ChatGPT and Claude prompts for Dubai real estate agents (copy and paste, with a team rule sheet)

The rule sheet first: what agents may paste and what must be checked

Most prompt packs for Dubai agents are lists of prompts. What they leave out is the policy. I would adopt this one-page sheet before anyone uses a prompt.

1. Data in, not memory. Any prompt that touches a number takes a pasted block of facts and ends with "use only the data above; if a figure is missing write MISSING; list every number you used".
2. Green, amber, red content. Green: tone rewrites, subject lines, translation drafts. Amber: listing copy from verified fields, area summaries from pasted data. Red: prices, fees, yields, ROI, legal or visa statements, permit numbers. For red items AI may only format text a human supplied from a source.
3. No client personal data in public tools. Strip names, phone numbers, Emirates ID and passport numbers, and deal terms before pasting.
4. Every output is a draft. A named person reads it before it goes out.
5. Ads need a permit. AI cannot generate or check a permit number.
6. Arabic and other translations get a native read before publishing.
7. Bad outputs get reported so the shared prompt library improves.

The owner signs off the sheet, keeps it in the team's shared folder and checks it quarterly.

How a good prompt is built

Every prompt below follows the same shape:

• Role: "You are a copywriter for a Dubai real estate brokerage."
• Data: a block of verified facts in square brackets for the agent to fill.
• Format: length, structure, language.
• Constraints: what not to do. The most important constraint is "do not add any fact that is not in the data".

Replace the bracketed items. Keep the constraints.

Listing copy prompts for Bayut and Property Finder

Bayut and Property Finder are the two biggest UAE property portals, where agents post listings to reach buyers and renters. Each has its own length limits and rules, which change, so check the current requirement in the portal's agent area instead of trusting a number in a prompt pack. For a longer treatment see AI real estate listing descriptions.

You are a copywriter for a Dubai real estate brokerage. Write a listing description for [Bayut / Property Finder]. Use only these facts: [type, bedrooms, size in sq ft, community and building, floor, view, furnishing, parking, status (ready or off-plan), asking price, any verified features]. Tone: clear and factual, no exaggeration. Length: [check the portal limit and enter it]. Do not mention any fee, yield, investment return, visa eligibility or price comparison. Do not add any amenity not listed. If a fact is missing, write MISSING. Finish with a list of every fact you used.

Tick the final list against your unit sheet. An invented amenity is the most common error.

An ad also needs a permit. DLD (Dubai Land Department, the emirate's property registry) and RERA (the Real Estate Regulatory Agency, its regulator) require a Trakheesi permit, the official advertising permit, before real estate marketing, with the permit number in every ad. Per Bayut's agent portal, applying needs a DLD Form A (the signed listing agreement between owner and broker) or an NOC (no-objection certificate) from the legal owner, and an ad preview. AI cannot create or verify the number. The permit number goes into the final ad by hand.

Viewing follow-up and re-engagement prompts

Write a WhatsApp message of under 80 words to a buyer after a viewing. Facts: viewing was [date] at [unit and building]; the buyer liked [two things they said]; their concern was [concern]; the next step is [step and date]. Tone: warm, professional, no pressure. Do not state any price, fee or return. Do not invent anything the buyer said.

Write a short message to a lead who went quiet [number] weeks ago after asking about [area and type]. Offer one useful thing: [a new listing / a price change / a viewing slot]. Make it easy to say no. Do not use urgency or fear of missing out. One question only.

Make sure marketing messages only go to people who have asked for them. The rules on consent are in AI limits and consent risks for UAE real estate teams, and the speed side is in speed to lead on WhatsApp.

Off-plan comparison summary prompts

Off-plan means buying a property that is still being built. A comparison summary is useful, provided the numbers come from the developer's documents.

You are an analyst helping a Dubai agent. Compare these two off-plan projects using only the table below: [project, developer, location, unit types, price from, payment plan, handover date, service charge if stated, source and date for each row]. Write a plain-language comparison of 150 words for a buyer. Highlight differences in payment plan and handover. Do not rank them, do not predict resale value, rent or returns, and do not state anything not in the table. Flag any missing row with MISSING. Finish with a list of every number used and its source row.

The agent checks the table against the developer's price list and the sales and purchase agreement before the summary goes anywhere.

Investor update email prompts

Write an update email to an existing investor client. Facts: [property, purchase date, current rental status, rent per the tenancy contract, any service charge changes, recent comparable sales with source and date]. Tone: clear, calm, non-promotional. Do not forecast prices or give a return figure unless I have supplied it under "Approved figures". Approved figures: [figures with source]. Include a one-line "what I'd watch next quarter".

For a monthly investor note, see AI investor communication for Dubai brokerages. Fee figures such as DLD registration costs should be keyed in from the official page; I cover them in my DLD fees calculator article.

Area guide prompts that use only data you paste

Write a 300-word area guide for [community] for first-time buyers. Use only these facts: [facts you have checked, with sources]. Describe it in plain language, for example who it suits and what a typical day looks like. Do not give prices, yields, fees, school ratings or distance times unless they appear above. If I have not given a fact, write MISSING instead of filling it in.

General chatbots produce confident-sounding area copy from memory: amenities that closed, metro stations that do not exist, a school that moved. This prompt prevents that.

What AI must never state without a checked source

• Fees and taxes: keep the fee as a placeholder in the prompt and fill it from the official page. Prompt packs often repeat a fee percentage that nobody has checked. Do not copy those.
• Yields and ROI: only figures you have calculated from rent and price data you can show. For AI price estimates, see how accurate AI property valuation is in Dubai.
• Visa or residency eligibility: check with the relevant authority.
• Legal or contract points: confirm with a lawyer.
• Permit numbers: never generated.
• Market claims: "prices will rise" is a forecast, not a fact.

The test: if a client acted on this sentence and it were wrong, would it cost them money? If yes, a human sources it.

Keeping client data out of public tools

Before pasting, remove names, phone numbers, Emirates ID and passport numbers, addresses and deal terms. Replace with "Buyer A". Check each tool's data-use setting. For example, Claude's consumer privacy page says that on the Free, Pro and Max plans chats may be used to improve models if the user chooses to allow it, and that incognito chats are not used. Business products are covered on a separate page that I have not read. For ChatGPT, I could not read OpenAI's privacy pages, so check the current page yourself. Many guides state that free and paid personal plans train on data while business plans do not, but I cannot confirm that, and it can change. Pay for a business plan if the team uses AI daily, and confirm its terms in writing. Wider data duties are in PDPL and client data for Dubai brokerages.

How software helps you get more from these prompts, and what stays human

A pack of prompts in a document is fine to start. Over time, three things make it work better.

• A shared prompt library: one place where each prompt has an owner, a "last checked" date and an example of a good output. Agents report bad outputs there.
• Prompts built into your tools: a button in your CRM that fills the listing prompt from the unit record, so nobody retypes facts. That removes the main error source (retyping). I build this kind of connection for brokerages, and the data stays in your own system.
• A review step: drafts go to a queue and a named person approves them.

What stays human: choosing the facts, checking every number, adding the permit, and deciding whether the message should be sent at all. For the bigger picture, see AI for real estate, and for how to think about AI use in a business more widely, ChatGPT for business.

ChatGPT or Claude?

Both do these jobs. The choice is about workflow, not a winner. Pick the one your team will actually keep within the rule sheet, pay for the business tier, and set up the data-use options properly. Do not run half the team on each unless you need to.

Running a shared prompt library for the team

1. Create one document or sheet with columns: name, purpose, prompt, content class (green, amber, red), owner, last checked.
2. Start with the prompts above and prune to what agents really use.
3. Re-check every quarter, and whenever a portal, regulator or fee page changes.
4. Add one real example of a good output beside each prompt.
5. Have agents report any made-up fact immediately.

How would I approach this?

I would adopt the rule sheet before anything else, then give agents these prompts and nothing more for a month. I would keep every fee and return as a placeholder that is filled from a source, and I would check ten real outputs a week against the facts. If agents keep retyping unit details, that is the point to connect the prompts to your CRM, not before.