Housing Grant (EHG), the family grant, and the proximity grant for families living near their parents. That part is fine. But then it will try to apply those grants to the wrong scenario, or stack them as though every grant applies simultaneously, or quote amounts that have been revised.

The reality is that HDB grants are means-tested, family-nucleus-tested, and property-type-specific. The maximum Enhanced CPF Housing Grant for eligible first-timer families can go up to $80,000 for lower-income households, but the amount tapers sharply as household income rises. A couple earning $9,000 a month may receive less than half of that maximum. In some recent policy revisions, the income ceilings and grant tiers were also refreshed — meaning the exact figure ChatGPT quotes from its training data may no longer reflect what HDB will actually disburse.

More importantly, ChatGPT does not understand the trade-off between grants and restrictions. A heavily subsidised BTO or Plus-model flat comes with a 10-year Minimum Occupation Period (MOP), resale restrictions, and subsidy clawback conditions. The chatbot might calculate that a BTO is cheaper — and it is, numerically — without explaining that the cheaper subsidised flat is also the one you cannot sell or rent out freely for a decade. For a 27-year-old who may want to migrate, marry, or upsize within seven years, that restriction is a cost that does not appear on any grant form.

When ChatGPT Is Actually Useful

For fairness, we should note where the AI performed well. ChatGPT is an excellent glossary. It can explain the difference between leasehold and freehold, define ABSD, summarise the TDSR framework, and tell you that District 10 contains Bukit Timah and that District 15 runs along the East Coast. It is a good starting point for understanding terminology — provided you verify every number before building a plan around it.

ChatGPT is also skilled at generating checklists. Ask it "what do I need to prepare before buying a resale HDB flat?" and it will produce a plausible, and largely accurate, list: obtain an HFE letter, check your CPF grant eligibility, get an Option to Purchase, arrange a valuation, apply for a loan, and so on. The structure is sound. It is the details that drift.

Task	Use ChatGPT for	Do NOT trust it for
Understanding terms	Explaining freehold vs leasehold, MOP, ABSD	Applying a specific ABSD rate to your citizenship and property count
Budget planning	Listing cost categories (stamp duty, legal fees, moving costs)	Producing the actual cash you need for a specific project
Loan maths	Explaining what TDSR and MSR mean	Knowing which ratio applies to your property type
Picking areas	Naming districts with certain attributes	Saying which estate fits your budget today
Market direction	Summarising historical policy shifts	Predicting whether prices will fall next quarter

In one test, ChatGPT correctly advised that a 30-year-old single citizen earning $7,800 should not rush into a 1-bedder private condo as an "investment" without checking rental yield and ABSD implications for their next purchase. That is genuinely sensible advice. The bot is capable of reasoning well in the abstract. It only fails when the abstraction meets your real transaction.

The Affordability Framework ChatGPT Does Not Give You

You do not need to abandon AI. You need to surround it with a rigorous framework. The Hiva reality check for any "can I afford it?" question looks like this:

Work through those steps with your own numbers and the phrase "ChatGPT says I can afford it" becomes irrelevant. The market decides, not the language model.

A Worked Example: Marcus and Lynn, Revisited

Let us apply the framework to Marcus and Lynn, the married Singaporean couple from our test set with a combined income of $9,200 and combined savings plus CPF of $140,000. ChatGPT told them they could comfortably afford a resale 4-room HDB in the high $600,000s, or possibly stretch to a compact OCR condo.

Run the real numbers under MSR: at 30% of $9,200, their maximum monthly mortgage instalment is $2,760. Assume a 3.5% interest rate and 25-year loan. That supports a loan of roughly $548,000. Their $140,000 in CPF and cash can fund the difference, meaning a resale HDB in the mid-$600,000s is feasible — but only if they have no other significant debts. That part of ChatGPT's advice survives contact with reality.

The condo part, however, does not. A compact OCR 2-bedder might cost $1.1 million. Under TDSR, the maximum loan they can support is roughly $835,000, leaving a shortfall of nearly $265,000 to be covered by cash and CPF. Marcus and Lynn only have $140,000. ChatGPT did not run this arithmetic in a way that exposed the gap. It simply said "affordable" and moved on.

Notice what the framework reveals: the bottleneck is not income — it is the savings required for the downpayment and closing costs. A buyer earning more can borrow more, but the cash portion of the downpayment — at least 5% of the purchase price in cash, plus the rest of the 25% downpayment, plus BSD — is the wall most young buyers actually hit. ChatGPT rarely frames the problem this way because it does not have visibility into your CPF statements or bank account.

What We Learned About Asking AI for Property Advice

After running these personas through ChatGPT and checking its outputs against the actual regulatory and market data, three lessons stand out.

First, treat every price ChatGPT mentions as a hypothesis, not a fact. The chatbot's training data is stale by design. Property prices in Singapore have moved rapidly in recent years, and any AI recommendation based on district averages from a year ago needs to be checked against live transaction data for the specific project or estate you are considering.

Second, always supply the constraints before asking the question. Citizenship, PR status, number of residential properties owned, gross income, other debts, age, property type, intended occupation period — feed all of it to ChatGPT in your prompt. The bot will still make errors, but you will have reduced the surface area for them.

Third, never let an AI's confidence substitute for a regulatory calculation. The MSR, TDSR, and ABSD rules are mathematically precise. A bank or HDB will not bend them because ChatGPT wrote a compelling paragraph. Run the numbers yourself using the official frameworks.

Food for Thought

Before you close this tab, consider these questions — for yourself, or the next time you open a chat window:

What is your actual "all-in" price? If you plan to buy a $1.2 million condo, can you state — without opening a calculator — the total stamp duty, legal fees, and cash downpayment required? If not, you are not ready to trust a chatbot's budget.
Does your AI prompt include your other debts? Many young Singaporeans carry car loans, credit card instalments, or education loans. These erode your TDSR headroom silently. Did you tell ChatGPT about them? Did it even ask?
What is your property's exit plan? ChatGPT can tell you whether a unit is affordable. It cannot tell you whether you will still want to own it in ten years. If you plan to upgrade, migrate, or rent it out later, have you checked the MOP, the resale restrictions, and the ABSD implications of your next purchase?
Is your advice based on a transaction or an average? When someone — human or machine — tells you a neighbourhood is affordable, ask them which project, which block, which floor, and which lease commencement date. If they cannot answer, they are not giving you property advice. They are giving you a generalisation.
Who is responsible when the numbers are wrong? ChatGPT will not refund your lost downpayment if its advice miscalculates your stamp duty. The ultimate check has to be yours.