Buying Off-Plan

What AI Gets Wrong About Dubai Property, With Examples

A client was told a Dubai Hills townhouse costs AED 3.5M. It is nearer 4.9M. Where language models fail on property, and how to use them properly.

I should declare an interest immediately: this site is drafted with AI assistance, and it says so in the byline of every article. So this is not an argument that the technology is useless. It is an argument about where it breaks, based on three conversations I have had this year.

Failure one: the price was two years stale

A client came to me certain he could buy a three-bedroom townhouse in Dubai Hills Estate for AED 3.5 million, because that is what he had been told. The realistic figure was closer to AED 4.8–4.9 million.

He had not been given a bad deal. He had been given a bad expectation, which is worse, because he spent weeks rejecting properties that were correctly priced. He believed every agent showing him a 4.8 was trying it on.

This is the most common failure and the most corrosive. A model has no live price feed. It produces a number that reads like a fact because it is stated like one, with no uncertainty attached. A human adviser saying “roughly three and a half, but check, that is from memory” transmits doubt. The model does not.

How to avoid it: never accept a price without a source and a date. Ask for the actual transactions — DLD records what things sold for, and any competent agent can pull comparables for a specific community in ten minutes.

Failure two: the recommendation expired

Ask a model for the best place to invest in Dubai and JVC comes back with remarkable consistency.

Three years ago that was defensible. JVC offered among the strongest gross yields in the city at a genuinely low entry price. A great deal was written saying so, and that writing is what the model learned from.

What it cannot see is what happened next. The same pipeline that made JVC affordable kept running. Fitch has flagged oversupply risk across that band of the market. JVC still shows roughly 7.2% gross on Bayut’s H1 2026 data — the number is real — but the forward risk profile is not what it was when those articles were written.

The deeper problem: a model reproduces the centre of gravity of what has been written. In property, the volume of writing about an area peaks after it has already performed. So the consensus you get is reliably a lagging indicator, and the areas that are genuinely early are the ones with almost nothing written about them yet. You are being handed yesterday’s trade in confident prose.

Failure three: technically correct, practically useless

Ask for the highest rental yield in Dubai and you will be pointed at International City, at something like 9–10% gross. That is arithmetically true.

Here is what does not come with it. Much of that stock is old and tired. A significant portion functions as workforce accommodation. Maintenance runs high and service quality is inconsistent. Tenant turnover, arrears and management overhead are materially worse than in mid-market Dubai. By the time you have paid for all of that, the net is nowhere near the headline, and you have bought yourself a second job.

The model was not lying. It answered the question asked — highest gross yield — and gross yield genuinely is highest there. It simply had no way to volunteer that this is the one number in property most likely to mislead you, and that the gap between gross and net is widest exactly where gross looks best.

This is the failure I find most instructive, because there is nothing to correct. The answer was right. The framing was missing, and the framing was the whole thing.

What it is actually good for

I use these tools daily and I would be worse at my job without them. Where they earn their place:

• Process and definitions. What an Oqood is, how escrow works under Law 8/2007, what a grace period clause does, what fees apply at transfer. Stable, well-documented, unlikely to have changed.
• Structuring your thinking. “What should I ask before buying off-plan” produces a genuinely good checklist.
• First-pass drafting, which is exactly what happens on this site — followed by a licensed human checking every figure against a named source and removing what does not survive.
• Explaining something you did not follow. Paste the clause, ask what it means, then verify the answer with someone who carries liability for being wrong.

Where they do not belong: current pricing, area selection, developer assessment, yield expectations, and anything where the answer determines what you buy.

The distinction that matters

There is a real difference between AI as a drafting tool with a human accountable for the output, and AI as the adviser.

On this site the first thing is true, and disclosed. Every number carries a source and a date. Every area call is mine, including the ones that cost me commission. When I do not know something, the article says so rather than producing a confident estimate.

The failure mode in this article is the second thing — treating a fluent answer as a researched one. Fluency is not accuracy, and property is a field where a plausible wrong number costs six figures.

Bring me the answer you got and I will tell you which parts hold up. Genuinely — that is a useful conversation, and I would rather correct a number than watch someone act on it.

This article is general information about the Dubai property market, not financial, legal or investment advice. Figures change and unofficial estimates are labelled as such — verify current numbers with the Dubai Land Department or a licensed professional before committing funds.