How to optimize images for AI search, LLMs and shopping agents

01Let search and AI crawlers fetch your pages and images

To be shown as a link in AI Overviews or AI Mode, Google says a page needs to be indexed and eligible to appear in Search with a snippet, with no other technical requirement. Both features use Googlebot: there is no separate AI crawler to allow. Other assistants search the web with crawlers of their own, and each needs to be let in.

Crawlers for search and for model training are separate, and so are the robots.txt names that control them. Blocking GPTBot or the Google-Extended token keeps your content out of model training; it doesn't take you out of ChatGPT search or Google Search. Blocking OAI-SearchBot or Googlebot does.

# Search and AI search: let them fetch pages and images
User-agent: Googlebot
User-agent: Googlebot-Image
User-agent: Bingbot
User-agent: OAI-SearchBot
User-agent: PerplexityBot
User-agent: Claude-SearchBot
Allow: /

# Model training: a separate choice, with no effect on search
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
Allow: /

Sitemap: https://www.example.com/sitemap.xml

02Put real, fast images in the HTML

Google finds images in <img> elements, through their src and srcset attributes; inside a <picture>, keep an <img> with a src as the fallback. Google doesn't index images set as CSS backgrounds. Agents that read the page's structure see the same thing: an <img> with alt text is an image, a background is decoration.

Lazy-loading is fine with the browser's own loading="lazy", which keeps the address in the HTML, but never on the main product image: it's usually the largest element on the page, and how fast it appears is one of Google's Core Web Vitals (Largest Contentful Paint). Give it fetchpriority="high" instead, and set its width and height so nothing moves when it loads.

<img src="/images/blue-leather-handbag-front-1500.jpg"
     srcset="/images/blue-leather-handbag-front-750.jpg 750w,
             /images/blue-leather-handbag-front-1500.jpg 1500w"
     sizes="(min-width: 64em) 50vw, 100vw"
     width="1500" height="1500"
     alt="Blue leather handbag with gold buckle, front view"
     fetchpriority="high">

Google Search supports BMP, GIF, JPEG, PNG, WebP, SVG and AVIF, but Merchant Center doesn't take AVIF: keep JPEG, PNG or WebP files for your product feed. Compress every image, so pages load fast for shoppers and crawlers alike.

03Give every file a descriptive name

Google's image guidelines say a file name can give it clues about what an image shows: blue-leather-handbag.jpg says far more than IMG_4032.jpg. Keep names short and descriptive, separate words with hyphens, as Google recommends for addresses, and translate them on a localized site.

Name a product's photos as a set, so each one says what it adds: blue-leather-handbag-front.jpg, blue-leather-handbag-side.jpg, blue-leather-handbag-buckle-detail.jpg, blue-leather-handbag-worn.jpg. Most stores keep the uploaded name in the image's address, so rename before you upload.

04Write alt text for people, screen readers and agents

Alt text is an image's description in the page's code. Google calls it the most important attribute for giving it information about an image, and uses it with computer vision and the page's text. Screen readers read it aloud. In the accessibility tree that browser agents read, an image is its alt text: when an agent doesn't study a screenshot, that's all it knows about the photo.

Describe what the image shows, as you would to someone on the phone. A simple formula: the product, its key detail, then the view or context.

Where it goes depends on your platform. Image SEO for AI writes alt text into each file, in the IPTC field made for it, “Alt Text (Accessibility)”. Since version 7.0, WordPress reads that field when you upload an image and fills in its alt text: on WooCommerce, your alt texts arrive with your images. Shopify doesn't read alt text from the file: each image has its own alt text field, which you can fill from the alt-text.csv list in your ZIP, or in bulk through Shopify's product CSV.

05Surround each image with the words it can't say

Google uses the page around an image to understand it: the title, the headings, the caption, the text nearby. Put each product photo on the page about that product, close to its name and description, and caption lifestyle and editorial photos (<figcaption>).

AI answers quote text, not pixels. Write as text what a photo can't state: brand, model, material, dimensions, color names, price, availability. And don't put essential information only inside an image: a size chart, a list of ingredients or care instructions saved as a picture is hard for every reader, people included.

06Describe products and images in structured data

Structured data (JSON-LD) states facts in a form machines can't misread. For a product, use Product markup with its offer, brand, GTIN and images: the image is required for Google's merchant listings, and Google recommends several high-resolution images, in 16×9, 4×3 and 1×1 formats, of at least 50,000 pixels (width × height).

{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Mila handbag, blue calf leather",
  "brand": { "@type": "Brand", "name": "Maison Lune" },
  "gtin13": "3760000000017",
  "image": [
    "https://www.example.com/images/blue-leather-handbag-front-1x1.jpg",
    "https://www.example.com/images/blue-leather-handbag-front-4x3.jpg",
    "https://www.example.com/images/blue-leather-handbag-front-16x9.jpg"
  ],
  "offers": {
    "@type": "Offer",
    "price": "240.00",
    "priceCurrency": "EUR",
    "availability": "https://schema.org/InStock"
  }
}

For images you want credited, Google's image license metadata adds an ImageObject with the image's address and at least one of creator, credit text, copyright notice or license. With a license link, Google Images can show a “Licensable” badge.

Two rules matter more than any property. The markup has to match what the page shows: same price, same images, same availability. And there is no AI-specific schema: Google says its AI features don't need special structured data. Mark up what's true and visible, nothing else.

07Sign your images: credits and license in the file

IPTC metadata is the standard photographers, news agencies and image libraries use to store credits inside image files. Google Images reads its Creator, Credit Line and Copyright Notice fields, and suggests filling in at least those three. Add a Web Statement of Rights, a link to your license terms, and Google can show the “Licensable” badge without any markup on the page.

Credits travel with the file to marketplaces, resellers and anyone who downloads it, where page markup can't follow. They're how a copy of your photo points back to you.

08Label AI-generated images honestly

Google Merchant Center requires every product image made with generative AI to say so in its metadata, such as the IPTC digital source type TrainedAlgorithmicMedia, and asks you not to remove it, in main, additional and lifestyle images alike. Google Search's “About this image” shows the AI information in C2PA Content Credentials, which ChatGPT and other AI tools write into their images, in Google Images, Lens and Circle to Search.

In the European Union, the AI Act's transparency rules (Article 50) apply since 2 August 2026. AI tools must mark what they generate in a machine-readable way (those already on the market have until 2 December 2026), and realistic AI images of real people, objects or places that could pass for genuine have to be disclosed. When an AI image shows your real product, disclosing it protects your listings.

09Clean the files, keep what helps

Photos straight from a phone or camera carry more than pixels: GPS coordinates, the camera's serial number, dates, the editing software and its history, sometimes a thumbnail of the original crop. None of it helps search, all of it adds weight, and some of it, like where a photo was taken, is nobody's business.

Remove it, then write back only what helps: your credits (step 7), the alt text and description, and the AI label when step 8 applies. Check what your platform keeps: CDNs often strip metadata from the resized copies they serve (Cloudflare has a setting to keep Content Credentials), while the original upload usually keeps it.

10Shoot for visual search

Visual search starts from pixels. Google doesn't publish separate advice for it, beyond its image guidelines, nor how Lens recognizes products. It does publish what it expects from product photos in Merchant Center, and those rules make photos easy to match.

Google's visual search now breaks a photo into the objects it contains and runs a search for each: in a lifestyle photo, every product you sell is something a shopper can be matched to.

11Feed the shopping graphs

Shopping answers draw on product data as much as on pages. Google's AI Mode shopping is built on its Shopping Graph, more than 50 billion product listings, and Merchant Center free listings can appear in Google Images and Lens. Since January 2026, the Universal Commerce Protocol (UCP) lets shoppers check out from AI Mode and Gemini, with the retailer still the seller.

In a Merchant Center feed, image_link is the main image, additional_image_link adds up to ten more, and lifestyle_image_link up to five photos of the product in use (at least 600 × 600 pixels, without text). Microsoft Merchant Center feeds help products appear in Copilot. ChatGPT accepts merchant product feeds that ask for a direct JPEG or PNG image of the exact variant, plus optional additional images, and already includes Shopify and Etsy stores. Shopify's Agentic Storefronts share each product's title, description, options, images, price and availability with AI assistants.

12Make product pages readable by agents

Google's guide to AI features notes that browser agents may read a page's screenshots, its code and its accessibility tree, the structure assistive technologies use. Chrome's documentation for developers puts it simply: agents review the accessibility tree to find what they can act on. What helps a screen reader user helps an agent.