Article contents
When a buyer consults an AI, it recommends areas, then figures, and only then agencies
A generative engine asked about a property project answers in a stable order: areas or neighbourhoods, criteria and price ranges, the procedure, and agencies only if the question invites it or if third-party sources name them. For an agency, the consequence is direct: it only gets into the answer by becoming the source for one of these building blocks, most often the local one.
This structure follows from the mechanism described in our reference article on being cited by ChatGPT, Claude and Perplexity: the buyer's question is broken down into sub-queries, and each sub-query brings back passages from different sources. “Where to buy near the sea between Alicante and Valencia with 250,000 euros” becomes one search on coastal towns, another on prices per square metre, a third on purchase costs, a fourth on recommended agencies. An agency whose website only answers the fourth has almost no chance; one whose pages answer the first three is cited, and the fourth follows.
In our audits of agencies and developers, in Spain as in France, the starting point is the same: property listings, an “About us” page, a form. Nothing that answers a buyer's sub-query.
What buyers really ask generative engines
Buyer prompts are long, contextualised, and mix several intentions in a single sentence. They rarely resemble the queries that classic SEO targets. Here are the formulations that come up most often in the panels we track, by project stage.
| Project stage | Example of a real prompt | What the engine looks for as a source |
|---|---|---|
| Exploration | “I want to buy a flat near the sea between Alicante and Valencia, budget 250,000 euros, two bedrooms, to live in for six months a year. Which towns would you recommend and why?” | Town or neighbourhood pages with prices, resident profiles, transport, seasonality |
| Trade-off | “Dénia or Jávea for a second home: what are the differences in price, atmosphere and seasonal letting?” | Dated local comparisons, with sourced figures |
| Procedure | “What are the steps and costs to buy a property in Spain when you live in France? NIE, bank account, notary, taxes?” | Structured, up-to-date procedure guides, with amounts and timescales |
| Provider | “How do you recognise a reliable estate agency on the Costa Blanca? Which French-speaking agencies in Alicante have good reviews?” | Directories, reviews, local press, agency entity pages |
| After the purchase | “How much are the property tax (IBI) and the community fees for an 80 m² flat in Valencia?” | Recurring-cost pages, by town, dated |
Table: scroll horizontally.
Two observations. The language of the prompt is the buyer's, not the market's: a French buyer purchasing in Valencia asks in French, and the engine favours sources in French, even if it finds few; an agency that publishes in its buyers' language occupies a space with little competition. And most of these prompts do not ask for an agency but for knowledge of the ground: it is that knowledge, published in citable form, that earns a recommendation.
- 01Project describedBudget, use, constraints, in one or two long sentences.
- 02Areas suggestedThe engine cites town or neighbourhood pages, often third-party guides.
- 03Criteria and figuresPrices, costs, taxation, timescales: the block where a local agency can be the most precise source.
- 04ProcedureLegal and administrative steps, taken from dated guides.
- 05Agencies namedOnly if third-party sources and reviews converge.
Why portals occupy the answers, and where room remains
On questions of area and price, generative engines first cite the major listings portals, public statistics and a few specialist media, because these sources are accessible, updated and corroborated by one another. An agency website will not dislodge them on “average price per square metre in Valencia”. It can, however, become the retained source on what these players handle poorly: the detail of a neighbourhood, the difference between two neighbouring towns, the real costs for a non-resident, the concrete timeline of a transaction.
Illustrative data. What to take from it: agencies are marginal on prices, but they can hold a real place on neighbourhoods and procedure, where portals remain generic; agency choice, for its part, is decided mainly on third-party sources (directories, reviews, press), absent from this chart.
The reasoning is that of a local publisher: where is the portal vague, outdated or absent? There, a precise, dated and signed page is retained ahead of the generic page of a more powerful site, because selection happens at passage level.
The neighbourhood page: the unit of citation for an agency website
The most frequently cited page on an agency website is neither the home page nor a listing: it is a town or neighbourhood page written as an answer. It answers the question asked (“Should I buy in Ruzafa or in El Cabanyal?”), gives dated and sourced figures, describes who is happy there and who regrets it, and it is signed by an agent who works there.
The difference from a classic neighbourhood page comes down to a few writing choices, which the following example, with fictitious figures, makes visible.
The second version contains a price, a period, a condition of use and an alternative: it is extracted as is, with the agency's name as the source. The first looks like thousands of portal pages. The figures on your page must come from your transactions and from the public data you cite, never from a convenient estimate.
The sections a neighbourhood page must contain
- Prices by property type, in dated ranges, with the public source used for cross-checking.
- Buyer profiles: who buys here, for what use, who would do better to look elsewhere.
- Local constraints: parking, noise, flood zones, tourist-letting rules.
- Comparison with two nearby neighbourhoods, because the trade-off sub-query is almost always there.
- A marked-up FAQ, in the buyers' own words.
What gets cited and what does not on an agency website
Property listings are almost never cited: ephemeral, duplicated on portals, with no answer to a general question. Procedure guides, on the contrary, are among the most frequently reused pages, because the “how to buy” sub-query appears in most projects and because these guides age quickly with those who do not maintain them.
| Page type | Cited by generative engines? | Why |
|---|---|---|
| Property listing | Rarely | Ephemeral, duplicated, no general answer |
| Home page | Rarely | Promotional copy, no data |
| Town or neighbourhood page | Often, if it has figures and dates | Answers the area and trade-off sub-queries |
| Procedure guide (NIE, costs, notary, taxation) | Often, if kept up to date | Answers the “how” sub-query, present in almost every project |
| “About us” page | Indirectly | Feeds the entity, not the answer |
| Quarterly market analysis | Often, if sourced | Freshness and figures, two selection criteria |
Table: scroll horizontally.
The procedure guide deserves particular care from agencies selling to non-residents. A guide “Buying in Spain when you live in France” that details the NIE, opening a bank account, the preliminary contract (contrato de arras), the notarial deed, the purchase taxes with their rate at the update date and the usual time between offer and signing answers five sub-queries at once. It must be re-read at every change in regional taxation, failing which the engine will reproduce the error as faithfully as a correct figure.
The agency entity: what the engine checks before naming you
Before recommending an agency, a generative engine seeks to establish that it exists, that it is authorised, that it operates where it says it does and that third parties speak of it in the same terms. This verification goes through structured data, then through the consistency of external presences. An agency whose name, address and area of operation differ between its website, its Google profile and the portals is read as several weak entities, none reliable enough to be recommended.
- 1Legal identityCompany name, registration in the agents' register where one exists (regional registers in Spain, professional licence in France), physical address.Otherwise: the agency is indistinguishable from undeclared intermediaries.
- 2Markup
RealEstateAgentwithareaServed,address,sameAs, languages spoken, andPersonfor each agent who signs.Otherwise: the engine does not know where you operate. - 3Readable reviewsUp-to-date Google profile, recent reviews, agency replies, consistent ratings across platforms.Otherwise: no evidence that clients exist.
- 4Third-party presenceProfessional directories, local press, associations, portals, with the same name and the same speciality.Otherwise: your positioning remains a self-declaration.
{
"@context": "https://schema.org",
"@type": "RealEstateAgent",
"name": "Mediterránea Casas",
"legalName": "Mediterránea Casas Alicante S.L.",
"address": { "@type": "PostalAddress", "addressLocality": "Alicante", "addressCountry": "ES" },
"areaServed": ["Alicante", "El Campello", "Sant Joan d'Alacant"],
"knowsLanguage": ["fr", "es", "en"],
"sameAs": ["https://www.linkedin.com/company/…", "https://g.page/…"],
"employee": [{ "@type": "Person", "name": "…", "jobTitle": "Estate agent, Playa de San Juan area" }]
}The example describes a fictitious agency. What matters is not the exhaustiveness of the markup but the concordance: the same areas of operation on the website, the Google profile and the directories, the same languages, the same address. For a multilingual agency, each language version must point to the same entity, and not to three “agencies” the engine will not know how to reunite.
Reviews: the evidence the engine looks for before naming an agency
On the “which agency” sub-query, generative engines rely on reviews more than on any other signal, because they are the only source that speaks about the agency without being the agency. They read volume, recency, content and the replies given. A review that describes a precise situation (“flat in El Campello, couple from Lyon, help with the NIE and the bank account, signing in seven weeks”) carries more weight than ten two-word reviews: it contains the same terms as buyer prompts.
Three practices stand out from our audits. Ask for the review in the days following the signing, with an open question about what was done. Reply to every review by restating the context (neighbourhood, property type, buyer's country) without revealing private information. And never buy or filter reviews: engines cross-check platforms, and a perfect rating here, a mediocre one there, is a signal of inconsistency, not of quality.
Reviews published only on the agency's website corroborate nothing: they are read as proprietary content. They must exist on at least one platform that AI search crawlers can read.
Measuring an agency's citation share, by language and by stage
An agency does not steer its generative visibility with Google rankings for “estate agency Alicante”, but with a panel of buyer prompts run every week, and a citation share read by language and by project stage. For a Costa Blanca agency selling to French, Belgian and Dutch buyers, that means a panel in three languages, covering exploration, trade-offs, procedure and agency choice. The logic is the same as the one described for “which product to choose” answers in e-commerce, with stages specific to the sector.
- Weeks 1 to 2Panel and access
- 40 to 80 buyer prompts, by language and by stage
- Check that AI search crawlers can access the neighbourhood pages and the guides
- First citation share, by engine
- Weeks 3 to 7Local pages and entity
- Six to ten neighbourhood pages rewritten as answers with figures and dates
- Procedure guide for non-residents, dated and sourced
RealEstateAgentandPersonmarkup, harmonisation of presences
- Weeks 8 to 12Reviews and corroboration
- Collection of detailed reviews and systematic replies
- Presence in directories and the local press with the same description
- Second measurement, read by language and by stage
Two readings are useful to management. The citation share on neighbourhood questions says whether the website has become a source of local knowledge; the share on agency choice says whether the entity and the reviews have convinced. The first progresses before the second: the engine recommends an agency after having cited it as a source several times. If the second remains at zero while the first rises, the answers must be read word for word: the engine often says something false or incomplete about the agency.
What to remember
- An AI advising a buyer recommends areas, figures and a procedure before naming an agency: be the source on the former to be named on the latter.
- Buyer prompts are long, in their own language, and mix exploration, trade-offs and procedure; neighbourhood pages with figures answer them, property listings do not.
- Portals dominate prices; an agency wins on local detail, the choice between neighbourhoods and the real costs for a non-resident.
- The agency entity is read in four layers: legal identity,
RealEstateAgentmarkup, readable reviews, consistent third-party presence. - Citation share is measured by language and by project stage; citation as a source precedes recommendation as an agency.
Frequently asked questions
Can an estate agency be recommended by ChatGPT or Perplexity?
Yes, but rarely directly. Generative engines name an agency when independent sources (reviews, directories, local press) designate it consistently and when its website has already been cited as a source on neighbourhood or procedure questions. An agency present only through its listings is almost never named, whatever its ranking on Google.
Do property listings need rewriting to be cited by AI?
No. Property listings are ephemeral and duplicated on portals; they answer no general question and are almost never retained as sources. The effort goes into town and neighbourhood pages, procedure guides and dated market analyses, which answer buyers' sub-queries and age less quickly than a listing.
Which language should you publish in when selling to foreign buyers?
The language of the buyer's prompt. A French buyer purchasing in Valencia asks in French, and the engine looks for sources in French, often few in number on local topics. An agency that publishes its neighbourhood pages and its procedure guide in French, English or Dutch occupies a space with little competition, provided each version points to the same entity.
Do Google reviews count for generative engines' recommendations?
Yes, it is the most used signal on the question “which agency to choose”, because it is the only source that speaks about the agency without being the agency. Engines read volume, recency, detailed content and the agency's replies. A review that describes the neighbourhood, the property type and the support received carries more weight than a rating without text.
How long does it take for an agency to appear in AI answers?
In our client engagements, the first citations on neighbourhood questions generally arrive two to three months after the publication of pages with figures and dates, if AI search crawlers have access to the website. Recommendation as an agency comes later, once reviews and third-party presences converge, which often requires an additional quarter.
An SEO practitioner since 2001, Kamel Malek runs SEO360 (Alicante, Valencia, Madrid, Paris). He has published three books on visibility in generative engines, including Generative Engine Optimization and Rétablir les faits.
