Agent-Readiness for Shopify catalogues

It makes your products retrievable and correctly described when someone asks ChatGPT, Gemini or Perplexity what to buy in your category. We audit what the assistants say today, rebuild your product data as structured attributes, publish it as Shopify metafields, JSON-LD and feeds, and measure the change. It does not make an AI buy for your customers. Forrester (2026) finds very few consumers let an agent complete a purchase, so we sell being findable, not being transactable.

Because the assistant never entered your product into its candidate set. It retrieves from structured product data and filters on attributes, so a product whose data says sulphate-free only in a paragraph of copy is invisible to a question that asks for sulphate-free. Exclusions are the most common gap, and an answer has three or four slots with no page two. A competitor with more complete data takes the slot, whichever product is better.

Three things have to be true. Your products carry the attributes shoppers ask about as structured fields, including exclusions such as free_from, rather than as sentences in a description. Those fields are published where assistants read them: Shopify metafields, JSON-LD on the product page and the Merchant Center feed. And your catalogue is in ChatGPT's product feed, which OpenAI states runs on ACP and includes Shopify merchants through Shopify Catalog. The audit shows which of the three you are missing.

Probably yes. OpenAI states that ChatGPT product discovery runs on ACP feeds and that Shopify merchants are included automatically through Shopify Catalog. We check this in the audit before charging for anything connected to it. The feed is usually already being sent. The gap is what it says about each product: if your attributes are missing, the feed delivers that absence to the assistant faster. Enrichment is where the work is.

The AI visibility audit is planned at 1–2 weeks elapsed and three days of effort: score the catalogue, run 15–25 category questions across ChatGPT, Gemini and Perplexity twice each, then deliver the gap report and fix list. The build is planned at 4–8 weeks: taxonomy design 1–2 weeks, enrichment 2–4 weeks with every field verified by a person, publishing about a week, then verification at delivery and again at 90 days.

It fits Shopify or Shopify Plus brands with 500 or more SKUs in a category where customers filter on real attributes: ingredients, materials, fit, compatibility, certifications. It does not fit marketplace-only sellers, who do not control their product schema, or commodity catalogues where products differ only by size or colour. With under about 150 SKUs the audit alone is usually enough, and we will say so rather than sell you a build.

It is different, and the two are complementary. SEO gets pages ranked on content and authority signals. An assistant does not read your product page; it retrieves from structured product data and filters on attributes, so a feed with no free_from field cannot be found however well the page ranks. This is sometimes sold as GEO or AEO. The structured attributes we add also improve conventional search, so nothing in the build works against your SEO agency.

We run the same 15–25 category questions across the assistants before we start and after we finish, each twice on different days because assistant answers vary. The result is a side-by-side: where you appeared before, where you appear now, and which competitors you displaced. We also tag traffic arriving from assistant surfaces in your analytics. The limit: we can see what you own and any first-party identity you capture. We cannot see who the assistant's user was.

No, and nobody can. Assistant answers change from day to day and the feed protocols moved twice within six months. What we commit to is the method: the same questions before and after, screenshots of both, and a report at delivery and at 90 days. If the enrichment has not changed where you appear, the report shows that too.

You do. Shopify, Merchant Center, ACP and any model-provider accounts are in your name and named at proposal stage, because credentials held by an agency stop working when that agency does. LLM usage during enrichment runs on your account or is passed through at cost. Allowing AI crawlers in robots.txt is your decision, signed off in writing. Afterwards, a quarterly re-scan takes half a day per quarter inside a retainer, or your team maintains the data with the handover document.