TL;DR: Dubai's regulator now uses AI to watch property adverts, and it has publicly said it automatically modified 29% of the 279,000-plus ads it monitored. If your brokerage or portal uses generative AI to write listings, the safe design is to treat the Trakheesi permit as the source of truth and let AI work only on wording around verified facts. This article sets out how to build that, what to audit, and what to do this quarter. It is not legal advice; confirm specifics with your compliance counsel.
The regulator is now an AI reader of your listings
Most conversations about AI in Dubai real estate start with what a brokerage can generate: descriptions, translations, social captions, video voice-overs. The more important shift is on the other side of the table. The Dubai Land Department (DLD) says it launched an AI-powered Real Estate Advertising Governance Platform at GITEX 2024, and in April 2025 announced that it had monitored more than 279,000 adverts across Property Finder, Dubizzle and Bayut, with 29% of the monitored listings automatically modified using artificial intelligence (Dubai Land Department, 24 April 2025).
Two caveats are worth stating plainly. First, the DLD announcement does not explain what "automatically modified" means in practice, or how the system matches an advert to a permit. Second, outlets describe the time window differently: Arabian Business says "since its launch", while Gulf News says between October and the announcement (Arabian Business; Gulf News). We should not build a compliance programme on guesses about the regulator's model. We should build one on what is documented and assume it is consistent and machine-readable.
My point of view is simple. When a regulator reads listings with software, the winning strategy for a brokerage is not to generate more copy faster. It is to make every advert reconcilable, field by field, against a permit record. AI that writes freely works against that goal. AI that is constrained by it works for you.
What Trakheesi and Madmoun actually require
Trakheesi is the DLD/RERA system through which brokers obtain advertising permits. According to Gulf News, brokers apply for a permit through Trakheesi, which generates a unique ID for each listing, and commentary quoted in the same piece says every property advert needs a permit before it is posted, including on social media (Gulf News, 13 December 2024). The same article reports that details in an advert must be accurate to the property, and that a property can be listed only once by an authorised broker or company, with duplicate adverts banned.
Madmoun sits on top of that. The DLD describes it as a service accessible through Trakheesi, represented by a QR code issued for each real estate advertisement permit. From 24 April 2023, real estate companies were expected to show the QR code on print and audiovisual adverts, so customers can verify authenticity and RERA approval and see the property's authorised details, including whether it has been sold or rented (Dubai Land Department, 18 April 2023). The April 2025 announcement restates that companies are required to display the QR code on all adverts, visual or written.
Other rules reported in the press matter for automation design:
- Sold properties. A "Just Sold" advert may be placed only after the property has been formally transferred to the buyer, per Gulf News reporting on DLD requirements (13 December 2024).
- Takedowns. In February 2024 the DLD ordered agents to remove fake property listings within three working days, or face fines (Arabian Business, 16 February 2024).
- Broker limits. Gulf News also notes recent measures limiting how many brokers can advertise a given property and requiring a transacted property to be taken down once the deal is done.
I have not quoted fine amounts. The sources we reviewed do not give figures for these rules, and penalties change, so confirm current amounts with the DLD or counsel.
Where generative AI breaks these rules
None of the sources above say that using AI to write adverts is prohibited. The risk is indirect. Generative tools tend to do the following, each of which collides with a documented requirement:
Inventing or smoothing facts. A model asked to "make this listing more compelling" may round a size, add a view, imply a renovation, or describe a payment plan that does not exist. The requirement is that details are accurate to the property and consistent with the permit. Fluent copy that drifts from the permit record is a compliance defect, however good it reads.
Creating duplicates. Teams that generate variants for each channel (portal, Instagram, WhatsApp broadcast, email) can end up with several adverts for one unit. If one property may only be listed once by an authorised broker, variant explosion needs an architecture, not good intentions.
Losing the QR code. Image tools, video generators and caption pipelines frequently crop, resize or recompose creatives. A repurposed carousel slide or a clipped video end-card that drops the Madmoun QR code turns a compliant advert into a non-compliant one. This is a pipeline bug rather than a creative one.
Missing the lifecycle. A listing is not static. It is sold, rented, withdrawn, or transferred between brokers. If AI-generated content is syndicated to channels you do not control, stale adverts persist after the three-working-day expectation for fake or outdated listings has passed in spirit, if not in law.
Over-claiming in virtual staging and imagery. AI-edited photos and virtual staging can misrepresent condition. The DLD's description of Madmoun refers to property condition and specifications being visible to customers, which is a reminder that the verified record, not the render, is the reference.
An architecture that makes compliance a property of the system
The design principle is: facts flow in from verified records; AI only writes around them. In practice that means five components.
1. A listing record with a permit as its primary key
Every advertisable unit has one canonical record: the permit ID, the broker and company, the authorised property details, status (available, sold, rented, withdrawn), and the date the permit was obtained. Nothing is published unless it joins to a valid permit. If your CRM and listing tools do not share a single record today, that is the first project. Our piece on building a unified real estate data platform covers the pattern.
2. Structured generation, not free-text generation
Give the model the authorised fields as structured input and ask for copy that cites only those fields. Then validate the output. A simple, effective check extracts every number, area, bedroom count, price-like string and amenity claim from the draft and confirms each exists in the record. Anything unmatched fails the draft and goes to a human. This is a deterministic gate after a probabilistic step, and it is the single most valuable control in the whole design.
3. A creative pipeline that owns the QR code
Treat the Madmoun QR code as a mandatory layer in a template, not a manual add-on. Your image and video generators should compose creatives from templates in which the QR region is locked, with an automated check (a QR detector run on every rendered asset) before anything is queued for publishing. Fail closed: no readable code, no publish. For where listings and virtual tours meet marketing assets, see how we approach listing and virtual tour platforms.
4. Lifecycle events that fan out
Status changes must propagate. When a unit transfers, the system should withdraw or update every live advert, and "sold" messaging should wait for the formal transfer, as Gulf News reports the DLD requires. Build this as an event stream: transfer recorded, adverts for that unit enumerated, takedown or status update dispatched to each channel, and confirmation logged. Auditability matters more than speed here, because the first question in any dispute is what you published, when, and under which permit.
5. An evidence log
Store the prompt, the structured input, the model output, the validation result, the human approver and the permit reference for each published advert. If a platform or regulator asks why an advert said something, you want a lookup, not a reconstruction. Use sensible retention periods and keep personal data out of these logs wherever possible.
Data protection: leads, chatbots and profiling
Listing copy is only half of the AI surface. Most Dubai brokerages also use chatbots, lead scoring and campaign targeting. The UAE federal personal data protection law (Federal Decree-Law No. 45 of 2021) defines profiling as a form of automated processing that assesses aspects of a person including financial performance or condition and behaviour. As reproduced by Modulos, Article 18 gives data subjects the right to object to decisions resulting from automated processing that have legal impact or adversely affect them, with exceptions where the processing is agreed in a contract, required by other legislation, or based on prior consent, and Article 18(4) requires a human element in reviewing automated decisions on request (Modulos summary of the UAE PDPL). Article 17 also covers objection to processing for direct marketing, including related profiling.
Two honest limits apply. Baker McKenzie's handbook, last reviewed in December 2024, reported no regulator guidance on AI, automated decision-making or profiling and no enforcement activity at that time (Baker McKenzie). And the status of the executive regulations, which affect how the law is applied, should be checked against official sources and counsel rather than assumed. If you operate from free zones such as DIFC or ADGM, separate regimes may apply. The practical takeaway is to design lead scoring so that a human can review and override any decision that materially affects a buyer or tenant, and so that you can explain what data fed it. Our note on AML and client due diligence software discusses a related area where automated checks and human review have to coexist.
What to audit before your next campaign
Run this list across your marketing and listings stack. It is short on purpose.
- Permit coverage. Can you show, for every live advert on every channel including social, a valid permit ID? Orphaned adverts are the first thing to find.
- Duplicate detection. How many live adverts does each unit have, and which broker or company owns each?
- QR integrity. Sample your last 50 published creatives, including video end-cards and carousels. Is the QR code present, readable and pointing to the right permit?
- Fact-diff. Take 20 AI-assisted listing descriptions and diff them against the authorised record. Count unsupported claims.
- Sold and withdrawn lag. From transfer or withdrawal to the last channel updated, how long does it take? Compare that with the three-working-day expectation the DLD set for fake listings in 2024.
- Chatbot and scoring review. Which automated decisions affect buyers or tenants, and can a human review them on request?
- Vendor contracts. Do your AI and content vendors process personal data, and where, and under what terms?
What to do this quarter
You do not need a large programme. You need a sequence.
Weeks 1 to 3: map and measure. Inventory every place an advert is created or edited, including agents' personal tools, which is where shadow AI usually lives. Run the audit list above on a sample and record the failure rate. Without a baseline, you cannot show improvement.
Weeks 4 to 7: put the gate in. Stand up the canonical listing record and the post-generation fact check. Even if the first version covers only your highest-volume channel, it removes the biggest risk: unverifiable copy reaching the public. Add the QR presence check to the creative pipeline in the same sprint.
Weeks 8 to 10: close the lifecycle loop. Connect transfer and withdrawal events to channel updates, with a log of each takedown. This is also the stage to review chatbot and scoring flows for human review paths.
Weeks 11 to 13: rehearse. Run a mock inquiry: pick three live adverts at random and produce the permit, the source record, the generation log and the approver within an hour. The gaps you find are your next quarter's backlog.
A contrarian note to finish this section: resist the urge to buy a bigger model to fix this. The failures described above are mostly data and workflow failures, and a more fluent model will make them harder to see, not easier.
Build versus buy
Off-the-shelf listing tools will cover the basics, and many portals add their own checks. Custom work earns its place where your workflow is unusual: multiple brands, off-plan inventory with developer-supplied content, in-house creative teams, or integration with a CRM that is the real system of record. A real estate platform built around a single listing record, a broker network portal that keeps agents inside the approved workflow, and AI assistants with guardrails are the usual building blocks. If you are weighing generative tools for listing copy and staging, our earlier look at generative AI for real estate marketing sets out the trade-offs, and our compliance and risk consulting page describes how we scope assessments. Costs depend on integrations, channels and volumes, so scope and pricing are agreed on a consultation rather than quoted up front.
Talk to us about your listing pipeline
If you are planning AI-assisted listings, creatives or lead handling for UAE audiences, we can help you map the workflow, design the verification gates and build the integrations. Book a meeting with the Syslabs team to discuss scope; pricing is agreed after a consultation.
Sources
- Dubai Land Department, "Dubai Land Department strengthens transparency with AI-enabled real estate advertising governance", 24 April 2025.
- Dubai Land Department, "Dubai Land Department provides Madmoun service to verify validity of real estate ads via QR codes", 18 April 2023.
- Arabian Business, "Dubai property watchdog monitors over 279,000 listings with new AI platform", 24 April 2025.
- Gulf News, "Dubai property ads: Land Department gets AI to correct listings", 24 April 2025.
- Gulf News, "Dubai brokers will need to follow new strict rule on 'Just Sold' ads", 13 December 2024.
- Arabian Business, "Dubai real estate fake listing deadline", 16 February 2024.
- Modulos, UAE PDPL reference; Baker McKenzie Global Data and Cyber Handbook, UAE (last reviewed December 2024).
