Why ChatGPT and Claude Get Dubai Property Prices Wrong

An AI assistant misprices Dubai property when its evidence does not match the home in the question. The model is not careless. It simply reads the wrong numbers, or the right numbers mixed with the wrong ones.

Each gap below has a direct correction in Dubai real estate data. DLD records the price, size, floor, date, property type and bedrooms of each sale, and whether the sale was ready or off-plan. Ejari records the annual rent, the start date and whether a contract is new or a renewal.

The overview of AI for Dubai real estate explains how an assistant reads this data through a plugin.

A language model knows the Dubai market as it stood at its training cutoff. After that date it learns nothing new unless a tool feeds it fresh data. In a fast market, a year-old picture gives the wrong direction as well as the wrong level.

The PIX© residential index shows why. It rose between 10.6% and 13.7% a year in every month of 2025. It peaked at 219.98 in January 2026. By September 2026 it stood at 205.79, which is 6.45% below the peak and 3.96% below September 2025.

An assistant trained on 2025 text learned a market that rose by double digits every year. It has no way to know that prices turned in June 2026. Ask it whether now is a good time to sell, and it may reason from a trend that has already reversed. The PropertyIndex plugin reads the index monthly, so the assistant starts from the current level. Reading the direction by segment and community is covered in monitoring the Dubai property market with AI.

A dashboard or a scheduled market report needs the same data without a chat. The PIX© API is the right tool for that job. It returns DLD sales, Ejari rentals and the PIX© market index as JSON, and it is licensed for scheduled retrieval and backend integrations.

An asking price is what a seller hopes to get. A DLD sale price is what a buyer paid, recorded by the Dubai Land Department when the sale completed. The two numbers measure different things and often sit apart.

Web search pulls in listing pages because they rank well for property questions. A listing shows one seller's hope on one day, and it often stays up after the price drops or the home sells. When an assistant averages a set of listings, it reports the market as sellers would like it to be.

DLD sales remove that bias. Each sale carries the price, the size, the floor and the date, so the assistant can state a median price and a price per square foot that buyers actually paid.

A Dubai community average describes no single building. Dubai Marina holds dozens of towers of different ages, heights and specifications. The towers compared below differ in price per square foot by a factor of almost three.

DLD recorded 1,789 sales of ready apartments in Dubai Marina at a median of AED 1,892 per sqft between October 2025 and September 2026. LIV Marina sold at a median of AED 3,324 per sqft. Marina Pinnacle sold at AED 1,159. The community figure sits 43% below the LIV Marina median and 63% above the Marina Pinnacle median.

Some wrong answers come from evidence that is real but mixed. Two splits matter most in Dubai: off-plan against ready sales, and new rental contracts against renewals. Both are recorded on every transaction, so both can be separated.

A price without a period and a count cannot be checked. “Around AED 2,000 per sqft” could describe last month or three years ago. It could rest on hundreds of sales or on two.

Sample size changes how far a figure can be trusted. In the Dubai Marina chart above, the LIV Waterside median rests on 121 sales. The Marina Gate 1 median rests on 20. Both are useful, but one unusual sale moves the smaller median far more. A grounded answer shows the count so the reader can weigh it.

The PropertyIndex plugin returns the period and the count with every figure. It also returns the middle half of prices, which shows how wide the spread is. When the evidence for a building is too thin, the assistant can widen the period or step up to the sub-community and say that it did.

A grounded AI answer about Dubai property names its evidence. Four checks catch nearly every weak answer, whether it came from ChatGPT, Claude or a person.

• Count. The answer states how many sales or rental contracts sit behind each figure. A median from 121 sales and a median from 3 sales do not deserve the same weight.
• Period. The answer names the dates the evidence covers, such as October 2025 to September 2026. A figure with no period could be from any year.
• Source. The answer names Dubai Land Department sales or Ejari rental contracts. Listings, news articles and "market reports" are different numbers.
• Layout. The answer matches your home: the same building, property type, bedrooms and ready or off-plan basis. A community-wide figure is a starting point, not a price.

If an answer fails two or more checks, treat it as an estimate. Ask the assistant to restate it with the count, the period and the source. An assistant with the PropertyIndex plugin can do that. One without data access has no count or period to give.