Image SEO in 2026: alt text, file names, and AI visual search
AI Overviews and multimodal models now read images, not just index them. Here is how to tag and structure images so both search and AI engines understand them.
Most SEO checklists still treat images as an afterthought: compress them, add some alt text, move on. That was fine when the only thing looking at an image was Googlebot deciding whether to show it in Image Search. It is not fine anymore. AI Overviews pull images into their answers, ChatGPT and Gemini can read and reason about a picture directly, and Google's own systems increasingly treat an image and its surrounding text as one unit of meaning. Images are a content type again, not decoration.
Why images matter again
Two things changed. First, multimodal models can look at an image and describe what is in it without relying on your alt text at all, which means a mismatch between what an image shows and what your page claims is now easier for a model to catch. Second, AI Overviews and AI-generated answers frequently include a thumbnail pulled from a source page, and that selection depends on the same signals search has always used: file name, alt text, surrounding context, and structured data. Get those wrong and your image is invisible to both systems. Get them right and an image becomes a second entry point into an answer, alongside your text.
File names and alt text still do the heavy lifting
The fundamentals have not changed, they just matter more. A file called IMG_4821.jpg tells a crawler nothing. A file called red-running-shoes-side-profile.jpg tells it what to expect before it even renders the image.
- Use descriptive, hyphenated file names that match the image content, not a camera or CMS default.
- Write alt text that describes what is actually in the image, in plain language, not a keyword list.
- Keep alt text under about 125 characters. Screen readers and crawlers both truncate long strings.
- Skip alt text only for purely decorative images, and mark those with an empty alt attribute rather than omitting it.
- Put the image near the text it illustrates. Models weigh nearby text heavily when deciding what an image is about.
A common mistake is writing alt text for the crawler instead of the reader: stuffing in a product name and three keyword variants instead of describing the picture. That approach was never good practice, and it is worse now that a model can directly compare your alt text against the pixels and treat a mismatch as a quality signal.
Structured data and image sitemaps
Schema markup gives search engines explicit metadata instead of making them infer it. For product images, use the image property inside your Product schema rather than relying on the page's largest image by default. For articles, set a clear primary image in your Article or NewsArticle schema. If you run a large image library, such as a stock site or a recipe blog with step photos, an XML image sitemap is still worth maintaining. It lists image URLs Google might otherwise miss, particularly ones loaded lazily or pulled in through JavaScript.
- Add an explicit image field to any existing Product, Article, or Recipe schema.
- Use absolute URLs for image references, not relative paths.
- List your primary content images in an image sitemap if lazy loading or JS rendering could hide them.
- Confirm the image referenced in schema is the same one visible on the page. A mismatch reads as a quality problem.
Performance is still part of image SEO
A well-tagged image that takes three seconds to load hurts your Largest Contentful Paint and, indirectly, your rankings. Serve modern formats like WebP or AVIF with a JPEG fallback, set explicit width and height attributes so the browser can reserve space and avoid layout shift, and lazy-load images below the fold. None of this is new advice, but it is easy to let slip when a site redesign or a new CMS template quietly drops the width and height attributes a previous version had.
How AI visual search and multimodal models use your images
Tools like Google Lens, ChatGPT's image understanding, and Gemini's multimodal search can take a photo as the query itself, not just as something attached to a text query. When that happens, your alt text and surrounding page context are what connect the visual match back to your content and your brand. This is still an early and fast-moving area. The exact ranking weight AI Overviews give to image relevance versus text relevance is not published, and vendor behavior here changes without much notice, so treat any specific claim about how much an image swings a result with caution. What is not speculative is the baseline: a clearly named, well-described, correctly tagged image gives every one of these systems more to work with than an untagged one does.
An image with no alt text is invisible to a crawler and ambiguous to a model. Neither system can cite what it cannot describe.
None of this requires a redesign. Start with your highest-traffic pages: rename the images, write real alt text, add the image field to your schema, and check that width and height attributes are set. Do that for your top twenty pages this week and you will have covered the images most likely to show up in a search result or an AI answer.