Niche language matters
Luxury, relocation, waterfront, downsizing, probate, second-home, and historic-home intent did not resolve consistently.
Real Estate AI Recommendation Benchmark 2026: a 10-prompt pilot baseline.
This pilot tests a simple commercial question: when a buyer or seller asks for a specialist agent in a specific US market, does the observed AI-answer dataset return a close, market-specific recommendation—or drift into broader, adjacent, or irrelevant material?
Intellin.
“Austin relocation”
“Miami luxury”
“Los Angeles probate”
Coverage was fragmented across the 10 prompts.
Seven prompts returned related records, but only a small minority produced a close market-level selection answer. Three returned no observed match.
Test the exact market-and-specialty questions tied to the business; do not infer visibility from a generic ranking.
01
Coverage was fragmented.
Seven of ten prompts returned at least one related record in the observed dataset, while three returned none. Only a small minority produced a close market-level agent-selection answer.
Several results widened the geography, changed the intent, or returned educational material instead of a named specialist. General search visibility should not be assumed to equal recommendation visibility.
This is not a national market-share claim. It is a dated pilot showing why teams should test the exact questions tied to their own market and niche.
02
Ten prompts. Four kinds of outcome.
The result column records what the dataset surfaced. The interpretation column shows how closely it matched the commercial question.
| Criteria | Observed dataset result | Interpretation |
|---|---|---|
| Austin · relocation | Related Austin agent-selection material appeared | Close intent, but not a stable named-specialist result |
| Miami · luxury | Market-level 'top Miami realtors' material appeared | Strongest close match in the pilot |
| Naples · waterfront | Statewide Florida agent material appeared | Geography widened and niche specificity was lost |
| Denver · listing | Mixed Colorado-agent and irrelevant records appeared | Query interpretation was noisy |
| Phoenix · downsizing | Adjacent 'best time to sell' content appeared | Transaction advice replaced specialist selection |
| Raleigh · relocation | No observed match | Coverage gap in this dataset |
| Tampa · luxury waterfront | Statewide Florida agent material appeared | Geography and specialty widened |
| Los Angeles · probate | Probate-specialist educational content appeared | Expertise concept resolved; named local selection did not |
| Scottsdale · second home | No observed match | Coverage gap in this dataset |
| Charleston · historic homes | No observed match | Coverage gap in this dataset |
03
Recommendation visibility is a systems problem.
Market authority is not enough on its own
Several records shifted from city-level questions to statewide rankings, broad educational content, or adjacent transaction advice.
AI systems need corroboration
Google's guidance emphasizes indexed, helpful textual content, internal links, accurate Business Profile data, and structured data that matches visible content.
The website must answer the commercial question
A team needs pages and evidence that connect who it serves, where it serves, what it knows, and why third parties support that claim.
Measurement must stay prompt-specific and dated
AI outputs vary by model, platform, query wording, location, source set, and time.
04
Transparent enough to challenge and repeat.
The purpose of the pilot is not to manufacture a headline. It is to define a useful baseline and the next, stronger test.
Scope
Ten US city-and-specialty buyer or seller prompts collected in July 2026.
Dataset
DataForSEO observed AI-answer records across supported platforms, including ChatGPT and Google where available.
Evaluation rule
Each prompt was classified as a close market-level selection, adjacent or broadened answer, irrelevant answer, or no observed match.
Limitations
This was not a live repeated-run test of every major AI engine. Dataset retrieval can normalize prompts to related queries. Results do not measure all users, markets, engines, or universal recommendation share.
Next iteration
Repeat a fixed panel across ChatGPT, Perplexity, Gemini, and Google AI or organic surfaces; run each prompt more than once; log sources and named agents; and invite independent operators to review the method.
05
Put the pilot in context.
Read the real-estate AEO pillar, see the service system, or compare provider models before testing your own market.
06
Check the evidence.
Claims and comparisons are grounded in current public guidance, independent directories, industry reporting, and dated observed data.
Google Search Central: AI features and your websiteGoogle's guidance on crawlability, internal links, textual content, page experience, Business Profile data, and structured data for AI features.Schema.org: RealEstateAgentThe canonical real-estate entity type and properties such as areaServed, address, aggregateRating, and organization relationships.HousingWire: The new real estate playbook is getting cited by AI, not clicked onIndustry context on AEO, local-data consistency, Business Profile stewardship, and prompt-level visibility measurement.
07
The honest answers.
What does this benchmark measure?
This pilot measures whether an observed AI-answer dataset returned a close market-and-specialty agent-selection answer for ten high-intent prompts. It records query relevance, geography, specialty, named entities, sources, platform, and whether no related record was observed.
Is this a national ranking of real estate agents?
No. It is a dated ten-prompt pilot designed to test recommendation coverage and query interpretation. It does not rank every agent, market, or AI platform and should not be read as a universal share-of-voice study.
Why did some prompts return adjacent answers?
AI-answer datasets can normalize a prompt to related queries, widen geography, or substitute educational content for provider selection. That behavior is itself useful evidence because it shows how easily niche commercial intent can be lost.
Can these results change?
Yes. AI answers are non-deterministic and change with the model, source set, location, wording, and date. Reliable measurement uses a fixed prompt panel, repeated runs, dated observations, and clear limitations.
How can a team test its own market?
The real-estate AI Recommendation Gap Report defines the buyer and seller questions that should lead to the team, records who and which sources appear, and connects the recommendation gap to a website inquiry path.
A national pilot cannot show what buyers see in your market.
Test the questions tied to your specialty, competitors, evidence, and inquiry path.
Start with evidence about your own market.
We'll show you where you appear, who is recommended instead, and what to fix first.