Discovery

Discovery in agentic commerce is how AI agents find and surface products in response to queries. Getting discovered by systems like ChatGPT, Perplexity, Claude, and Gemini means being surfaced across multiple routes: a search engine's index, the AI's own crawler, a merchant feed, or product data exposed through a commerce protocol.

Being found is three separate results, and they do not move together. A search engine can index your page, a specific AI system can retrieve it, and an answer can cite it, and any one of those can be true while the others are not. Clearing one does not clear the next, so it pays to treat them separately.

Indexed: are you in a search index?

A search engine's crawler has fetched your page, read it, and stored it in that engine's index. This is traditional SEO, and it still matters: several AI systems answer from a classic search index rather than crawling the whole web themselves, so being in Google's or Bing's index is often the floor they build on. It is also per engine. Sitting in Google's index says nothing about Bing's, and an AI system reads whichever index it is wired to.

Retrievable: can a system fetch you?

A specific AI system can actually reach and read your product when it needs to. This is not guaranteed by being indexed. Each system, ChatGPT, Perplexity, Claude, Gemini, runs its own crawlers on its own refresh schedule, so a brand-new page can be in Google's index and still unreadable to an assistant whose crawler has not reached it yet. Which crawlers those are, and whether your store is quietly blocking them, is an access question covered on the Readiness guide.

Fetching your page is only one of the routes in. An agent may also read your product from a shopping index, a merchant feed, or a structured product object exposed through a protocol like ACP or UCP. If an agent relied on a protocol alone, being on that protocol is what would make you discoverable to it. In practice agents pull from several of these at once, so being retrievable means being reachable through whichever routes a given agent actually uses, not any single one.

Cited: does the answer include you?

The AI system names you in the answer the user sees. A system can reach your page and still cite a competitor instead, so being retrievable does not get you cited on its own.

How SEO shapes discovery

All three states build on SEO. Being indexed, retrievable, or cited rests on the fundamentals search engines have always rewarded: a page a crawler can reach, structured and complete information, and content worth surfacing. The structured, machine-readable side of that is covered on the Readiness guide.

Generative Engine Optimization (GEO) has become part of the discovery conversation in its own right, a practice aimed specifically at being surfaced inside AI answers.

The term comes from a 2023 research paper by Pranjal Aggarwal and colleagues, later published at KDD 2024. They defined a generative engine as a system that gathers information from several sources and summarizes it with an LLM to answer a query, and GEO as the work of improving a page's visibility inside those generated answers.

The term is contested. Google's own guidance treats optimizing for its AI features as continuous with regular SEO and calls "GEO" rebranding, and across the agentic commerce, SEO, and ecommerce communities there is real disagreement over whether AI visibility needs any work beyond strong traditional SEO.

What is not in dispute is that visibility is split across many systems with different crawlers, indexes, and refresh timing, so one ranking number no longer covers it. Being on Google does not mean every assistant can retrieve and cite you, which is why discoverability gets checked across several AI systems rather than one.

The stakes here are highest for the long tail. The most differentiated products an agent goes looking for often sit with smaller merchants, the ones with no household-name recognition to carry them into an answer on reputation alone. For a well-known brand, discovery is partly handled by fame. For everyone else, being retrievable and legible is close to the whole game.

Content and context

Agent queries are richer than search queries ever were. A person types "best running shoes." An agent is handed something like "lightweight running shoes for long races under $200." Matching a query like that turns on two different things a product's data can express.

Content is the makeup of the product: what it is, what it is made of, what it weighs, what it costs. In that query, "lightweight" and "under $200" are content. Context is what the product is for, the situation it suits. "For long races" is context. A rich query usually carries both, and a product matches when its data answers both.

{
  "@type": "Product",
  "name": "Mount to Coast C135",
  "offers": { "price": "190.00", "priceCurrency": "USD" },
  "additionalProperty": [
    { "name": "Cushioning", "value": "Ultralight, 42mm dual-layer" },
    { "name": "Recommended use", "value": "Long-distance racing" }
  ]
}

Content maps cleanly onto the product attributes most catalogs already carry. Context is the part usually left implicit, the use case a person infers from a photo or a description but an agent cannot assume. Writing both into structured, schema.org-style product data is the most discussed way to make a product legible to a query like this. The markup itself is on the Readiness guide; the point for discovery is that a product gets matched on what it is and what it is for.

Richer data also widens the range of queries a product can turn up in. A complex request rarely stays one query: in Google's AI Mode, a technique called query fan-out breaks a single prompt into many smaller searches, runs them in parallel, and recombines the results into one answer. "Lightweight running shoes for long races under $200" becomes separate searches for lightweight racers, long-distance shoes, options under $200, and more. Every attribute and use case a product exposes is another of those sub-queries it can match, which is why more complete product data covers more of the surface.

Test your visibility

There is no single command for discovery. Each AI system keeps its own index and its own view of the web, so the test is to ask the assistants directly, one at a time.

Are you surfaced? In ChatGPT, Perplexity, Claude, and Gemini, ask the questions a shopper in your category would ask, and see whether you appear and are cited. For a running-shoe store, that might be:

• what are the best trail running shoes under $200?
• lightweight road running shoes for a marathon
• durable running shoes for long distances on different terrain

Being surfaced in one is not being surfaced in all... their indexes and refresh schedules differ, so a brand can show up in one assistant and be absent from another on the same day.

Are you retrievable? Paste one of your product URLs into each assistant and ask it what the page says. If it cannot read the page, or gets the details wrong, you are indexed but not retrievable by that system... a live page a given assistant cannot fetch at that moment. That is the gap between being on Google and being usable inside an AI answer.

Are you indexed at all? Google Search Console and Bing Webmaster Tools tell you whether the traditional index has you, which is the floor the AI systems often build from. Indexed is the baseline; retrievable and cited are separate results stacked on top of it.

What's durable, what's in flux

The surface changes constantly: which assistants lead, how often each one crawls, what any given system indexes, and how much discovery runs through protocols like ACP and UCP. What does not change is the fragmentation. No single system speaks for the whole web, so visibility is never one number, and a product that is present and checkable across many of them is insulated from any one of them changing.

Further reading

Where I have gone deeper on these on The Desk:

• Good SEO Is a Head Start With AI Shopping Agents ... structured data as the durable lever for being found and chosen by agents.