Yes, alt text and image file names still matter for AI search, and they matter more now that so much of a page's proof lives in photography and video instead of paragraphs. An AI crawler cannot see a photo. It reads whatever text sits next to it, and a business that skips that text is handing a blank page to the exact system deciding whether to recommend it.
Most sites treat alt text as an accessibility checkbox or skip it outright. Both choices cost more than a compliance flag. A crawler reading a page for facts to quote treats an image with no alt text as a gap in the record, not a missed formality.
What an AI Crawler Actually Reads From an Image
GPTBot, ClaudeBot, and PerplexityBot do not render a page the way a browser does, and they do not look at pixels the way a person does either. They read the HTML around an image: the alt attribute, the file name, the surrounding caption, sometimes a figure element tying a caption to a specific photo. That is the entire dataset an image contributes to a crawler's read of a page, covered briefly in our post on what an AI crawler actually sees when it hits your homepage.
A photo of a finished kitchen remodel tells a human visitor everything in half a second. It tells a crawler nothing unless the alt text states what the photo shows, and the caption states what claim it is making. Beautiful photography with no text layer is invisible to the system most likely to decide whether a buyer ever sees it in an answer at all.
Alt Text Written for a Screen Reader Is Already Written for a Machine
The accessibility rule for alt text and the AEO rule for alt text are close to the same rule, describe the photo the way a sighted person would describe it to someone who cannot see it. Specific, factual, no keyword stuffing, no filler like "image of." A screen reader user and a language model are both working from text standing in for a picture, and both are better served by the same discipline.
Alt text stuffed with repeated keywords, Phoenix kitchen remodel, Phoenix kitchen contractor, Phoenix remodeling, fails a screen reader and reads as noise to a crawler that has gotten good at telling the difference between a real description and an attempt to game one. Write what is actually in the frame, including the detail that makes it specific: the material, the room, the city if the shot is location specific.
File Names and Captions Are Part of the Same Text Layer
An alt attribute is not the only text an image carries. The file name a crawler fetches is visible in the HTML, and a name like IMG4821.jpg says nothing while scottsdale-kitchen-remodel-quartz-island.jpg adds a second, independent confirmation of the same fact the alt text states. A caption sitting visibly under the image is a third. None of these fields have to work alone, and a fact confirmed in three places reads as more reliable than the same fact asserted once.
This is the same logic behind the entity consistency work: agreement across surfaces raises confidence, and a photo's file name, alt text, and caption are three small surfaces that either agree or contradict each other on what the image is showing.
Where the Gap Costs the Most: Portfolio and Production Pages
A production reel or a photography portfolio page is often the least text dense page on a studio's site and the most image dependent, which makes it the page where this gap costs the most. Our post on staff portraits as AEO work covers the same principle from the identity side: a photo only does machine readable work once it is described in text a crawler can read. A portfolio full of stunning, unlabeled stills is a portfolio a generative engine cannot describe to anyone asking who does this kind of work well.
The fix is not less photography. It is a caption and alt text pass on every image that is doing real selling work: what the shot is, what it demonstrates, and where relevant, the city or venue it was shot in.
Structured Data Reinforces the Same Text, It Does Not Replace It
ImageObject schema can label a photo's creator, date, and description in structured form, the same reinforcement pattern covered in our post on the schema worth shipping. It is a confirmation layer, not a substitute. Schema stating a fact that no visible text on the page supports does the image no good, because a crawler that finds the claim in code but not in the rendered page has no independent confirmation to check it against.
A Practical System, Not a One Time Pass
Alt text and file naming are easy to do once at launch and easy to forget on every image added afterward. A working system names files descriptively before upload, writes alt text as part of publishing rather than after, and treats a new portfolio piece or blog image the same way a new page gets a title tag, as a required field, not an optional one.
Getting this right does not guarantee an image gets pulled into an AI answer. Structure only. No ranking or citation guarantees. What it does is remove one more gap between what a business actually built and what the systems deciding who gets recommended are able to read.
Auditing alt text, file naming, and the caption layer across a site's highest traffic and most image heavy pages is part of the technical work inside the UM Media SEO / AEO / GEO Foundation, from $7,500. Structure only. No ranking or citation guarantees.