ChatGPT recommends real-estate agencies. How do you become part of the answer?

Let us start with the figure that changed things. In 2025, 6 % of consumers had already asked an AI to recommend a local business. By March 2026, it is 45 %. AI has become the third channel of local discovery, behind Google and Facebook, and ahead of Yelp or Tripadvisor.

Among 30-44 year-olds — the bracket that sells a first property or moves house — the proportion rises to 64 %.

The difference between ten links and three names

A Google results page shows ten links. The seller compares, opens three tabs, forms an opinion. A generative answer does not work that way: it gives two or three names, with one sentence of justification for each.

The whole question is whether yours is one of them. There is no second page.

What AI Overviews actually cite

A study published by Ahrefs in September 2026, covering more than three million queries, ranked the fifty most-cited sites in Google's AI Overviews. The result is surprising: YouTube comes first with 22.9 % of mentions, followed by Reddit at 18.5 %, Facebook at 10.1 %, Instagram at 5.6 %.

In other words, generative engines cite platforms where people talk far more than institutional websites. An agency with no presence outside its own site starts with a structural handicap, whatever the quality of its markup.

The point nobody tells real estate

Here is a nuance we would rather state, even though it tempers our own commercial pitch. According to Semrush measurements, real estate is among the sectors where AI Overviews trigger least: under 3 % of queries, against nearly 26 % for scientific subjects.

That does not make the work pointless, for two reasons. First because the share of affected queries is growing, and the widening observed through 2025 hit commercial and transactional queries first. Second because AI Overviews are only part of the subject: ChatGPT, Gemini and Perplexity answer questions about choosing a supplier every day, outside Google.

What concretely makes a source get cited

An academic study presented at the KDD conference in 2024, conducted by Princeton, Georgia Tech and the Allen Institute, tested nine content-modification strategies across roughly ten thousand queries. Three stand out clearly.

• Citing your sources: up to 40 % more visibility in generative answers
• Adding dated statistics: around 37 %
• Including attributed quotations: around 22 %
• Conversely, keyword stuffing has no measurable effect, and sometimes a negative one

This is good news: what makes content citable by a machine is exactly what makes it useful to a reader. A sourced and dated figure, an attributed claim, a paragraph that answers a question without requiring you to have read the previous one.

Where to start, in order

First check that your site is readable: server-side rendering, structured data, a robots.txt that does not block AI crawlers. Many recent sites serve an empty page to anything that does not run their JavaScript — and most AI crawlers do not run any.

Then give them something to cite: self-contained answers, figures with their source, a named author. Finally, exist elsewhere. An active Google profile, a few videos, a presence where your trade is discussed. That is the signal most correlated with citation, and the slowest to build.