The Short Answer
ChatGPT, Perplexity and Google mainly read text. If you present a property with nothing but property photos, a floor plan as an image and a PDF brochure, you give these systems very little to read. The result: when a buyer asks, the AI tends to name the portal where the same property appears as text, or another agent. Good photos still matter, but they only work once someone has already found your site.
A Buyer Asks, and Your Photos Say Nothing
A buyer from Manchester types into ChatGPT: “Sea-view house in Jávea, three bedrooms, pool, up to 650,000 euros, which agents have something like that?” The answer comes in seconds. It lists two portals, a large international agency and perhaps an office in Moraira.
Your property is a perfect fit. You paid a photographer, waited for the evening light, the terrace has never looked better. Your website shows a gallery of 25 photos, with a “Download brochure” button underneath.
For the buyer, it would be exactly the right house. For ChatGPT, there is almost nothing on the page: a title, perhaps “Villa Jávea Ref. 1042”, and a download link.
What AI Search Actually Reads on Your Site
OpenAI describes its search crawler OAI-SearchBot in its official crawler documentation quite simply: it is used to surface websites in ChatGPT’s search results. What it evaluates for that is, above all, the text on the page.
Google works in a similar way, just with better documentation. In its image search guidelines, Google explains that it uses alt text together with computer vision and the page content to understand what an image shows. Captions and image titles count too. The image on its own is never the whole story: the text around it is what decides.
For a property listing, that means the “3 bedrooms” inside a floor-plan JPG, a price on a designed banner, or the energy rating in a photo of the certificate are not a reliable source. You do not know whether a model will read them from the image in any given case, and you cannot control it.
Why ChatGPT Overlooks Your PDF Brochure
Many estate agents on the Costa Blanca put the real information into the PDF brochure. From the office’s point of view, that makes sense: the brochure looks good, prints well and goes out by email.
To be fair: Google can index PDF files as long as they contain real text. A brochure is not invisible by default. In practice, it still has four weaknesses:
- No page of its own. The PDF hangs off a button, often without its own linked address. Search systems find it less easily than a normal page.
- Text as an image. Many brochures are exported from design software, and some are scanned. The PDF then contains pictures of text instead of text.
- Behind the form. If you only hand out the brochure in exchange for an email address, crawlers never see it.
- Not maintained. Price reduced, under offer, sold: the listing page gets updated, the PDF on the server often does not.
Why ChatGPT Ends Up Naming Idealista
On 15 October 2026 I checked the robots.txt files of Idealista, Fotocasa and Kyero. None of the three portals specifically blocks AI search crawlers there. Whether they limit bots at another technical level cannot be said for certain from the outside.
On all three portals, each property’s key details are text: price, square metres, rooms, area. If the same house appears on your own site only as a photo gallery, the portal is simply the better source for an AI search. Idealista gets the mention, and the enquiry goes to any agent who advertises there.
This is my assessment based on how the pages are built, not a measurement of AI answers. But you can easily check it yourself, more on that below.
What Every Listing Page Needs
The good news: for most of these points you do not need a new website, just a better template for your listings.
- One address per property. Every listing has its own page with a readable URL, for example
/sea-view-villa-javea-3-bedrooms/, not just?ref=1042. - Key details as text. Price (stating whether taxes and purchase costs are included), built area, plot, bedrooms, bathrooms, town and urbanisation, energy rating. As real lines of text or a table, not in an image.
- A description written by a person. Two or three paragraphs in the buyer’s language: location, light, distance to the beach and the town, what makes the house special. In English for British buyers, in German for German buyers, not machine-translated.
- Alt text that describes. “Kitchen with island and garden view” instead of “IMG_4471”. One short sentence per image.
- Captions on the key photos. Google names them explicitly as a signal.
- The floor plan as an image plus a room list as text. Living room 38 m², bedroom 1 with en-suite bathroom, and so on.
- The brochure as an extra. With a real text layer, no form gate, and the same information as the page.
How AI reads a website and decides whether to name a business, I explain in more detail in How AI Reads Your Website to Name Your Business.
The Two-Minute Test
Open any listing on your website. Select the whole page (Ctrl+A or Cmd+A), copy it and paste it into an empty text document.
What you see there is roughly what a crawler finds as text. Are the price, rooms, area and a description there? Or only menu items, “Ref. 1042” and “Download brochure”?
If it is the second, you know why ChatGPT does not name your property. How to then check directly in ChatGPT whether your agency is mentioned at all, I explain in Check If My Business Appears in ChatGPT.
Photos for People, Text for Machines
Good property photos are not a luxury. They decide whether a buyer gets in touch once she is on your site. But the finest evening light is worth nothing if she never gets there.
I’m Oliver, I live in Dénia and build websites for the self-employed and estate agencies on the Costa Blanca, with listing pages that people and AI search understand equally well. Whether ChatGPT, Gemini and Perplexity already know your agency, the free AI visibility check shows you. And if you want to know what exactly your listings are missing, write to me.