Your images should tell the story your brand wants to own.
A brand can invest years in building a position, then publish a stream of beautiful images that barely express it. That is a useful creative question before it becomes a question about search.
Multimodal AI can interpret images as well as text. Asking a model what a picture communicates can expose a gap between the story a marketing team intends and the story visible in the work. That reading is an interpretation to examine, not a measurement of the brand's position in AI search.
A limited exercise with Volvo imagery.
Volvo is a useful example because safety is central to its history. Nils Bohlin's three-point seatbelt design was patented, and Volvo made the patent available to other car manufacturers. Volvo's own history describes the invention and the shared patent.
In the exercise reported in the original version of this article, ChatGPT 5.2 was used to interpret a selected set of 27 Volvo images. The recorded reading identified no direct safety visuals and only one or two weak, indirect cues; the other images were read mainly through design, urbanity and electric mobility.
That finding raises an art-direction question: if safety is the intended message for this particular set, where is it visible? It does not establish how Volvo is represented across its entire communication, what a person would infer, or how an AI search service would answer a question about the safest car.
An image reading is not a ranking mechanism.
An image prompt, model training and a search service retrieving sources are different processes. Publishing a new picture does not mean it immediately trains every model. A model's description of that picture does not reveal the weight a search system gives it.
Historical reporting, product information, independent evidence and the question being asked can all provide context beyond a selected image set. We cannot infer that brand heritage disappears when a picture fails to illustrate it.
For Google's AI Overviews and AI Mode, Google says established SEO practices still apply. It recommends accessible text, helpful content, internal links and relevant high-quality imagery. There is no special AI schema or additional optimization required, and eligibility does not guarantee inclusion.
Make the creative question practical.
Start with the brand meaning you want a specific campaign or product page to communicate. Then examine the images with and without their surrounding copy. Can a reader identify the product, understand the intended use and recognise something distinctive about the brand?
- Define the message. Select the brand pillars that matter to this brief. A campaign does not need to explain the entire company in every frame.
- Review the evidence. Compare human readings with model interpretations using recorded prompts and a fixed image set. Treat disagreements as questions for review.
- Improve the work. Strengthen the relevant visual story while keeping product facts, captions and page copy accurate. Then test the result against the original communication objective.
Measure search visibility separately from image interpretation. Record the service, query, date and sources shown; repeat the observation before drawing a conclusion. Measure enquiries and qualified conversations as business outcomes, rather than treating an AI description as a proxy for demand.
Clarity is a creative advantage.
Our point of view remains that a brand should be deliberate about what its imagery communicates. A beautiful image can still miss the brief. AI gives the team another way to question that image, alongside experienced human judgement.
For another practical consequence of changing discovery, read what happens when a product image leaves its original context. If your next production needs a clearer visual direction, explore how Scenes combines real products and AI-created settings.
Method and correction, 14 November 2026. The original exercise named ChatGPT 5.2 and a 27-image sample, but did not publish a fixed prompt, complete image list, test date, repeated runs or independent human comparison. The counts above describe that earlier reported exercise and have not been independently reproduced here. It was not a controlled comparison or a test of search rankings. This revision corrects the seatbelt patent account and removes the earlier claims that all published images train AI or that visual repetition determines AI recommendations.
Talks, interviews and articles · Discuss what this means for your brand
