Sean Walsh
Sean is a Director at Intelligency heading up our digital marketing and client services operations. Sean has 15+ years experiencing working both in-house and agency with brands including Lloyds, Alstom, Hitachi, Lufthansa, Viaplay, DFDS Seaways and Mercedes-Benz.
AI search systems do not look at your images the way a human browsing your website does. They scan them, extract what they can identify, and then use that information to decide whether your page is a credible source for the answer they are assembling. A photo that looks great to a person can be functionally invisible to a machine if it cannot detect what the image shows or confirm that the page text backs it up.
This is not a niche technical concern. With AI Overviews now reaching 2.5 billion monthly users, the question of what AI systems can see on your website has real commercial consequences. And images are increasingly the entry point, not an afterthought.
The change worth understanding
Google holds a patent filed in 2023 and published in April 2026 describing a mechanism where the cited source for an AI-generated answer is chosen by image match first, with surrounding text then pulled in to build the answer. In other words: the picture on your page may be what gets your page selected, and the text is used to fill out the response.
This is a patent application, not confirmed live behaviour. Google files thousands of patents and many never ship exactly as described. But as a directional signal it matters, because it is consistent with what practitioners are observing across AI search environments: visual content is playing a larger role in how AI systems identify and retrieve relevant sources, not just how humans navigate pages.
The scale of visual search reinforces why this direction makes sense. Google Lens processes around 20 billion visual searches a month. Pinterest handles 1.5 billion visual searches a month, converting at 62% better than text searches because a camera finds what words struggle to describe. When that volume of search intent is arriving through images, the systems that respond to it need to be able to read images reliably – and they are increasingly being designed to do exactly that.
What AI sees when it looks at your images
There are two distinct questions to ask about any image on your website. The first is whether the AI can correctly identify what is in the photograph. The second is whether you put the right things in the photograph to begin with.
These sound similar but they are different problems. The first is about legibility – whether the machine can read the image accurately. The second is about intent – whether the image actually communicates the story you want associated with your brand or service. A striking lifestyle image that reads beautifully to a human might tell a machine a very different story if the composition is ambiguous, if the background objects do not match your brand, or if the image contains text that cannot be read at the resolution it is being served.
A practical example: a dental clinic that uses a stock image of a general healthcare setting tells an AI system something much less specific than a clinic that uses its own photography showing its actual treatment rooms, equipment and team. The AI is reading both, but the signal from the first is weaker and shared with every other site using the same image. Stock images are a known problem for AI citation specifically because they appear on thousands of sites – the uniqueness of your image is part of what makes it citable.
The specific things that affect machine readability
There are several factors that determine how well an AI system can read and use your images. Most of these are straightforward to address:
- Image uniqueness. Your own photography, showing your actual products, premises, people or work, performs better than stock images in AI citation environments. If AI systems use image matching as a selection signal, uniqueness is part of what makes your image the one that gets selected rather than a generic version of the same thing.
- Text legibility within images. If your images contain text – packaging, signage, labels, credentials – machine vision systems read it. If that text is blurry, small, or obscured by glare, the system misses it. The same applies to before-and-after images common in aesthetics and dental contexts: the treatment result needs to be visually clear enough for the machine to register what it shows.
- What else is in the frame. AI systems do not just see the main subject. They see everything in the frame. What surrounds your product or service in a photograph forms part of the context signal the machine reads. A treatment room image with equipment visible in the background that does not match the service being described is sending a mixed message. This is worth considering when reviewing or commissioning photography.
- The text immediately around the image. The patent mechanism describes the surrounding text as what gets pulled into the AI answer after the image match. If your most relevant, specific information is buried three paragraphs below an image rather than in a caption or direct label alongside it, the machine may use weaker text to build its answer. Putting the key fact directly next to the image is a simple change with a direct effect.
- Alt text. Alt text for AI search is not just an accessibility or keyword exercise. It is a claim the machine can extract and use. Writing it as a clear description of what the image shows and what it means commercially is more useful than a keyword string, and more useful than a generic description that could apply to any image.
Which types of site this affects most
The multimodal AI search direction is most immediately relevant to businesses where the visual representation of a product or service is central to the buying decision. Aesthetics, dental, healthcare, hospitality, workspace, food and education all fall into this category. For these businesses, the photograph is not decoration supporting a text-based page – it is a primary source of information that shapes what a prospective customer believes about quality, credibility and relevance.
E-commerce is the clearest case: if an AI system can use image matching to identify a product and pull in the surrounding text to build an answer, the businesses with clear, unique, well-labelled product photography on pages with specific supporting copy have a structural advantage over those relying on shared manufacturer images and thin descriptions.
But the principle is broadly applicable. Any business that cares about being recognised and recommended by AI systems should consider whether their images are helping or hindering that goal. The question is not whether your images look good to your customers – it is whether they give AI systems the right signals to identify, retrieve and cite your pages.
What to do with this information
The most impactful single change for most businesses is replacing generic stock images with original photography. Not because stock images look bad, but because uniqueness matters in AI citation and stock images are inherently shared across thousands of sites. Your own photography of your actual work, space or team is the version only you have.
Beyond that, the practical checklist is short. Caption your images with specific, factually accurate descriptions rather than mood-led copy. Place the key information the image illustrates in the text directly alongside it, not separated by paragraphs of other content. Write alt text as a clear description of what the image shows, not as a keyword list. Check that any text visible within your images – credentials, labels, treatment names – is clear enough to be read at the sizes you are serving.
None of this requires expensive tooling or a complete photography overhaul in one go. The highest-value starting point is auditing your most important pages – typically treatment or service pages, product pages, and homepage – to check whether the images on those pages are unique, legible and contextually accurate. That review will surface the specific gaps worth addressing before worrying about anything more advanced.
