Something has changed in how buyers find cars, and most of the industry hasn’t noticed yet, writes Martijn Versteegen, CEO at IMAGIN.studio.
For twenty years the rules of online visibility were reasonably stable. You optimised your text, built authority, appeared in a list of ten blue links and the buyer chose.
Images were part of the experience once someone arrived. They were rarely the reason they arrived – but that model is now being replaced.
AI search does not return ten blue links
When a buyer asks an AI which compact electric SUV suits a family of four, they don’t receive ten links. They receive an answer, assembled from a small number of sources that the system decided to trust. Everyone else is invisible. Not ranked low, but completely absent.
This matters more for automotive than almost any other sector, because our industry has spent decades building on shared imagery. Manufacturers supply press images, marketplaces syndicate them, and dealers upload them. The same hero shot of the same trim, in the same colour can appear on hundreds of sites simultaneously.
That was efficient, but now it is a liability.
Modern AI systems are built on deduplicated data, with research from Meta on semantic deduplication demonstrating that large portions of training data can be removed without degrading performance, because so much of it is redundant.
The filtering does not work on file names or metadata. It converts images into mathematical representations and compares actual visual content. You can crop, resize and rename, yet the system still recognises it as the same image it has already seen a thousand times.
Shared imagery is now a liability
When 500 retailers publish the identical manufacturer image, the system doesn’t see 500 pages. It sees one image and a great deal of repetition.
The same logic applies at the point of citation. Google holds a patent covering the estimation of information gain, which measures how much genuinely new information a page contributes relative to what has already been seen. Pages that add nothing score poorly – a page carrying images already published on more authoritative sites, is adding very little value by definition.
Academic work supports this. A study from Princeton and collaborating institutions on GEO found visibility improvements of up to 40% for content optimised for these systems, with larger gains for mid-ranking pages. Distinctiveness and presentation were among the strongest factors determining whether a source was cited at all.
None of this is a prediction. It is how these systems already work.
That said, it is important to note that there is still a lot of opportunistic noise in this area. Some of the more dramatic figures circulating are modelled estimates rather than measured results. The research on text is more mature than the research on images. Anyone claiming precise multipliers is guessing.
But the direction is not in question, and the direction is what should concern retailers and manufacturers. Visibility in AI-mediated search rewards content that exists nowhere else. Our industry’s standard practice produces content that exists everywhere.
Poor imagery now costs discovery
There is a second reason this is urgent, and it has nothing to do with algorithms. European buyers are researching brands they have never encountered. Search interest in Chinese manufacturers across the UK, Germany, France, Italy and Spain has risen dramatically over the past twelve months, in some markets more than tenfold.
These buyers have no reference points, and they cannot rely on decades of familiarity with a badge. Every judgement they make about whether a vehicle is right for them is formed from what they can see on a screen.
Poor imagery has always cost conversions. What is new is that poor imagery now costs discovery as well. A listing with a handful of badly lit phone photographs does not simply convert less effectively. It becomes progressively harder to find.
Genuine distinctiveness requires three things
Genuine distinctiveness requires three things, in practice.
The visual content itself must be mathematically distinct, not a lightly modified stock asset. The surrounding metadata and structured markup must be specific to that vehicle and that page. And the technical delivery must present a genuinely distinct file. Most shared imagery fails all three. Adjusting one does not fix the others.
The industry has treated vehicle imagery as a presentation layer – something applied at the end, once the important decisions are made. That framing no longer describes how buyers find cars. Imagery has become part of the technical infrastructure that determines whether a vehicle is discoverable at all, and businesses that recognise this early will be visible.
Those that do not will wonder where their traffic went.
Author: Martijn Versteegen, chief executive, IMAGIN.studio
Ensure you always receive AM insights. Make us a preferred source of news on Google