What came back when I asked.
I ran this twice with two framings, because the answer depends on who the engine thinks is asking. Ten runs on 2 and 3 September 2026 from a Dubai IP address, logged out, clean browser session each time. Ten more on 6 September 2026 under identical conditions, with one sentence changed.
Prompt A · brokerage · 2 to 3 Sep 2026
"I run a real estate agency in Dubai. Recommend SEO/GEO firms that specialize in real estate. I am open to hiring international firms."
Prompt B · developer · 6 Sep 2026
"I am a property developer in Dubai. Recommend SEO/GEO firms that specialize in real estate. I am open to hiring international firms."
RealEstateSEO.ph appeared zero times out of ten in both. That is the honest starting position, I would rather publish it than not, and it was published before anything else on this page.
It is a baseline rather than a standing result. Those runs were taken on 2, 3 and 6 September 2026, before this page and the four pages under it existed, so what they measure is the position the work started from. The re-run on 2 December 2026 is the one that says whether it moved. What did appear, and how often:
Revised on 11 September 2026. Runs dated before this are the pre-revision baseline, and that includes the 10 September 2026 citation check further down this section. The 2 December 2026 re-run measures the page after revision. The prompt labels and the citation line below were added on 18 September 2026 and change no count. Opening order and hero stats revised on 18 September 2026. No count changed.
| Firm | Developer prompt | Brokerage prompt |
|---|---|---|
| SEO Sherpa | 8 of 10 | 9 of 10 |
| Arab SEO | 8 of 10 | not in top 5 |
| BIG LAB | 7 of 10 | 9 of 10 |
| SEODXB | 7 of 10 | 6 of 10 |
| Refrakt Digital | 6 of 10 | not in top 5 |
| NEXA | 5 of 10 | not in top 5 |
| Media Search Group | 4 of 10 | not in top 5 |
| Kairox Realty | 2 of 10 | 7 of 10 |
| FlyDragon | 2 of 10 | 6 of 10 |
| RealEstateSEO.ph | 0 of 10 | 0 of 10 |
Measurement stamp: ChatGPT, logged out, Dubai IP, 10 runs per prompt, prompt strings frozen. Brokerage cell 2 to 3 September 2026, developer cell 6 September 2026. Duplicate byte-identical outputs discarded. Both prompts, the IP condition, the run counts and the dates are published above so anyone can rerun this.
The two prompts are separate results. Arab SEO, Refrakt Digital, NEXA and Media Search Group were named on the developer prompt only, at 8, 6, 5 and 4 of 10, and none of the four reached the brokerage prompt's top five. SEO Sherpa, BIG LAB, SEODXB, Kairox Realty and FlyDragon appeared on both.
The same method, pointed at the questions buyers ask rather than the one a brokerage asks, is the Dubai brokerage census: 50 ChatGPT answers to 10 property questions, with every firm named resolved against the Dubai Land Department broker register.
To cite this table: RealEstateSEO.ph run log, ChatGPT logged out, Dubai IP: brokerage prompt 10 runs (2 to 3 Sep 2026), developer prompt 10 runs (6 Sep 2026). Published by a firm that sells in this category.
Next scheduled re-run: 2 December 2026
The 2 December 2026 re-run will log every cited URL in every run. It reports two rates separately: how often this page is cited as a source, and how often this firm is named as a candidate.
One disclosure it has to carry. From 10 September 2026 onward, ChatGPT was citing this table when it built its own shortlists, so the re-run partly measures the effect of publishing it. Runs that cite this page will be reported separately from runs that don't.
Within days, the run log was being cited.
On 10 September 2026, days after this page went live, I asked the brokerage prompt again from a Dubai IP, logged out, ten runs. On the brokerage prompt, with its international clause, this run log was cited by name in 7 of 10 of them, with the appearance table above reproduced accurately. I ran it a second time with that one sentence removed and nothing else changed, ten more runs, and it was cited in 5 of 10. The developer prompt was not rerun that day.
The firm itself was named as a shortlist candidate in 1 of 20 runs across the two brokerage variants: 1 of 10 on the brokerage prompt, and 0 of 10 with the international clause removed.
That gap is the finding, and it is worth more to you than the appearance counts are. Publishing measured, sourced, reproducible work earns a citation fast. Earning a citation is not the same as earning a recommendation. The two are decided by different things, and this page is the evidence for the distinction, because it is the same page in both results, measured a week apart with the same method. A study gets quoted. A candidate gets shortlisted. If you are buying this work, that is the difference to ask any firm to show you, and it is the reason the rest of this page reads as an offer rather than a paper.
This run log was cited by name in 7 of 10 runs on the brokerage prompt, 10 September 2026.
It was cited in 5 of 10 runs on the brokerage prompt with the international clause removed.
The firm was named as a candidate in 1 of 20 runs across both brokerage variants: 1 of 10 with the clause, 0 of 10 without.
Measurement stamp: ChatGPT, logged out, Dubai IP, 10 runs per variant, prompt strings frozen. Both variants are the brokerage prompt, run 10 September 2026. The variants differ only by the sentence "I am open to hiring international firms," which the two prompts above both carry and the second variant drops. Duplicate byte-identical outputs discarded. A citation counted as this page or this domain named as a source in the answer body, which is a different measurement from the appearance count in the table, and the two are not comparable.
Twenty runs, three findings.
The appearance counts are the shallowest thing the run log holds. What the field did when one noun changed, what the firms in it actually had on their sites, and what the answers cited when they put a firm first, are the three that matter.
Changing one word changed the field
25 distinct firms were named across the developer-prompt runs.
Only 3 of the brokerage prompt's top 5 survived into the developer prompt's top 5.
Kairox Realty fell from 7 of 10 on the brokerage prompt to 2 of 10 on the developer prompt.
FlyDragon fell from 6 of 10 on the brokerage prompt to 2 of 10 on the developer prompt.
Twenty-five distinct firms were named across the developer-prompt runs. Only three of the brokerage prompt's top five survived into the developer prompt's top five. Kairox Realty fell from 7 of 10 on the brokerage prompt to 2 of 10 on the developer prompt. FlyDragon fell from 6 to 2 across the same change.
Arab SEO and Refrakt Digital appeared at 8 and 6 of 10 on the developer prompt without featuring in the brokerage prompt's top five at all. So a developer and a brokerage in the same city, asking the same question with one noun changed, are shown substantially different shortlists.
If you are a developer and your agency's pitch cites its visibility on brokerage-facing queries, that is a different query from the one your buyers run.
Then I read every firm that appeared, on both prompts.
Finding 01 · SEO Sherpa
A client wall beats a vertical page
Two of the brokerage top five have no real estate practice at all. SEO Sherpa has no real estate service page anywhere on its site. What it has instead is DAMAC and Meraas in its client wall. That turned out to be the pattern: naming one property brand a buyer recognises does more than a vertical page does.
Finding 02 · Kairox Realty
Eleven pages out-appeared forty
Kairox Realty is a marketing agency rather than a brokerage, despite the name. Its site is eleven pages. It appeared more often than firms with forty on the brokerage prompt, and dropped to 2 of 10 on the developer prompt.
Finding 03 · FlyDragon
The sharper prompt got the scope check
FlyDragon's own site states plainly that they build exclusively for United States residential real estate and lock ZIP codes to one agent per market. In the brokerage prompt the engine recommended them for Dubai six times out of ten anyway. On the developer prompt they dropped to two, and in those two the answer itself flagged the limitation, noting the firm's positioning is oriented toward agents and brokerages and that a developer should make them prove otherwise. The engine applied more scope scrutiny to the more specific prompt. That is a finding about how these answers are assembled, not about the firm.
What actually earned a first-position recommendation
This is the part worth more than the appearance counts. Across the ten developer-prompt runs, the firm placed first was BIG LAB five times, SEO Sherpa twice, iProspect twice, The Parthenon once.
BIG LAB was placed first in 5 of the 10 developer-prompt runs.
SEO Sherpa was placed first in 2 of the 10 developer-prompt runs.
iProspect was placed first in 2 of the 10 developer-prompt runs.
The Parthenon was placed first in 1 of the 10 developer-prompt runs.
Where the answer justified a first position, it did it with a named client and numbers attached:
| Firm | Named client | Evidence cited |
|---|---|---|
| BIG LAB | Mira Developments | A GEO programme |
| iProspect | Dubai Properties | 32 percent more organic traffic and 35 percent more property bookings |
| The Parthenon | LYM Real Estate | Roughly 11,000 to over 150,000 monthly impressions with citations across seven AI engines |
| Keyreative | The Myriad Dubai | 120 to 135 organic enquiries a month, without ever placing first |
Not one firm was placed first on the strength of a service page describing what it does.
The same answers then undercut their own evidence. Multiple runs noted that those figures are agency-reported and told the buyer to ask for anonymised Search Console or GA4 data and client references before accepting them. So the engine ranks on named case studies while coaching the buyer to distrust the numbers inside them.
I do not have a Dubai client case study. I do have live Search Console accounts I will screen-share on a call, which is the specific evidence those answers tell you to ask for.
What the answers told buyers to do.
Three things came up in nearly every run, and they are worth reading whether or not you ever talk to me. The same material is worked into a list of questions to ask any firm you shortlist, including this one.
01
Do not buy SEO and GEO as two separate services
The answers were consistent that entity authority, structured data, technical foundations and third-party citations serve both, and that a standalone GEO package usually means conventional content work with a new label.
02
Be wary of any firm promising guaranteed ChatGPT rankings
Or a guaranteed citation outcome. Nobody controls that. I have the same sentence further down this page.
03
Run a four or five firm pitch, and make each one audit a real development
Several runs specified giving every firm the same 20 to 30 commercially important prompts and asking each to baseline your current visibility across ChatGPT, Gemini, Perplexity and Google AI Overviews before proposing anything. That last one is worth acting on regardless of who you hire. It is also, more or less exactly, what my paid diagnostic is.
Smaller observation
Several runs cited Arab SEO specifically because it publishes developer pricing, at AED 8,500 (USD 2,314) per month and from AED 15,000 (USD 4,084) for a master plan. Publishing a rate card appears to make a firm easier to recommend. Mine is further down this page.
Where the answers come from
Several runs cited ranking pages with titles like "best GEO agencies in Dubai" or "top AI search optimization agencies in Dubai." A number of those are published by agencies that also appear in their own rankings. That is ordinary marketing, and it is worth knowing before you treat a list as independent. Check who published any ranking you are handed, including this page.
The Dubai search picture.
Two structural facts decide what a brokerage or a developer can win here. The first is the size and the shape of the transaction volume. The second is who already holds the transactional queries.
H1 2026, all transactions
All Dubai real estate transactions in H1 2026 came to AED 419.94 billion, USD 114.3 billion, across 112,850 transactions.
H1 2026, sales only
Sales alone in H1 2026 came to AED 286.44 billion, USD 78.0 billion, through roughly 86,000 deals.
Within that half year
Q1 2026 alone was AED 252 billion, USD 68.6 billion, across 60,303 transactions, up 31 percent year on year in value on a 6 percent rise in volume.
Dubai recorded AED 252 billion (USD 68.6 billion) in real estate transactions in the first quarter of 2026, up 31 percent year on year in value on a 6 percent rise in volume, across 60,303 transactions (Dubai Land Department). Emirates 24/7, citing DLD, put the first half at AED 419.94 billion (USD 114.3 billion) across 112,850 transactions, of which AED 286.44 billion (USD 78.0 billion) was sales through roughly 86,000 deals.
Off-plan is two thirds of the volume and under half the money
The split inside that sales number is the part that matters for search. Off-plan accounted for 58,840 of those transactions and AED 139.75 billion (USD 38.1 billion), which is roughly 68 percent of the deals. Ready property accounted for 27,160 transactions and AED 146.69 billion (USD 39.9 billion), roughly 32 percent of the deals and slightly more of the money. Those are two different buyers running two different searches, and most brokerage sites are built for one of them.
| Segment | Deals | Share of deals | Value |
|---|---|---|---|
| Off-plan | 58,840 | 68 percent | AED 139.75 billion (USD 38.1 billion) |
| Ready property | 27,160 | 32 percent | AED 146.69 billion (USD 39.9 billion) |
The second structural fact is that Bayut, Property Finder and dubizzle hold the transactional queries. A search for apartments for sale in Dubai Marina returns portals, not brokerages, and no amount of on-site work changes that in the near term.
What brokerages can hold is the advisory layer: the questions buyers ask before they know which unit they want, and increasingly ask an AI rather than Google.
The advisory layer, worked out in four pages
Each one takes a claim this page makes and tests it against the published data, with every figure sourced and dated.
What the portals actually hold
Listing counts for Bayut, Property Finder and dubizzle against a large brokerage, and why the brokerage's own stock is already inside the portal's total.
Why brokerages compete with each other
Three systems solve three different versions of listing duplication. DLD authorises, the portal ranks, the engine retrieves, and only the third is open to you.
Off-plan against completed
Different statutes, a 30-point financing gap, and which of the two segments has the stronger content opportunity as of September 2026.
Who is buying, and in what language
The DLD publishes origin in three buckets and no nationality register exists. What English SEO reaches, and the segments it does not.
Two engagements, one per segment.
/developers
If you are a developer
When a buyer in London or Mumbai asks an AI assistant which Dubai developer to trust, your name either comes up or it doesn't. For most Dubai developers it doesn't, and brand size isn't what decides it.
The work is entity architecture, market and area content, and measurement across ChatGPT, Gemini and Google's AI Overviews. I take one client per market segment in Dubai, and developer-side work is a segment of its own.
Foreign-buyer funnels, multilingual site architecture, and the trust content that closes a reservation from someone who has never seen the unit.
Out of scope: I don't build property listing pages or launch microsites. In Dubai those are regulated advertising under Trakheesi, and they belong with your permit holder.
/brokerages · 15 to 60 agents
If you are a mid-size brokerage
Property search inside AI assistants increasingly routes through the portals. What it doesn't route through them is the question that comes first: which area, what the process looks like for a foreign buyer, and who to actually work with.
That's the ground still open, and it's where a 15 to 60 agent firm can outrank a portal, because a portal can't answer it with anything specific.
I work with one client per market segment, on a three-month minimum. Segment definitions and current availability are on the exclusivity policy page.
What a Dubai engagement actually builds.
The answers I measured volunteer close to the same checklist every time, and a buyer running the four-firm pitch above will hand it to all four. This is mine, answered in the same terms, so you can put it beside theirs.
Architecture
Community and area pages
Dubai Marina, Downtown Dubai, Business Bay, Palm Jumeirah, Jumeirah Village Circle, Dubai Hills, Arabian Ranches and Emaar Beachfront each need a page that answers what buying there involves: the price band, the service charges, the handover position, who the stock suits. Most brokerage area pages are a description paragraph wrapped around a listing feed, which answers none of it.
Architecture
Off-plan launches and project hubs
An off-plan buyer searches the payment plan, the handover date, the escrow arrangement and the developer before they search the unit type. A project hub holds those in one place and stays canonical for that project through to handover, rather than being replaced by a fresh launch page every quarter and splitting its own history four ways.
Entities
Developer entity pages
Emaar, DAMAC, Sobha, Nakheel and Ellington are entities the engines already hold. A page about one of them either connects to that entity or competes with it. I build them to connect: consistent naming, the project list, and the developer, project and community relationships stated in markup rather than implied by a navigation menu.
Entities
Listing schema and entity architecture
RealEstateListing, Residence, Place and Organization nodes, joined by @id so a listing resolves to its project, the project to its developer, and the developer to the community it builds in. The relationships matter more than the node count: a graph an engine can walk is what makes several of your pages corroborate one answer instead of competing to be it, which is the mechanism the REN.PH section below shows working.
The markup layer has its own page: what an answer engine reads.
Technical
Large listing inventories
A brokerage with a few thousand live units has a crawl problem before it has a content problem. The work is deciding which facet combinations earn indexable URLs and which stay parameters, what gets a canonical, and what happens to a URL when the unit sells. Getting that wrong is the most common reason a large site's good pages are crawled too rarely to matter.
Technical
hreflang and multilingual structure
Return-tag pairs that actually reciprocate, an x-default that resolves, and a URL structure that survives adding a fourth language two years from now. I build that layer for Arabic the same as for English or Russian. The limit is that I don't write or research Arabic myself, and the FAQ below states it in full rather than repeating it here.
Content
DLD transaction data as an input
The Dubai Land Department publishes transaction volumes and values, which is source data a page can be grounded in, dated and cited. It's the same mechanism that got REN.PH cited off BIR zonal values. A market page built on published transactions ages honestly and can be re-dated each quarter. One built on an opinion can't be.
Content
Investor and process content
Golden Visa thresholds, service charges, freehold against leasehold, escrow, transfer costs, and mortgage eligibility for non-residents. This is what a foreign buyer asks before they've chosen an area, it gets asked of an assistant more often than of a broker now, and it's the set of questions a portal answers worst.
Local
Google Business Profile and entity consistency
Name, address and phone consistent across the profile, the site, the portals and the directories, with the profile's categories and service areas matched to what the site actually claims. It's unglamorous, it's wrong more often than not, and it's one of the few things that still moves a local pack.
Measurement
Frozen-prompt citation monitoring
The method at the top of this page, run monthly for you across ChatGPT, Gemini and Google's AI Overviews: a fixed prompt set, a fixed run count, a logged IP, and a record of who gets named and what got cited. That is how you find out whether the work moved the number instead of being told that it did.
What an assistant says about a firm has its own page: real estate GEO.
Every one of those is inside the tiers on the rate card below. The tier decides how much of it runs at once and in what order, not which of them you get.
What I do differently, and where it comes from.
I hold a broker's licence, so when I open a brokerage site I read the listing structure first and the crawl second. That order matters more than it sounds.
The broker's read
Duplicate content, or cannibalisation?
The most common thing I find on Dubai brokerage sites is not a technical defect. It is forty units from the same development, listed by six different brokerages, all inheriting one description from the developer's brochure, all competing with each other and with the portals for the same phrase. An SEO audit calls that duplicate content. A broker calls it the reason your own listings cannibalise each other. The fix is different depending on which problem you think you have.
I wrote up how I read a site on the licensed-broker page. The worked example there is a real brokerage, named with its owner's permission: Oasis Real Estate in Pattaya, Thailand, whose Managing Director, Johannes Pannier, let me run an independent check across the infrastructure his team already had in place and publish where I saw room to gain.
The engineering side
REN.PH, and how engines cite it
REN.PH is a Philippine property data platform I built and run: zonal values and broker profiles, structured and grounded against BIR source data, 65,291 pages in the live XML sitemaps as of 21 August 2026, counted as pages rather than data rows.
The part of that worth knowing about if you're evaluating me for Dubai isn't the page count. It's how AI engines cite it. On a Taguig zonal value query, checked twice four months apart, including once logged out of Google in a private browser to rule out my own search history skewing it, the AI Overview built its answer from REN.PH pages: a city hub page plus several barangay-level pages cited together for one answer.
It also earns traffic, which is the part a page count does not tell you. REN.PH took 121,293 Google Search Console clicks across 28 April to 18 August 2026, on a platform that had no organic traffic at the start of the year. That is the number I screen-share, and it is the same kind of evidence the answers in section 02 tell you to demand of anyone.
REN.PH took 121,293 Search Console clicks from 28 April to 18 August 2026.
REN.PH has 65,291 pages in live sitemaps.
The Taguig AI Overview was checked 2 times, four months apart.
What it proves, and what it does not
I want to be precise about what that does and doesn't prove. It shows I understand the mechanism, structured, cross-linked content grounded in one verifiable source getting treated by these engines as a single corroborating authority rather than isolated competing pages, and that I've watched it happen on a platform I built and control. It does not mean I can promise the same outcome on your site. Nobody legitimately can. REN.PH has years of structured government data behind it and no competing content ecosystem the way Bayut and Property Finder dominate Dubai property search. Different market, different competitive field, different odds. If an agency tells you they can guarantee an AI citation outcome, that's the line to walk away on, and I'd rather tell you that directly than have you find it out from someone else's warning about firms like me.
What I can commit to is the same measurement discipline shown at the top of this page, applied to your firm: a frozen set of prompts, run repeatedly, logged, so you know whether the work is moving the number rather than being told it is.
What doesn't transfer from REN.PH to Dubai: Philippine search intent, Philippine portal dynamics, Philippine buyer language. That's a different research problem, which is why the rest of this cluster exists rather than a copy of the REN.PH playbook with Dubai swapped in.