Dubai Real Estate Websites Are Invisible to AI Search
We crawled 708 brokerage domains drawn from the Dubai Land Department register. Only 5.5% are quotable by an AI engine, and the firms with the best performance ranks are no better online.
Dubai's real estate market runs on a public register: every licensed brokerage office is listed, with its licence number, its rank and, sometimes, a website. That makes it possible to ask a question the industry usually answers with opinion. How much of this market can an AI search engine actually read, and how much of it can it quote?
On 2026-10-20 we read the register, deduplicated it to 9,944 licences, extracted 708 unique domains, and requested each one with 9 crawler user agents. We then scored every site that served a page. The short answer is that the web presence of this market is thin, and the thin part is exactly the part AI answers depend on.
Key findings
- 9,175 of 9,944 licensed offices (92.3%) list no website. The register undercounts the number of firms with a web presence, not overcounts it.
- Only 26 of 472 scored sites (5.5%) reach a citability score of 50 or above. A further 89 (18.9%) reach 40. The median site scores 32 out of 100.
- Structured data averages 9.4 out of 100. The pages are readable but almost nothing on them is marked up in a way an answer engine can trust and lift.
- GPTBot is refused by 90 of 603 responsive sites (14.9%), against 12 for Googlebot on desktop (2.0%). The blocking is aimed at AI, not at crawlers in general.
- RERA performance rank does not predict web quality. Median AI visibility is 32 for Bronze, Silver and General-ranked firms alike, so a rank cannot be used to shortlist a digital supplier.
What the 9,944 licences look like
The register is the closest thing this market has to a census. It records the licence, the rank, and whatever contact details the firm chooses to publish. Most firms publish an email and a phone number. Almost none publish a website.
| Measure | Result |
|---|---|
| Licensed brokerage offices in Dubai Emirate | 9,944 |
| List no website on the register | 9,175 (92.3%) |
| List any social profile | 340 (3.4%) |
| Publish a contact email | 9,917 (99.7%) |
| List a phone number | 2,333 (23.5%) |
| Hold the buying and selling licence | 9,838 |
| Also hold a leasing brokerage licence | 6,504 |
Source: Dubai Land Department public register of RERA-licensed real estate brokerage offices, read 2026-10-20. Deduplicated by licence number; the raw export held 9,978 rows.
Which crawlers get refused
Each of the 708 domains was re-requested with 9 user agents. A refusal is a 403 or a CDN challenge. The denominator is the 603 sites that answered a crawler at all, because a site that never answered cannot have refused anything.
| Crawler | What it is | Refused | Rate |
|---|---|---|---|
| GPTBot | OpenAI, training | 90 of 603 | 14.9% |
| CCBot | Common Crawl, training | 47 of 603 | 7.8% |
| OAI-SearchBot | OpenAI, retrieval | 25 of 603 | 4.1% |
| PerplexityBot | Perplexity | 25 of 603 | 4.1% |
| ChatGPT-User | Triggered by a live chat | 19 of 603 | 3.2% |
| Googlebot smartphone | Google, mobile | 18 of 603 | 3.0% |
| Applebot | Apple | 14 of 603 | 2.3% |
| Bingbot | Microsoft | 13 of 603 | 2.2% |
| Googlebot desktop | Google, desktop | 12 of 603 | 2.0% |
The two training crawlers are refused at roughly three times the rate of the retrieval crawlers. GPTBot on its own is refused five times as often as Googlebot on a desktop user agent. The refusal is aimed at AI, not at crawlers in general.
What the sites score out of 100
The 472 sites that served a page and met the scoring criteria were scored with the geo-audit toolkit. The scale runs 0 to 100 and measures machine readability, not quality.
| Module | Mean score | What it measures |
|---|---|---|
| Structured data | 9.4 | JSON-LD types found against a recommended set |
| Citability | 22.7 | A summary, a FAQ, numbered structure, source links, clear definitions |
| llms.txt readiness | 30.2 | llms.txt, llms-full.txt, and AI crawler access in robots.txt |
| Technical | 37.6 | Indexability, speed, and the technical signals a crawler reads first |
| Content | 59.3 | Structure, depth, and how much a page gives a reader |
| AI search technical | 82.8 | Crawler access and the technical signals answer engines check |
The pattern is the finding. Reachability scores well, structured data scores badly, and citability sits between. The pages are readable and almost nothing on them is quotable, which is why the AI visibility scores in this market are low.
The quotable market
Only 26 of the 472 measured sites, 5.5%, score 50 or above on citability, the threshold at which an AI engine can quote a page. A further 89, or 18.9%, reach 40 or above. This number is a crawl you can download and check, not a secondhand estimate.
The rank does not predict the web
| RERA performance rank | Sites scored | Median AI visibility |
|---|---|---|
| Gold | 3 | 38 |
| Bronze | 89 | 32 |
| Silver | 36 | 32 |
| General | 344 | 32 |
The regulator's ranking says almost nothing about a firm's web presence. A performance rank is not a proxy for digital maturity, so it cannot be used to shortlist a digital supplier either. Across the full register the ranks split 4 Gold, 108 Silver, 636 Bronze and 9,196 General.
Methodology
- Registry. We read the Dubai Land Department public register of RERA-licensed real estate brokerage offices on 2026-10-20. The export held 9,978 rows, which deduplicated to 9,944 unique licence numbers.
- Domain extraction. We extracted 708 unique domains from the website and social fields and normalised them to registrable domains.
- Crawl. On 2026-10-20 each domain was requested with 9 user agents: GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, CCBot, Googlebot desktop, Googlebot smartphone, Bingbot and Applebot.
- Response classification. 603 sites answered at least one crawler, 490 returned a page, and 472 met the scoring criteria. 218 did not answer any crawler and were classified rather than dropped.
- Scoring. Each qualifying site was scored 0 to 100 across structured data, citability, llms.txt readiness, technical, content and AI search technical modules. Scores measure machine readability, not commercial quality.
- Open data. The per-site record is published as a CSV with the HTTP status, the final URL and every module score, including sites that could not be reached.
Limitations
- Crawler requests were sent from a single machine, so a CDN may treat the real crawlers differently. Sites that answered a challenge are counted separately from sites that returned a hard 403.
- The register is a snapshot. Firms add or remove websites between reads, and the crawl reflects one day.
- A refusal is a single-request observation, not proof of a permanent policy. Some sites may allow a crawler on a different path or at a different time.
- Scores measure machine readability, not the quality of a firm's service or the accuracy of its listings.
- The study does not test whether being quotable causes a citation. It measures the supply of quotable pages.
Download the data
The CSV holds 708 rows, one per unique domain. The JSON holds the 210 sourced statistics this study sits alongside. Both are free to reuse under CC BY 4.0 with attribution.
Cite this study
Amir Ali (Clienvora). (2026). Dubai Real Estate Website AI Visibility Study 2026 (v1.0.0). Clienvora Data Lab. https://www.clienvora.com/research/dubai-real-estate-ai-visibility/ Licensed CC BY 4.0. Free to republish with attribution and a link to this page.