Shopify image alt text for AI shopping agents

Image alt text is the cheapest 6 ranking-spread points in the entire 15-signal suite — and the one fewest stores collect. Every Shopify store CatalogScan has scanned has alt-text coverage below 30%; many are at 0%. The signal is purely "did your team type a sentence per image" and the payoff is direct: AI shopping agents use alt text to pick up product attributes that aren't in the description — color shown in the image, fabric drape, model height-to-product ratio, the angle of view, whether a shoe is photographed on or off a foot. When a query is "running shoe with a wide toe box visible in the side profile," the only signal in your catalog that can match it is alt text.

Last updated 2026-04-30 · Ranking-spread signal · 6 pts

What it is

Every product image in Shopify has an alt field, exposed in three places:

• The Shopify Admin → Products → product → image gallery → click an image → "Edit alt text."
• The Storefront API Image object's altText field.
• The public /products.json feed at products[].images[].alt.

When a product image renders in HTML — either via {% raw %}{{ image | image_tag }}{% endraw %} in Liquid or via your Hydrogen/Next.js component — the alt field renders as the alt="" attribute on the <img> tag. Browsers and screen readers and AI agents all read the same field.

<img alt="" src="...">
{ "alt": null }

<img alt="IMG_3492.jpg" src="...">
{ "alt": "main-shoe-1.png" }

<img alt="Wool Runner" src="...">
{ "alt": "Wool Runner" }

<img alt="Wool Runner in Natural Grey, side profile, on white background" src="...">
{ "alt": "Wool Runner in Natural Grey, side profile..." }

Why AI shopping agents care

• Attribute pickup the description misses. Descriptions tell agents what the product is. Alts tell them what the image shows. A description for a t-shirt might say "100% cotton, crewneck, regular fit"; the alt for the third image says "shown in heather grey, on a 5'10" model, hits at the hipbone." That second sentence answers a different class of customer query than the description.
• Per-image specificity. One product has 3-8 images, each captured from a different angle, on a different model, in a different colorway. Each alt is a separate text-matching anchor. Agents that surface specific images (Perplexity, Google AI Mode in image-rich answers) pick the image whose alt best matches the query — without distinct alts, you score the same on every image and the agent picks one effectively at random.
• Trust priming. AI rankers note alt-text coverage as a "well-tended catalog" signal — a proxy for the kind of operator who's also going to have GTINs, descriptions, and review schema in good shape. The alt-text signal correlates highly with overall catalog hygiene; rankers use it as a fast pre-filter.
• Accessibility halo. Alt text is also the WCAG accessibility requirement for images. AI rankers have begun lightly weighting accessibility-positive signals (proxy for legitimacy and trust), and broken alt-text coverage downweights you on the same axis. One field, two systems benefit.

How to test it on your store

One curl + jq measures coverage in one second:

curl -s https://yourstore.com/products.json?limit=30 \
  | jq '[.products[].images[].alt] | length as $t
        | map(select(. != null and . != "")) | length as $f
        | "\($f)/\($t) = \($f * 100 / $t | floor)%"'

Expected output: a fraction and a percentage. Anything under 80% is a problem; anything under 30% is the most common case. The free CatalogScan scan runs this same check across the first 30 products and reports the percentage with examples of the worst offenders.

What good alt text looks like (8-15 words, recipe)

Each is 8-15 words, leads with the product, then the colorway/variant, then the distinguishing visual attribute. Skip "image of," "photo of," "picture of" — the parser knows it's an image. Skip emoji and decorative punctuation — they're noise in the embedding match.

How to fix it

// Pseudocode — fill in client + auth
for (const product of allProducts) {
  for (const image of product.images) {
    const alt = `${product.title} in ${product.options[0].values[0]}, view ${image.position}`;
    await admin.productImageUpdate({ id: image.id, alt });
  }
}

5 mistakes we keep finding