A merchant did what every "protect your content from AI" guide told her to do. She edited robots.txt, blocked GPTBot, ClaudeBot, PerplexityBot and Google-Extended, and considered the matter closed.

Two weeks later her products were showing up in ChatGPT shopping answers, with current prices and live availability. She wrote to us asking how the bots got past the block.

They did not get past anything. Nothing crawled her store. Shopify handed the data over directly, exactly as designed.

The sentence in the docs that explains everything

Shopify's own agentic storefront documentation is unusually clear about this. Blocking AI crawlers in robots.txt "affects only open-web discoverability". It does not stop Shopify Catalog from sending your product data to the agentic storefronts you have activated.

That one sentence describes two completely separate pipes carrying your data into AI systems. robots.txt controls exactly one of them. Most merchants think it controls both, and every decision they make about AI visibility inherits that mistake.

Pipe 1: open-web discoverability

The first pipe is the one you already understand, because it works the way search engines always have.

AI crawlers visit your public storefront the way Googlebot does. GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended request your pages, read the HTML, and use what they find. This is how an assistant can cite your buying guide, quote your about page, summarize your care instructions, or pull a review hosted on your site when someone asks an informational question.

robots.txt is the doorman for this pipe. It is the file where you tell each bot "you may crawl" or "you may not". Block a crawler there and it cannot read those pages, which means the content on them cannot be cited or quoted by that system.

On Shopify you control this through the robots.txt.liquid template, and the decision is more granular than "AI yes or no". OpenAI and Anthropic each run separate bots for model training, for search indexing, and for live fetches when a user shares your URL. Blocking the training bot while allowing the search bot is a real option, and most stores have never made that choice deliberately.

One more property of this pipe worth knowing: these crawlers read raw HTML and do not execute JavaScript, so what they can see is only what your server sends in the first response, the rendering gap that now tops the technical SEO list.

Pipe 2: Shopify Catalog syndication

The second pipe involves no crawler at all, which is why robots.txt has no say in it.

Shopify Catalog takes your structured product data, the title, price, images, GTIN, variants and availability, and pushes it directly to the AI channels you have activated through a first-party data connection. Shopify hands the data over. Nothing visits your site, nothing requests a page, and nothing checks your robots.txt on the way, because there is no "way".

This is the pipe behind Agentic Storefronts. It is how your products reach ChatGPT, Microsoft Copilot and the Shop app with live pricing, and it keeps flowing even if you have blocked every AI crawler that exists.

Google runs a parallel version of the same pipe: AI Mode shopping answers pull from Merchant Center through the Google & YouTube channel, not from your pages. And the traffic returns the same first-party way, with orders landing in your admin carrying channel attribution.

The design makes sense once you see the intent. Crawling is pull: a bot takes what it can read. Syndication is push: the platform sends clean, structured data on your behalf, which is faster, fresher, and far more reliable than letting an agent scrape a product page. Shopify built the second pipe precisely so AI shopping would not depend on the first one.

What flows through each pipe

The split is not just technical. Each pipe carries a different half of your store.

Pipe 1 carries your words: blog posts, guides, comparisons, policy pages, the about story, on-site reviews. This is the material AI systems read to understand and describe your brand, and to answer the research questions that come before a purchase.

Pipe 2 carries your facts: the product records themselves. This is the material AI shopping agents compare when someone asks for the best option under a budget.

Which means there are four possible configurations, and only two of them are coherent.

Allow both pipes, and AI systems can recommend your products and explain them with your own content. Close both, by blocking crawlers and deactivating your agentic channels, and you have opted out as fully as the platform allows.

The strange middle states are where merchants land by accident. Deactivating channels while allowing crawlers means your content gets cited but your products are absent from shopping answers. And blocking crawlers while Catalog flows is the accidental default for stores that "blocked AI" last year.

In that state your products appear in AI conversations while the AI cannot read a single page of your context. It describes you from Reddit threads and competitor comparisons instead, because third-party content is what fills the gap when your own is unreadable.

For most stores, that last configuration is the worst of both worlds, and it is the most common one we see.

How to actually control each pipe

For pipe 1, edit robots.txt.liquid with intent rather than fear. Decide per bot family: training crawlers, search crawlers, user-fetch agents. If you want citations and AI-referred visitors, the search and fetch bots need access, whatever you decide about training.

While you are in the root-file layer, remember robots.txt now has two siblings there. agents.md and llms.txt describe your store to AI systems rather than gatekeep them, two different jobs handled by two different files, and neither is the ranking lever it gets sold as.

For pipe 2, the controls live in your admin under Sales channels, not in any file. You manage which agentic channels are active, and you can hide specific products from AI surfaces. Note the asymmetry Shopify documents: the ChatGPT storefront is a discovery channel without an off switch, so product-level curation and Catalog eligibility are your real levers there.

Then check your own configuration this week. Open your robots.txt in a browser, note which bots are blocked, and compare that against the channels active in your Agentic settings. If the two tell different stories, you now know which pipe each one controls.

One last thought on where this leaves your effort. Pipe 2 carries your products, and Shopify largely runs it for you. Pipe 1 carries your story, the guides and posts that make an AI describe you accurately and send you buyers who convert. That pipe only carries what you publish, which is the half of the system we spend our days on at vevy.