Technical AEO

    Images, Alt Text and Multimodal AI Search

    How AI search understands your images, from alt text and captions to visual search in Google AI Mode, with markup examples and an image audit checklist.

    Saarthak Srivastav

    Saarthak Srivastav

    · 6 min read

    Key takeaways

    • AI search now takes images as input, but it still understands your images mostly through the text around them.
    • Write alt text that states what the image shows and why it matters on that page; use empty alt for decorative images.
    • Never leave key information only inside an image: charts, pricing tables and specs need a text version.
    • Use real img or picture elements for content images; Google has said CSS background images aren't picked up for image search.
    • For visual search, clear product photos, consistent imagery across retailers and visible branding make your products easier to match.

    AI search can now look at images, but when it reads your website, it still understands your images mostly through text: alt text, captions, filenames and the copy around them. Google added image input to AI Mode in 2025, so a photo can start a search, and the pages it finds still have to explain their images in words. This post covers how AI systems interpret images, how to write alt text that works, the markup to use, and how to prepare for visual search.

    How AI systems use your images

    Images reach AI answers through two separate paths, and each one needs something different from you.

    1. Images as input. A user photographs a product, a plant or an error message and asks about it. Google announced multimodal search in AI Mode in April 2025, combining Lens with a custom Gemini model, and said AI Mode runs several queries about the whole image and the objects in it. In September 2025 Google described a "visual search fan-out" that also picks up secondary details in an image. The answer then draws on web pages that match what was identified.

    2. Images on your pages. When an assistant reads your page to answer a question, it usually works from the HTML. Google says it combines alt text, computer vision and the surrounding page content to understand an image. Assistants that fetch pages as text, which is most of them outside Google, get even less: the alt attribute, the caption and nearby copy.

    The common thread is text. Even a perfect product photo only becomes useful when the page around it says clearly what it is. For more on how AI Mode works generally, see the Google AI Mode guide.

    Writing alt text that works

    Good alt text states what the image shows and why it matters on that page, in one plain sentence. Google's image guidelines ask for useful, information-rich alt text in context and warn that keyword stuffing can be treated as spam.

    Image Weak alt text Better alt text
    Product photo alt="shoe" alt="Trailrunner 4 trail shoe in olive, side view showing the lugged sole"
    Chart alt="chart" alt="Bar chart: support tickets fell each quarter of 2026 after the help center relaunch"
    Screenshot alt="dashboard screenshot" alt="Example dashboard showing share of voice by assistant for the last 30 days"
    Team photo alt="team" alt="Example Co. support team at the Austin office"
    Decorative divider alt="decorative line" alt="" (empty, so screen readers and crawlers skip it)

    Rules of thumb:

    • Lead with the subject; include product or brand names where they belong naturally
    • Describe what's relevant to this page, not every visual detail
    • Don't start with "image of" or "picture of"
    • Keep it to roughly one sentence; put long explanations in a caption or body text
    • Write alt text in the page's language on localized sites

    Alt text is also an accessibility requirement, which is reason enough to do it well. See page speed, accessibility and AI fetchers for how accessibility and machine readability overlap.

    Put the information in text, not just pixels

    Anything a buyer might ask an assistant about must exist as text on the page, not only inside an image. This is often a bigger problem than weak alt text, and alt text can't fix it.

    Common offenders:

    • Pricing tables exported as images. Assistants can't quote plan limits they can't read. Use an HTML table.
    • Infographics and charts. Add a two- or three-sentence summary of the finding and, for data, an HTML table below the chart.
    • Spec sheets and comparison graphics. Rebuild them as text tables; see product pages for AI assistants.
    • Text baked into hero images. Headlines and value propositions should be live text.
    • Screenshots of instructions. In docs, write the steps out and use the screenshot as support.

    Markup that makes images readable

    Use standard HTML image elements with real attributes, so every crawler can find the image and its description. Google's Search Relations team has said that content images should be img or picture elements in the DOM, and that CSS background images aren't picked up for image search.

    <figure>
      <img
        src="/images/trailrunner-4-olive-side.webp"
        srcset="/images/trailrunner-4-olive-side-800.webp 800w,
                /images/trailrunner-4-olive-side-1600.webp 1600w"
        sizes="(max-width: 800px) 100vw, 800px"
        alt="Trailrunner 4 trail shoe in olive, side view showing the lugged sole"
        width="1600" height="1000"
        loading="lazy">
      <figcaption>Trailrunner 4 in olive. The 5mm lugs are designed for loose, muddy trails.</figcaption>
    </figure>
    

    Checklist for each important image:

    • img or picture element, not a CSS background, for anything that carries meaning
    • Descriptive filename (trailrunner-4-olive-side.webp, not IMG_4821.jpg); Google calls filenames a light hint, but it's free
    • Alt text and, where useful, a visible caption
    • Explicit width and height to avoid layout shift
    • Native loading="lazy" for below-the-fold images, but not for the main image above the fold
    • Image placed near the text that describes it
    • Stable image URLs that don't change on every deploy

    If your images are injected by JavaScript after load, many AI crawlers won't see them at all. See JavaScript rendering and AI crawlers.

    Image sitemaps

    An image sitemap can help Google find images that are hard to discover in your HTML, such as gallery images loaded on interaction. Since August 2022, Google only reads image:image and image:loc; the caption, title, geo_location and license tags were deprecated.

    <url>
      <loc>https://www.example.com/products/trailrunner-4/</loc>
      <image:image>
        <image:loc>https://www.example.com/images/trailrunner-4-olive-side.webp</image:loc>
      </image:image>
    </url>
    

    Declare xmlns:image="http://www.google.com/schemas/sitemap-image/1.1" on the urlset element. More on sitemaps in XML sitemaps and lastmod for AI search.

    Structured data

    Reference your key images in structured data so their role is explicit: image on Product, logo on Organization, image on Article. See product schema and merchant feeds.

    {
      "@context": "https://schema.org",
      "@type": "Product",
      "name": "Trailrunner 4",
      "brand": { "@type": "Brand", "name": "Example Co." },
      "image": [
        "https://www.example.com/images/trailrunner-4-olive-side.webp",
        "https://www.example.com/images/trailrunner-4-olive-top.webp"
      ]
    }
    

    To be the page that matches a photographed product, your product has to be visually recognizable and your pages have to name it clearly. Nobody outside Google knows exactly how Lens matches images to pages, so focus on the parts you control:

    • Clear, well-lit product photos from several angles, including the angles people shoot in stores (front of packaging, labels, logos)
    • Visible branding on the product and packaging where it's appropriate
    • Consistent imagery and names across retailers and marketplaces, so every listing reinforces the same product identity
    • Distinct pages for distinct variants (colors, sizes, models) with alt text that names the variant
    • Original photos rather than the same manufacturer image every reseller uses, where you can

    Video works the same way; see YouTube and video in AI answers.

    Example: auditing a product page's images

    This is an illustrative example with made-up details.

    A cookware brand audits its best-selling skillet page:

    Finding Fix
    Hero image is a CSS background Moved to a picture element with alt text naming the model and size
    Dimensions and weight shown only in an infographic Added a specs table in HTML; kept the graphic below it
    Eight gallery images with alt="skillet" Rewrote each to describe the angle and feature shown
    Gallery loads on click via JavaScript Rendered the first four images in the initial HTML
    Retailer listings use a different product name Aligned product names with retail partners

    None of these changes alter how the page looks to shoppers, and all of them make the page more readable to crawlers and assistants.

    The bottom line

    Multimodal AI search lets people start with a photo, but text still carries your images into AI answers. Write alt text that says what each image shows and why it matters, put every important fact in live text rather than inside graphics, use real image elements with sensible filenames and structured data, and keep product imagery clear and consistent everywhere your products appear. It's mostly accessibility done properly, and it pays off with crawlers, assistants and people alike.

    FAQ

    Frequently asked questions

    Does alt text help with AI search?

    Yes. Google describes alt text as the most important image metadata it uses, alongside computer vision and page content. Many AI systems that read pages as text only see your images through alt text and captions.

    Can ChatGPT and other assistants see images on my website?

    Some assistants can analyze images a user uploads, but when they read web pages through search and fetching, they mostly work from the HTML text. Assume the text around an image is what gets used.

    What is multimodal search in Google AI Mode?

    Google added image input to AI Mode in April 2025, combining Google Lens with a Gemini model so people can photograph or upload something and ask about it. In September 2025 Google described a visual search fan-out that runs several searches about details in an image.

    Should I still use an image sitemap?

    It can help Google discover images that aren't easy to find in your HTML. Google now reads only the image:image and image:loc tags; caption, title, geo_location and license were deprecated in 2022.

    How long should alt text be?

    Usually one short sentence. Describe what matters about the image in context, include a product or brand name where natural, and skip phrases like 'image of'.

    • images
    • alt text
    • multimodal search
    • Google Lens
    • image SEO
    Saarthak Srivastav

    Written by

    Saarthak Srivastav

    AEO & GEO, AEO-Rank

    Saarthak Srivastav writes and reviews articles on the AEO-Rank blog, drawing on what the AEO-Rank team sees tracking AI answers across ChatGPT, Gemini, Claude, Perplexity and Grok.