When Shopify Agents Filter for Video or 3D

On September 15, 2026, Shopify added a media-type filter to the Catalog API that shopping agents use to find products. An agent can now ask, at query time, for products that have video, a 3D model, or both — and a product without that media is not ranked lower. It is absent from the result set before any comparison happens. If an assistant requests that media type for a 3D preview, a photo-only product is excluded from that filtered search. Ordinary searches are unchanged.

The listing job is to check which products would benefit from video or a 3D view, then add native media where a buyer experience genuinely needs it. Here is what changed, why it behaves like a gate rather than a ranking signal, and exactly which media a Shopify listing can hold today.

What changed on September 15

Shopify’s developer changelog announced three concrete changes to the Catalog API — the interface agents use through the Global Catalog and Storefront Catalog to search products across Shopify:

• search_catalog, get_product, and lookup_catalog now return video and 3D model assets on a product’s media array. Previously that array carried images only.
• search_catalog accepts a new filter, catalog.filters.media. Each entry names a media type — image, video, or model_3d. Entries combine with AND, so a request for video and model_3d returns only products that have both.
• Products that match come back with media asset URLs in the same response, including a sources array with one entry per file format, so an agent can render the product without a second request.

One rollout nuance matters. The new media types require the draft version of the Shopify global Catalog extension (dev.shopify.catalog.global) declared in the agent’s profile. Without it, the media field returns images only. On the stable version the filter itself is honored, but the response still carries images only. Shopify states the draft schema moves to stable in the next UCP release. The filter itself is already honored on the stable extension; the draft extension is required to receive video and 3D asset URLs. Shopify does not say how often agents use this optional filter.

A gate, not a ranking boost

It helps to separate two different ways media affects AI discovery.

The first is readability — how well a multimodal model can interpret your product photo. That work still matters, and we covered it in depth in our guide to product images for AI search. Readability improves what an AI can do with your listing once it sees it.

The September 15 change is the second kind: inclusion. When an agent chooses catalog.filters.media, the requested type is a hard condition for that particular search. A perfectly lit, perfectly alt-tagged image cannot satisfy a video filter, because the filter is not scored — it is satisfied only by the presence of a media type attached to the listing. On Shopify, that media is a native field you control from the product page in your admin. No markup, no feed mapping, no protocol file. This is GEO for marketplace sellers in its purest form: the unit of work is a product you already edit, and the effect is whether the product appears in a media-filtered Catalog query.

The media a Shopify listing can carry

Shopify’s product media documentation defines exactly what a seller can attach natively:

• 3D models — GLB or USDZ files, up to 500 MB. Models over 15 MB are automatically optimized, and Shopify converts files so both a USDZ and a GLB version exist (covering iOS and Android rendering).
• Videos — uploaded files up to 10 minutes, 1 GB, and 4K resolution, in mp4, mov, or webm. Alternatively, a YouTube or Vimeo link can be embedded.
• Images — existing Catalog queries keep returning the same products, and the stable extension still returns image media.

One distinction the documentation leaves open, and you should plan around: the changelog describes video assets returned with a sources array of file formats — that is Shopify-hosted media. Shopify does not document whether an embedded YouTube or Vimeo link surfaces in the catalog media array. Until that is stated, treat native uploads as the reliable path for this filter. That is a FirstShelf recommendation, not a documented rule.

Similarly, your theme controls whether shoppers see 3D models and video rendered on your storefront — all Online Store 2.0 themes built by Shopify and the Horizon family support them. Theme support is a display concern for human visitors; the media you attach to the product is what the Catalog API reports to agents. Both matter, but attachment is the step that feeds the new filter.

What agents actually do with the filter

Shopify’s own rationale in the changelog names the buyers on the other side: augmented-reality try-on, 3D product viewers, and camera-first shopping. An AR try-on literally cannot use a product without a renderable 3D asset; previously an agent had to fetch products, inspect each one for usable media, and discard the rest. The filter removes that loop, which makes media-gated flows cheaper for agent builders to ship — This reduces work for clients that already need a particular media type; Shopify does not report usage rates for the filter.

The AND logic lets a client require both video and a 3D model. Merchants should prioritize products where the intended buyer experience genuinely uses those assets.

For a seller, the decision is allocation, not ideology. You will not model a 500-SKU catalog this quarter. Start with the products where a buying decision genuinely depends on seeing dimension or motion in action.

The same fields on three platforms

Media-as-a-product-attribute is moving across ecosystems, but each surface consumes it differently right now:

• Shopify — Catalog API clients can optionally filter by media type. Video and 3D asset URLs require the draft extension.
• Google — Merchant Center’s product data specification includes an optional 3D model link attribute (virtual_model_link), US-only, pointing at a .gltf or .glb file up to 15 MB. Google’s shopping surfaces carry the field with tighter limits than Shopify’s 500 MB.
• OpenAI — the Agentic Commerce Protocol product feed for ChatGPT accepts legacy video_link and virtual_model_link metadata from existing feeds but states plainly that they are not used for product discovery, and new feeds can omit them.

Read that trio carefully, because it is a worked example of why channel-blind advice fails. The same media work has different documented roles: an optional Catalog filter on Shopify, a 3D model attribute at Google, and metadata OpenAI says its feed does not use for discovery. Shopify Product GEO is the discipline of putting each fact in the field the receiving system actually reads — and this week, the system that reads media type is Shopify’s agent catalog. For context, our earlier piece on the three attributes agents hard-filter by covered Color, Size, and Target gender; media type is a fourth gate on the same tool, and it is brand new.

What to inspect and change on your listings

This is the operating section. Run it against your top products this week.

1. Count your media. Open your ten best-selling products and count how many have any native video or 3D model attached — not embedded from YouTube, uploaded to Shopify. That count gives you a baseline for products that would be excluded if a client requested those media types.
2. Choose three to five hero products. Prioritize SKUs where dimension, mechanism, or motion decides the purchase: furniture and decor, lighting, fitted apparel, anything with a moving part or a size-perception problem.
3. Shoot the cheap version first. A 30–60 second phone video at 1080p is within Shopify’s upload limits. After upload, verify that it appears in the filtered Catalog response. Production polish can come later; presence cannot be retrofitted.
4. Add one flagship 3D model, not a fleet. GLB or USDZ, under 500 MB. Shopify converts the file to both formats and optimizes anything over 15 MB. You do not need a 3D scanner to start — Shopify’s Help Center points to partners and a scanner option for merchants who want models built.
5. Upload natively. Shopify documents mp4/mov/webm product uploads. It has not specified whether external embeds meet the new filter, so verify rather than assume they do.
6. Do not touch your images. Existing searches are unchanged, so keep useful product images alongside any new media. This is additive work, not a replacement.

Separate what is observed from what we infer. Observed: the filter, the asset returns, the file limits, and the draft-to-stable plan are all in Shopify’s documentation. Inferred: merchants whose products suit video or 3D experiences may benefit from preparing those assets before more clients use the optional filter; adoption is unknown. Recommended: the hero-SKU allocation above. Nothing here guarantees placement, citation, or sales; it changes whether an agent filtering by media type can see the product at all.

How FirstShelf can help

FirstShelf’s Shopify Product GEO work focuses on product facts a seller can control, including useful media and accurate variant details. For this new filter, inspect the media attached in Shopify, then test a Catalog API client that requests video or 3D. That check shows whether a product appears in that filtered query; it does not measure general AI visibility or guarantee a recommendation.