Shopify Quietly Rolled Out Agentic Commerce and llms.txt on Every Store

Tuesday morning a client pinged me about a new tab in their Shopify admin. Agentic Storefront. Sessions, orders, the usual columns — except attributed to ChatGPT and Microsoft Copilot instead of Meta or Google.

The numbers were small. They're small on every store I've checked this week. That part is not the interesting part.

The interesting part is that when I typed theirstore.com/llms.txt into a browser, the page loaded. Nobody on the team made that file. Shopify is serving it. So is every other Shopify store on the planet right now, including yours. Open a tab and check.

If you've been reading this blog, none of this should land as a surprise. I wrote about getting products listed in ChatGPT shopping results back in October. About generating AI-driven product UGC with Sora and Veo the summer before that. About UCP and what it does to your Meta Ads funnel in April. What's new this week is that the wiring underneath all of it shipped to every Shopify store by default — quietly, with no announcement.

What it actually is

/llms.txt is a plain-text manifest pointing AI agents at your products, your search endpoint, agent instructions, and the MCP and UCP commerce hooks that let an agent buy something from your store without ever loading your homepage (per Shopify's agentic commerce docs). There's also /agents.md (markdown version of similar guidance) and /sitemap_agentic_discovery.xml (the index tying it together). All three are generated by Shopify by default. All three are editable — drop a templates/llms.txt.liquid file into your theme and you can override the default. Same pattern as robots.txt.liquid.

UCP was the protocol announcement. The dashboard and the files are the rollout — every store, every region, all at once, no announcement.

That last bit is the tell. You don't ship to every store on the platform if you're still deciding. This is infrastructure now.

"But isn't this just another product feed?"

Fair question, and worth heading off.

If you've run Shopify for any length of time, you already have data going out to Google Merchant Center for Shopping and Performance Max. You have a Meta catalog for dynamic product ads. You have product metadata in your sitemap and on every PDP. So what's actually different about /llms.txt?

Three things, really.

• GMC and Meta catalogs are scheduled syncs. You push your data up on a cron, they index it, and your view of the world is as fresh as the last successful sync. Anyone who's debugged a Merchant Center disapproval at 2 AM knows how brittle that is. /llms.txt sits on your store and gets read on demand when an agent has a query. There is no sync window to go stale during.

• Second, GMC and Meta catalogs are discovery surfaces. They get your products into Shopping placements, DPA retargeting, the Shop tab. The conversion still happens on your site — they're the top of a funnel that ends with a checkout you control. /llms.txt is different because it also exposes the MCP and UCP endpoints. The same file that helps an agent find you also helps the agent buy from you. The site visit doesn't happen. The agent transacts on your behalf.

• Third, you push GMC, you push Meta, and over time you'll push every other AI shopping surface that launches — unless someone standardizes the layer. /llms.txt is that standardization. One file. Every agent. Same protocol.

The feeds you have don't go away. This isn't a replacement. It's a layer above the feeds that increasingly handles the part of the journey your storefront used to handle.

The thing nobody's talking about

One of the brands I work with runs Shopify Markets — separate domains, two languages, two currencies. Standard setup for a lot of brands now.

The default /llms.txt doesn't handle this well.

The file sits at the root of each domain, but it doesn't translate links, currency, or language metadata across markets. An agent landing on the wrong domain gets a single canonical view of the catalog in whatever the store's default language is, with whatever the root domain's pricing happens to be. If you sell in two currencies on the same Markets setup, the structured data the agent is comparing against can drift from what the buyer would actually see at checkout.

It will get fixed. Right now, if you run multi-region or multi-language, your /llms.txt is doing something you didn't intend and you probably don't know it. Go look. Then decide whether to override it with a Liquid template until Shopify catches up.

This is the kind of detail the LinkedIn posts about agentic commerce skip past, because the people writing them are running one-domain, one-language stores.

The scanner

Shopify also shipped a public agentic-readiness scanner at commerce-readiness.shopify.io. No login, no install — paste any public store URL and you get a score back. It runs 31 checks across five categories: Agent Discovery, Product Intelligence, Transaction Readiness, Store Quality, and Operational Readiness.

What it looks for is unromantic. Do your well-known endpoints (/llms.txt, /agents.md, /.well-known/ucp) return 200. Does /products/{handle}.json resolve for every product. Are your sitemaps reachable, including locale variants. Do you have real pages at /pages/faq, /pages/shipping, /pages/about, /pages/contact — not redirects, not Notion embeds, not modal overlays. Real pages with real text on them.

If you've been doing product-data hygiene for a while, most of it passes. If you haven't, this is the first time anyone is putting a score on it.

What I'd actually do

Run the scanner against your store. Open yourstore.com/llms.txt and read it. If something looks wrong, fix the underlying data first and override the file second.

That's the whole list. Don't buy AI visibility software in 2026. The platform is doing the work for free, and the part the platform isn't doing is product-data work no third-party tool can do for you anyway. Same point I've made about AI tooling for Shopify stores before.

One more thing

The dashboard numbers will stay small for a while. Yours will. Mine will. That's fine.

The reason to do the work now is that the work compounds. A clean feed in May 2026 is a clean feed in May 2027 when the numbers stop being small. A messy feed now is the same messy feed then, plus eighteen months of opportunity cost.

This was always the work. The platform is just finally watching whether you've done it.