What is a discovery-ready Shopify store?

The definition

A discovery-ready Shopify store is a store prepared to be found and correctly represented on the three surfaces where customers now discover brands: organic search, AI answers, and shopping agents. Prepared means the data each surface reads, from site architecture and structured data to catalogue content, policy facts and the agent-facing files Shopify now serves on every store, is accurate and complete, and states the same facts wherever it appears.

We coined the term because the state it describes did not have a name, and the work it describes was being treated as three separate projects. An SEO agency owns the first surface, an AI visibility tool watches the second, and nobody owns the third, even though all three read the same store. Discovery-readiness is the condition of that one store being fit to be read by all of them.

Readiness is also a filter. A brand nobody is looking for has nothing to gain from being easier to find, so the standard assumes a store with real equity, real demand for its name, and a gap between the two.

The three surfaces, and when each one arrived

Organic search needs no introduction, and it still carries most category demand. A buyer who wants walking boots or a birthday necklace still types the query into Google more often than anywhere else, which is why the fundamentals of search remain the base of the discovery-ready standard.

AI answers arrived in two waves. Google announced the US rollout of AI Overviews in May 2024, and Pew Research Center measured what that did to behaviour: clicks on conventional results fell to 8 percent of visits when a summary appeared, from 15 percent without one, with links inside the summary clicked 1 percent of the time. ChatGPT search launched in October 2024, reached all logged-in free users that December, and dropped its sign-up requirement in February 2025. The buyer increasingly reads an answer instead of a results page, so the brands named in the answer collect the demand.

Shopping agents are the newest surface and the one that changed status this year. On 24 March 2026 Shopify announced that products from millions of eligible stores had become discoverable in ChatGPT by default through Agentic Storefronts, managed from the Shopify admin alongside Microsoft Copilot, AI Mode in Google Search and the Gemini app. Shopify Catalog structures the product data those platforms read, keeps prices and inventory current, and attributes orders back to the originating channel.

Being listed is automatic and being chosen is not

The March 2026 switch settled the question of whether your store is in the agentic channel. If it is eligible, it is in, and so is every competitor. What the switch did not settle is which store the agent recommends when a buyer asks for the best option in your category, because that decision runs on the quality of the data each store exposes.

An agent does not browse the way a person does. It never sees your theme, your art direction or your homepage film. A human shopper in a well-stocked shop with no signage will wander the aisles and work it out, but a buyer's agent does not wander. It reads what the store publishes in structured form, and where it finds nothing useful it moves to the store next door and reads that instead.

That is why the work underneath discovery-readiness is data work. The stores that did their foundations properly, clear architecture, complete product data, accurate policies stated where machines can read them, were most of the way there before the channel existed.

What each surface reads on a Shopify store

The three surfaces share a common base. Collection architecture tells every reader what the commercial territory is. Product titles, descriptions and metafields carry the detail an agent builds a recommendation from. Schema markup makes prices, availability and reviews machine-readable. Policy pages state the terms a buyer's agent has to relay correctly.

On top of that base, Shopify has shipped a layer that exists only for agents, and most merchants have never looked at it:

• Discovery files on every store: /agents.md, /llms.txt and /llms-full.txt, all serving identical Shopify-generated content by default, editable through theme templates since 28 May 2026.
• The Universal Commerce Protocol manifest at /.well-known/ucp, generated from store configuration, which tells agents what the store can do.
• The storefront MCP endpoint, which lets any compatible agent search the catalogue, read policies and manage a cart over plain JSON.
• The Knowledge Base app, the store's question-and-answer feed for AI surfaces, auto-populated from store settings whether or not the merchant has ever opened it.
• An agentic discovery sitemap linking the three discovery files, listed inside the standard sitemap index.

Be realistic about the weight this layer carries. Ahrefs analysed 137,000 domains and found 97 percent of published llms.txt files never receive a single fetch, and Google has said plainly that it does not use them. The files matter because Shopify's own agentic systems read them, and because they are where a store's data discipline becomes visible. The heavier levers for being found and represented correctly are the catalogue data, the Knowledge Base and the schema, with the files as the cheap, fast layer on top.

The readiness test you can run this afternoon

You can establish where a store stands in under half an hour, without a developer and without logging into anything beyond your own admin.

1. Read your discovery files. Load /agents.md, /llms.txt and /llms-full.txt on your own domain. If all three are identical apart from one line naming the path, the store is serving Shopify's default, which contains the transaction protocol and no brand content at all.
2. Ask your store a question. In the Shopify admin, open the Knowledge Base app if it is installed and read the auto-generated answers against your actual policies, starting with returns and shipping. If it is not installed, that is the first finding.
3. Check the schema. Run your homepage and one key collection through Google's Rich Results Test and note what is present. On most stores, product schema exists and FAQ, aggregate rating and return policy markup do not.
4. Read a top collection as a stranger. If the page is a product grid with no introduction, no buying guidance and no answers to the category's obvious questions, an agent following that link learns nothing it can cite.
5. Run two category queries in ChatGPT. Use the questions a buyer would ask, without your brand name, and note who is recommended, in what language, and whether the details are accurate. The result is your current position in the channel.

Record what you find, because the same five checks run after the work is done are the before-and-after. Everything in the list is inspectable from outside except the Knowledge Base, which is why we treat the feed it powers as the first thing to capture properly, and there is a longer testing guide covering how to query the endpoints agents call.

What we find when we run it

We run this test as the opening step of every agentic readiness engagement, and the pattern across live stores is consistent. The discovery files are the untouched default on almost every store we have read. The Knowledge Base, where installed at all, is running on auto-generated answers nobody has reviewed.

The auto-generation is the part merchants underestimate, because the feed serves answers whether or not anyone has curated it, and it serves the errors too. When we captured a live store's policy feed against a set of 28 questions a shopper would ask, nearly half came back with no answer at all. Among the answers that did come back, the returns policy quoted a window that did not match the store's published terms, a click-and-collect question was answered with a raw internal location ID instead of an address, and a gift service was advertised that the store does not offer.

Every one of those answers was being served to shopping agents that day, on a store whose team had no idea the feed existed. None of it was anyone's fault, because nothing had ever prompted anyone to look. That is the state the March 2026 switch put most of the platform in: enrolled in the channel, represented by defaults.

The shape of the fix is worth seeing side by side. Before curation, the click-and-collect question came back as a system record, along the lines of "Location ID: 627296…, Pickup price: 0.00, Pickup time: twentyFourHours". A curated answer to the same question reads like a person wrote it: collection is free from the flagship store, usually ready within 24 hours, with the address written for a buyer. The illustrative wording will differ by store, and the gap between the two is the same everywhere: one is data that leaked, the other is an answer someone stands behind.

A worked example: one fact, four places

The clearest way to see what readiness means in practice is to follow a single fact through the store. Take a returns window of 21 days, the kind of term every store has and every buyer's agent will be asked about.

Four places, one fact, and the number has to be identical in all of them. A mismatch is worse than an omission, because an agent that reads 21 days in one place and 14 in another has learned that the store's data cannot be trusted, and an agent that finds nothing at least knows to say so. This is also why the work resists being split across three suppliers: whoever owns the returns page rarely owns the schema, and nobody owns the feed.

Multiply that discipline across shipping terms, guarantee terms, sizing, stockists, bestsellers and the questions specific to your category, and you have the actual substance of discovery-readiness. It is unglamorous, it is entirely doable in-house once someone owns it, and it is precisely the kind of detail that decides which store the agent recommends.

The order of the work

The layers read each other, so the order matters more than the effort. Knowledge Base answers are generated from store settings and policies, catalogue search returns whatever the product data contains, and the discovery files are only as useful as the facts behind them. Fixing the surface layer first means restating errors more visibly.

1. Architecture and catalogue first: consolidated collections that map to real demand, product data with the detail a recommendation needs, accurate policy pages.
2. Structured data second: product, FAQ and policy schema stating the same facts the pages state.
3. The agentic layer third: Knowledge Base answers reviewed and corrected, a custom agents.md carrying the brand and its actual terms, robots.txt letting AI crawlers in.
4. Verification last: the same five checks re-run, and the category queries repeated a few weeks later.

This is the same fundamentals-first argument we make for organic search, extended one surface further. Nothing about the agentic layer changes the hierarchy, it adds a new reader at the end of it, and the guidance on what a custom agents.md should contain follows the same principle.

What discovery-readiness will not do

It will not guarantee citations or recommendations. Shopify is explicit that Knowledge Base curation corrects what AI platforms say about a store, without changing how often it appears. Appearance is decided by the model's wider view of the brand, which is built from authority, coverage and entity clarity over time.

It will not restore the click volumes of 2023. Pew's data shows AI summaries reduce clicks even for the sources they cite, and no amount of readiness reverses that. The realistic goal is to be the brand the answer names, with the store data that lets an agent transact correctly when the buyer acts on it.

And it is not a state you reach once. The protocol versions in the agent files update, models change how they read, and Shopify is still building the channel out. Readiness is maintained the way any fundamental is maintained, which is also the honest reason it belongs with the rest of the foundational work.

Questions we hear about discovery-readiness