When a shopper asks an AI agent to "show me sustainable running shoes with an aggressive tread," the recommendation they receive depends entirely on how well the machine can parse and interpret your product media. To solve this, the AI visibility platform Pendium helps merchants adapt their catalogs so multimodal LLMs can evaluate 3D models and video assets alongside traditional text descriptions. By standardizing your 3D files to the GLB format, configuring the Shopify Model3d object with technical alt text descriptions, and structuring native videos through correct Liquid elements, your store can secure the single recommendation slot that AI shopping assistants offer to high-intent buyers in 2026.
How multimodal AI agents read Shopify store catalogs
In early 2026, Shopify enabled native support for agentic commerce, bringing files like /agents.md and /llms.txt directly into the merchant ecosystem. Multimodal models like GPT-4o, Claude 3.5 Sonnet, and Gemini Pro do not just crawl raw text; they actively ingest and categorize product catalogs by evaluating product images, videos, and 3D assets in tandem. If your shop has high-quality media but lacks structured metadata, those assets remain completely invisible to scraping bots.
At the International Conference on Learning Representations (ICLR) 2025, a Shopify engineering presentation at ICLR 2025 explained how the platform utilizes multimodal Large Language Models to organize and build a unified, structured understanding of billions of product listings. When an AI crawler indexes your store, it attempts to match visual data—such as fabric texture, physical dimensions, or shoe tread patterns—with complex semantic queries. Without explicit labels connecting these visual features to the database, the AI assistant cannot verify if your product matches the buyer's criteria.
This represents a fundamental shift from traditional search engine optimization. In our analysis at Pendium, we find that stores often pair rich media with empty or generic alt attributes, forcing AI crawlers to discard the products from structured comparison tables. To prevent this, merchants must pair structured media formats with structured data frameworks like JSON-LD schema. For instance, knowing how to map Shopify warranty data to JSON-LD for AI durability queries ensures that the physical properties seen in your product videos are mathematically validated in your store's text schema.
Configuring the Model3d object on Pendium-monitored storefronts
Shopify uses a dedicated API object called Model3d to represent and render hosted three-dimensional assets. When third-party AI agents crawl a merchant's store, they access this data via the Storefront API. According to the Shopify API documentation for the Model3d object, accessing these 3D listings requires the unauthenticated_read_product_listings access scope, meaning your store's theme must serve these objects cleanly without blocking bot scripts.
Populating the alt field for machines
The alt string inside the Model3d object is your primary surface for explaining a three-dimensional asset to an AI agent. While human shoppers orbit the model in their browser or use augmented reality to project the item into their living room, the LLM reads this alt string to understand what the asset represents.
Avoid subjective marketing phrases like "gorgeous modern sofa." Instead, feed the machine technical, descriptive parameters.
| Asset Type | Human-centric Alt Text (Avoid) | Machine-optimized Alt Text (Preferred) |
|---|---|---|
| Footwear | Waterproof hiking boot for outdoor fun | Mid-top waterproof hiking boot with Vibram rubber outsole, aggressive 5mm multi-directional lugs, and reinforced synthetic upper |
| Furniture | Stunning blue accent chair | Mid-century modern armchair in navy blue boucle fabric with tapered solid walnut legs, high-density foam padding, and 120-degree seat angle |
| Equipment | Durable camera backpack | 20L weather-resistant camera backpack with modular padded dividers, dual side-access panels, and dedicated 16-inch laptop sleeve |
Using descriptive, physical dimensions within your alt text allows AI assistants to build a precise spatial and functional profile of your product.
Setting the media content type
The mediaContentType field in the Model3d object must be explicitly declared as MODEL_3D so the indexing engine does not confuse it with a flat image or a static background thumbnail. Shopify stores often use lazy loading scripts or speed-optimizing apps to reduce initial page load. However, as noted in our research on why Shopify speed apps hide your products from AI, these speed-optimizing tools can accidentally hide the native script calls of the Model3d object behind JavaScript blocks that AI bots cannot trigger. Ensure your theme loads the Model3d parameters directly in the server-side Liquid payload rather than executing them exclusively post-interaction.
Formatting product video metadata using Liquid objects for Pendium visibility
Product videos are a massive conversion tool for human buyers, and they provide incredible context for multimodal engines looking at your page. Simply dropping raw MP4 files into a standard Rich Text editor or description field is a major mistake. AI engines crawl specific theme loops and API arrays, which means your videos must be declared through official Shopify Liquid objects to be indexed correctly.
Native video requirements
To ensure your product videos render fast enough to prevent AI bot timeouts during crawling, you must keep them under specific parameters. The Shopify Help Center guidelines state that your product media can be up to 5000 x 5000 px, or 25 megapixels, with a file size limit of 20 MB for quick loading. For most mobile and desktop themes, a square 1:1 aspect ratio at 1080x1080 px or 2048x2048 px works best.
Keep the duration between 5 and 30 seconds, focusing on physical properties like fabric drape, texture, and physical movement. Since Shopify product galleries auto-play video on mute by default, design the visual structure so that an AI engine or a human customer can extract the full context of the product's function in the first five seconds without needing sound.
Handling YouTube and Vimeo embeds
If you host your videos externally, do not simply use custom HTML iframe tags. Instead, write native Liquid loops in your product.media templates using the external_video object. This ensures the output maintains structured metadata hooks.
{% for media in product.media %}
{% case media.media_type %}
{% when 'external_video' %}
<div class="product-media-wrapper" data-media-id="{{ media.id }}">
{{ media | external_video_tag: class: 'youtube-vimeo-player' }}
</div>
{% when 'video' %}
<div class="product-media-wrapper" data-media-id="{{ media.id }}">
{{ media | video_tag: controls: true, autoplay: false, loop: true }}
</div>
{% endcase %}
{% endfor %}
By rendering your external and native videos through these standardized Shopify elements, you supply clean, machine-readable nodes containing direct links via video_source arrays. This allows scraping engines to fetch and review the video components instantly.
Standardizing modeling software pipelines for AI recommendation engines
Creating 3D models for commerce requires a different optimization pipeline than standard gaming or film modeling. Multimodal search platforms require highly optimized files that load instantly. If an AI agent attempts to index your storefront but hits a 500 MB unoptimized model, it will drop the connection and omit the product's spatial data from its database.
According to the Shopify partner resources, different 3D design tools have highly specific purposes. When designing product models, understand which software outputs meet commercial guidelines:
- ZBrush is an intuitive sculpting program great for designing soft surfaces and high-detail organic features, but it is not ideal for building clean, optimized meshes.
- Blender is a free alternative that is perfect for previewing, animating, and optimizing your meshes to reduce vertex counts before final export.
- Substance Painter is a highly robust tool that allows artists to paint PBR textures directly onto 3D surfaces and export the final assets natively in the GLB format.
Always export your completed files as GLB. This format bundles the 3D geometry, node hierarchies, and physically based rendering (PBR) texture maps into a single file. Shopify accepts both GLB and USDZ formats, but you only need to upload the GLB file to your admin portal; Shopify will automatically compile the USDZ equivalent for iOS and macOS AR systems. To keep your page speeds high and your AI crawlers happy, compress your final GLB files to under 15 MB.
Building a complete data map with Pendium for AI visibility
Standardizing your 3D assets and video objects is only one part of the optimization process. Multimodal engines do not look at visual media in isolation; they compare the visual metadata against the rest of your storefront's structured parameters. When you optimize your media, you should also ensure that your logistical data matches perfectly.
For example, implementing clean delivery structures by mapping Shopify delivery data to schema for AI search recommendations ensures that when Perplexity or Gemini serves your optimized product video, it can also confidently list accurate shipping windows and geographic availability.
If you want to know how AI engines are currently perceiving your store's media files and structured catalog data, you can run a free diagnostic scan. Using the Agent Experience Engine — Control How AI Perceives Your Brand at Pendium, you can evaluate how ChatGPT, Claude, Gemini, and other major systems view your brand.
The free AI Visibility Scan takes just two minutes and analyzes your online presence without requiring a credit card. By simulating real buyer personas, the platform reveals exactly where competitors are winning AI recommendations due to metadata gaps—giving you the exact blueprint you need to optimize your Shopify media catalog.
Visit Pendium.ai to run your free AI Visibility Scan today, or book a live platform demo with the team at the Pendium demo portal.