Your CGI Is Now Search Data: How Rightmove's AI Reads Listing Images

Rightmove's AI now scans listing images and text to match buyers to features like "river view" or "exposed brick". Here is what that means for how you brief CGI.

September 29, 2026

Buyers are searching for features, not postcodes

Rightmove launched AI Keywords in November 2025. Buyers type a phrase such as "exposed brick", "river view" or "underfloor heating", and the system, in Rightmove's words, scans images and text in listings to surface matching homes.

In February 2026 the portal added a beta conversational search built on Google's Gemini models, where a buyer describes what they want in plain language and gets matched listings back.

Rightmove has not published how images are weighted. Treat everything below as an informed reading of the direction of travel, not a ranking formula.

Why this matters for new-build marketing

A listing image used to do one job: make a person stop scrolling. It now has a second job. It is a data source that a machine reads before any human sees it.

For an off-plan scheme, the CGI is often the only imagery in the listing. So what is visibly in the frame (the balcony aspect, the waterside view, the kitchen island, the floor finish) is the only evidence the system has to work with.

One caveat. Rightmove's announcement does not say whether these tools apply to new-build listings in the same way as resale. Check with your Rightmove account contact before you build a strategy on it.

Four briefing changes worth making now

Show the selling feature clearly. If the scheme sells on a river view, one image should show that view unobstructed and recognisable, not a moody crop where it is a blur behind a curtain.

Match the images to the description. If the copy says "private terrace", a CGI should show the terrace. Text and image that agree are easier for any system to read.

Specify materials on purpose. Brick, timber, stone and flooring finishes are searchable terms. Confirm them in the design pack before the studio models them, so the image shows what will actually be built.

Keep the CGI honest. Rightmove also runs an experimental "Style with AI" tool that lets buyers remove furniture and restyle rooms. Buyers are being trained to expect a real, clean shell behind the styling. Imagery that shows what will genuinely be delivered holds up better next to that.

What this does not mean

It does not mean stuffing every frame with keywords. A crowded image sells nothing to a person, and there is no evidence that more objects help a ranking.

It does not mean CGI is favoured over photography. Nobody has published that. It means the content of the image matters more than it did.

Competitors are moving too. Zoopla announced an OpenAI enterprise agreement in April 2026 and reported early tests showing an 80% jump in listing views and a 150% lift in leads among users exposed to its AI features. That is Zoopla's own early testing, so weigh it accordingly, but it shows where the portals are investing.

How to brief your visualisation studio

Send a short feature list with the brief: the five things a buyer would search for to find this home. Ask the studio to make sure each one is clearly visible in at least one hero image, and to confirm the materials against the spec.

That costs nothing extra at brief stage and is expensive to fix after the images are signed off.

Quick answers

Does Rightmove's AI really read listing photos?
Rightmove says its AI Keywords feature scans images and text in property listings to find homes matching a buyer's search phrase.

Does CGI rank better than photography on property portals?
There is no published evidence of that. What the image shows matters more than how it was made.

Do these AI tools apply to new-build listings?
Rightmove's announcement does not say. Ask your account contact before planning around it.

What should a developer change in a CGI brief?
Name the features buyers search for, make sure each is clearly visible in a hero image, and confirm materials against the design spec before modelling.

‍

← Back to all posts