Two ways Shopify feeds ChatGPT. Most merchants only optimize one.

On March 24, 2026, Shopify turned on Agentic Storefronts platform-wide. Products started syndicating into ChatGPT, Gemini, and Copilot. There was no opt-in, no setup wizard. Most merchants didn’t notice.

On Shopify community forums, merchants keep asking the same question: why are competitors showing up in ChatGPT with buy buttons when their own products aren’t? They’ve usually spent months on “AI search” hygiene: schema markup, an llms.txt file, a blog full of keyword-rich content.

Their SEO was fine. The problem was that ChatGPT Shopping doesn’t use any of those signals. It pulls from a separate data pipeline.

Two paths, different pipelines

There are two ways your products end up in a ChatGPT conversation. They use different data sources, different infrastructure, different ranking signals. Most merchants conflate them.

Path A: Agentic Commerce. ChatGPT queries your Shopify catalog directly. Products show up as interactive cards with images, prices, variant selectors, and a “Buy” button. The customer checks out without leaving the conversation. Shopify calls this Agentic Storefronts; OpenAI calls it the Agentic Commerce Protocol (ACP).

Path B: Web Discovery. AI crawlers (OAI-SearchBot, GPTBot, Googlebot) visit your storefront the way any search engine would. They read your HTML, parse your JSON-LD, index your pages. When someone asks a question, ChatGPT generates a text answer and links to your site. This is the evolution of SEO into what people now call Answer Engine Optimization (AEO).

Both paths can show up in the same conversation. Someone asks “what are the best trail running shoes under $150?” and ChatGPT Shopping pulls product cards from Path A while ChatGPT Search adds context from Path B, maybe your blog post about choosing trail shoes. The paths complement each other, but they run on separate pipelines.

Path A: how your catalog becomes a shopping card

Three systems sit between your Shopify Admin and the card in ChatGPT Shopping:

Your Shopify Admin (product data)
    ↓
Agentic Storefronts (enabled by default since March 2026)
    ↓
Shopify Catalog (ML enrichment via vision-language models)
    ↓
Catalog MCP Server (real-time queries from AI platforms)
    ↓
ChatGPT Shopping (product cards + in-conversation checkout)

Shopify doesn’t pass your data through unchanged. Their Catalog layer runs vision-language models that infer attributes you never set (brand, color, material, size) from your images and descriptions. They run these predictions at platform scale. Your product images aren’t only for customers anymore. They’re inputs for the ML models that decide how your products get categorized and surfaced.

The speed difference from web crawling is striking. Change a price, update inventory, add a variant, and it shows up in ChatGPT Shopping within hours. With traditional crawling, you might wait days for Googlebot to revisit a page.

Every field in the ACP spec becomes input for the AI’s reasoning. That is a different job from a search ranking, where a missing field only means the row has less information. In a conversational system, a missing field means the AI has less material to explain why your product fits a specific customer question. A product with a complete structured attribute set is not merely better ranked. It is more describable.

There’s no ad bidding either. Google Shopping is pay-to-play at scale. ChatGPT Shopping, as of April 2026, ranks products on data quality and semantic relevance alone. OpenAI started testing ads in January 2026.

What semantic matching cares about

ChatGPT users don’t type “blue running shoes size 10.” They say “I need waterproof shoes for trail running in the Pacific Northwest.” Your product descriptions need to address use cases and constraints, not list attributes. Scenario-specific language outperforms keyword repetition.

Structured attributes do more work than free-text. A properly tagged material: Gore-Tex in your Shopify product attributes is more useful to the AI than mentioning “Gore-Tex” three times in a paragraph. Structured data is machine-parseable. Prose is not, or at least not as reliably.

Policy fields matter too, and most merchants ignore them. Shipping speed, return policy, warranty information: when a customer asks ChatGPT “can I return these if they don’t fit?”, the AI needs that data to answer. A product without those fields leaves the AI guessing, and the AI has other products that don’t make it guess.

Path B: your website as an AI source

Path B is what most merchants picture when they hear “AI SEO.” The territory is familiar, but the rules have shifted.

AI crawlers visit your storefront, read your pages, build an index. When users ask questions, the AI generates answers from that index and cites your store as a source. The SEO fundamentals still apply: crawlability, page speed, structured data, internal linking. User behavior has shifted underneath them. SparkToro’s 2024 study put US zero-click searches at 58.5%, and AI Overviews often sit above the traditional results on commercial queries. AI platforms prefer to answer inside the answer, not to send the user elsewhere. Your content has to be good enough to be cited, and the citation is the event that matters, not the click-through.

Content format has started to matter more than volume. FAQ sections and listicles let a language model lift a clean quote without having to paraphrase free-form prose.

Structured data is table stakes. Product schema with GTIN, price, and availability lets an LLM read the identifiers directly. A page without JSON-LD hands the LLM free text and asks it to guess, and LLMs with options rarely pick the page that asks for more work.

The highest-leverage move on Path B is blog content. Product pages describe one product. Blog content can address scenarios, comparisons, and “best X for Y” queries that drive AI citations. If you sell running shoes, a post about what shoe features matter for wet-trail running gives AI platforms something to cite when users ask that question.

A Shopify store without a blog has no surface for those long-tail queries to land on, so the queries find their answers somewhere else.

On robots.txt, the practical question is whether OAI-SearchBot and GPTBot can reach product pages and blog posts. On llms.txt, the file tells language models which pages matter. As of April 2026 no major AI platform has confirmed using llms.txt for ranking, so it reads as a signaling move more than a discovery mechanism.

How the two paths differ in practice

Most of the effort still goes into Path B because the SEO industry spent 25 years teaching merchants to think about web crawlers. Agentic Commerce is about a year old. The habit gap is easier to explain than the outcome gap.

Path A has more immediate transactional impact. When your product shows up as a shopping card with a buy button inside ChatGPT, the conversion path is one click. When your website gets cited in a text answer, the user still has to click through, find the product, add to cart, and check out. One step versus four.

Path A covers the transaction. Path B covers the description of the seller that the AI carries into that transaction.

Brand authority crosses both paths. When ChatGPT weighs two similar products (your $89 trail shoe against a competitor’s $92 version), signals from Path B shape how the AI frames its recommendation. A brand that ChatGPT “knows” from web content gets described with more confidence than one it sees only in a catalog feed.

Long-tail discovery lives on Path B. Nobody types “gift for a trail runner who overpronates and lives in Seattle” into ChatGPT Shopping. But they might ask that in a conversational search. If your blog answers that question with specific product recommendations, you’ve built a discovery channel that no catalog feed can replicate.

Cross-pollination between the two paths is the direction the incentives point. As AI platforms mature, they will cross-reference catalog data with web content. A product with rich catalog attributes and authoritative web content about its use cases should outperform one with catalog data alone. The platforms have not confirmed this yet.

The merchants showing up in ChatGPT Shopping now are the ones treating catalog data quality and web content as two separate problems.