Agentic storefront setup

When someone asks an assistant to find them a linen shirt under a hundred pounds that ships to Germany, something has to answer. What it answers with depends on which stores it can parse: which ones state their price, availability, materials and shipping in a form a machine can read without guessing.

This is not a new discipline so much as an old one that suddenly matters more. Structured data has been good practice for a decade. The difference is that it used to buy you a slightly richer search result, and now it decides whether you are in the answer at all.

What agents need to read

Product data that is complete and structured. Price, currency, availability, condition, GTIN or MPN where it exists, and a product identifier that stays stable. Shopify emits a reasonable baseline. What is usually missing is the detail that decides a recommendation: materials, dimensions, fit, care, country of origin. If that information only exists inside a paragraph of marketing copy, or worse inside an image, it does not exist.

The fix is metafields, structured properly, and rendered into the product schema rather than only into the page.

Variant-level accuracy. A shirt is not in stock. A medium in navy is in stock. Agents that recommend an out-of-stock variant produce a bad experience and learn not to trust the source. Availability needs to be right at variant level, in the structured data, not just in the theme.

Shipping and returns as data. "Where does it ship and what does it cost" is one of the most common qualifying questions, and on most stores the answer lives in a policy page written for humans. Expressing shipping destinations, costs and return windows in structured form is unglamorous and disproportionately useful.

Content that answers questions directly. Assistants extract answers. Product and category content written as clear statements gets used. Content written as brand atmosphere does not. This does not mean writing badly, it means making sure the facts are present in plain sentences somewhere on the page.

The crawling side

Agents have to be able to fetch the page. Several things commonly stop them.

Content rendered entirely client-side is the most common. If the product detail only appears after JavaScript runs, some agents will see an empty page. Server-rendered content is the safe position.

Aggressive bot protection is the second. Rules tuned to block scrapers frequently block legitimate assistant traffic too. This needs a deliberate decision about which agents are allowed rather than a default deny.

The third is robots.txt and the newer agent-specific directives. What you allow is a commercial decision, not just a technical one, and it is worth making it consciously. Some brands want to be in every assistant's index. Some do not want their catalogue used for training. Those are different questions and they can be answered differently.

What this is worth

Honestly: nobody has reliable numbers yet, and anyone quoting you a conversion uplift for agentic commerce is guessing. What can be said is that the work is almost entirely things that are worth doing anyway. Structured product data improves search results and shopping feeds. Server-rendered content improves performance. Accurate variant availability reduces support load.

That is the reasonable case for doing it now. It is cheap, it compounds, and most of the benefit does not depend on any prediction about agents being right.