Adobe Catalog Agent: AI product discovery is a data job

The storefront your buyer sees is no longer the only storefront you have. Adobe shipped Catalog Agent for Adobe Commerce this week — a machine-readable layer that feeds structured product data to ChatGPT, Copilot, Claude, and Gemini without changing a pixel of your site. It is the least glamorous kind of AI product discovery feature, and that is exactly why it is worth reading.

What actually happened

Adobe's announcement describes Catalog Agent as enrichment that runs behind the existing storefront. It pulls from the Commerce catalog and exposes a structured layer for AI crawlers and LLM-driven discovery: specifications, attributes, pricing, availability, inventory levels, variants, compatibility, and product relationships. Product pages, imagery, and the buying journey stay as they are. Retailbiz reports it is available natively across Adobe Commerce deployment models for existing customers — no third-party integration, no replatform.

The number driving it, from Adobe Digital Insights: traffic from AI sources to US retail sites grew 125% year over year between April and June 2026, on top of a 693% jump during the 2025 holiday season. Treat that as Adobe measuring a market it sells into — but the direction matches what Shopify reported about AI traffic and structured catalogs, and it matches what we see in client analytics.

Why AI product discovery matters for your business

Here is the part that transfers to any stack. An LLM recommending your product is not reading your hero image or your carefully art-directed PDP. It is reading fields. Does this fit a 2019 F-150. Is it in stock in size 11. What is the wattage. Ships from where. If those answers live in a description paragraph, a spec PDF, or an image, they do not exist to the thing deciding whether you get recommended.

That is a data hygiene problem wearing an AI costume, and it is fixable without buying anything:

Fill the attribute fields you have been skipping. Dimensions, materials, compatibility, model years, care instructions. Most catalogs have these columns empty because no human shopper ever complained. Machines do not complain either — they just recommend a competitor who filled them in.

Publish real availability, not "in stock." Quantity and location beat a boolean. Agent-driven buyers filter hard on what can actually ship.

Emit Product and Offer schema, and keep it truthful. JSON-LD on every product page, with price and availability that match what your checkout charges. Mismatches get you dropped, not corrected.

Do it in your own system. Adobe's version is a feature of Adobe Commerce. If you are on Shopify, WooCommerce, or something custom, the underlying work is identical and vendor-neutral: one authoritative product record, complete attributes, machine-readable output. We have argued this every time it comes up — own the product data and the checkout — because the surfaces keep changing and the catalog is the asset that survives them.

Retailers spent a decade optimizing for a search engine that showed ten blue links. The next channel reads your database. Stock it.

Sources: Adobe, Retailbiz.