Rollo
The method

How we measure whether AI can find you.

Brokers rightly ask: who checked, what did they type, and which AI? Here is exactly how the free check — and the study behind it — works, what “consistently findable” means, and what it does and doesn't prove. No black box.

What we measured

Can a buyer's AI name an individual agent — or only brands?

When a buyer asks an AI assistant to recommend an agent, three things can happen: it names you, it names someone else, or it names no individual at all and falls back to agency brands or portals. We measure how often each happens for a named, licensed broker — across the assistants buyers actually use.

How we run it

Many runs, four engines, scored by visibility — not a single answer.

1
Sample
~500 Dubai-licensed brokers drawn from the public RERA/BRN register — named individuals, not agencies. The sample and its date are recorded.
2
Engines
ChatGPT, Perplexity, Gemini and Claude — the assistants buyers use. We record the model and date of every run, because they change over time.
3
Buyer-intent queries
The questions a real buyer types — “recommend a broker in [area]”, “who's a trusted RERA-licensed agent for [area]” — plus name-specific checks (“is [broker] a real, licensed agent?”). Phrasing is varied so no single wording skews the result.
4
Repeat
Each query is run many times per engine. A single AI answer is unreliable: independent research found a less-than-1-in-100 chance that two runs of the same prompt return the same list, so the recommended practice is to run each prompt dozens to hundreds of times and average. We do.
5
Score by visibility %
We don't track a single ranking. We track visibility %— how often, and how prominently, an agent is named across all those runs, weighted so a prominent mention counts more than a passing one.
The definition

What “consistently findable” means.

An agent is consistently findable when they clear a set visibility-% threshold across the repeated runs for their own area's buyer-intent queries and the assistant surfaces a credential it can verify — a RERA/BRN identity — not just a name in passing. Below that threshold, the agent is not consistently findable: a buyer who asks today and again tomorrow can't rely on being shown them.

The headline number — of ~500 licensed brokers, only a few were consistently findable — is simply the count of agents who cleared that bar.

What we found

The same handful of brands. Almost no individuals.

Across the runs, the assistants named the same few agency brands and rarely a specific licensed agent.

AI platformAcross ~500 Dubai brokersResult
ChatGPTBuyer-intent recommendation queries91% returned no specific agent name
PerplexityBuyer-intent recommendation queries87% named only agency brands, not individuals
Google AIBuyer-intent recommendation queriesRERA-verifiable presence in <3% of cases
Source: Rollo query analysis, 500 Dubai-licensed brokers, June 2026
Why this happens

AI leans on brands it already knows.

This isn't a quirk of our test. Published research shows AI language models systematically favour established, well-known brands over local and individual ones, and give incomplete answers about nearby businesses. An individual agent with no machine-readable, verifiable record is exactly who the model leaves out. That is the gap a verified Rollo presence is built to close: a record AI can read and check, tied to your RERA licence and BRN.

What this is — and isn't

The limits of the check.

  • It's a measurement, not a guarantee. Identical questions give different answers run to run, so we report a visibility % over many runs — never a single ranking or a promise.
  • It's dated. Every figure is measured on a specific date against the then-current models. Results drift as the models update — which is why the check is re-runnable.
  • It varies by location and account. AI answers change with a buyer's location, history and settings. We hold the query and market steady; we can't control every personalisation signal.
  • It's not an endorsement, a licence verification, or a complete market ranking. It measures what AI returned, nothing more. It is self-initiated by Rollo and not run on behalf of any named agent.
Prior work

The research this builds on.

  • Aggarwal, Murahari et al., GEO: Generative Engine Optimization (Princeton / Georgia Tech / IIT Delhi / AI2, KDD 2024) — visibility in AI answers as a prominence-weighted metric. arXiv:2311.09735
  • SparkToro (2025), AIs are highly inconsistent when recommending brands or products — why single AI rankings are noise and visibility % across many runs is the right measure. sparktoro.com
  • “Global is Good, Local is Bad?”: Understanding Brand Bias in LLMs — AI models favour established brands over local ones. arXiv:2406.13997