Shopify agents.md and llms.txt: What We Found Testing Real Stores

Originally published on the PingAura blog.

TL;DR

• Standard Shopify storefronts now serve /llms.txt and /agents.md automatically. Both return an Agent Instructions page that tells AI agents how to shop the store.
• The page points agents to the Shop skill (personalised discovery, accelerated checkout with Shop Pay) and to the Universal Commerce Protocol (UCP): a /.well-known/ucp profile, an MCP endpoint at /api/ucp/mcp, and WebMCP tools in the browser at checkout.
• Headless and custom storefronts do not get this automatically. On the headless stores we checked, both files returned 404.
• If you run a standard Shopify theme, an llms.txt generator app is now largely redundant. To change the file, Shopify supports an agents.md.liquid theme template, which also serves /llms.txt and /llms-full.txt.
• In our tests, Shopify returned 403 to GPTBot, ClaudeBot and PerplexityBot on app proxy paths (/apps/...), while the same bots could read the storefront itself. Content an app serves through the proxy is invisible to those crawlers.
• Being reachable by agents is necessary but not sufficient. Whether AI recommends you still depends on product data, content, reviews, trust pages, crawler access and third-party citations.
• Check your store with the free Shopify AI Agent Readiness Checker and Product Page AI Citability Checker.

Introduction

For the past year, merchants have been told to add an llms.txt file so that AI assistants can understand their store. A small industry of apps grew up to generate one.

Shopify has made that step automatic for standard storefronts. Well-known Shopify brands on the standard storefront now serve both /llms.txt and /agents.md: an Agent Instructions page written for AI agents rather than people.

Plenty has been written about the rollout. We wanted to know what it actually does, so on 2 November 2026 we tested it: live standard and headless stores, the UCP profile and MCP endpoint, and a Shopify development store of our own where we sent requests as each major AI crawler. This post shares what we found, who gets the files, and what still decides whether an AI assistant puts your product in front of a shopper.

1. What Shopify Now Serves on Standard Storefronts

On the standard Shopify storefronts we checked on 2 November 2026, /agents.md returns a Markdown page that Shopify calls the canonical agent-facing description of the store, and /llms.txt mirrors it, word for word apart from one line pointing back to /agents.md. It is headed Agent Instructions and opens by telling the agent that it can browse and shop the store on its user's behalf.

The page has four main parts:

• Shop skill (recommended). Agents are pointed to Shopify's Shop skill. Once the user connects their Shop account, the agent gets personalised discovery (search by text or image, similar products, purchase history) and accelerated cart and checkout using Shop Pay with saved addresses and payment methods. It can also look up orders, track deliveries and reorder.
• UCP integration. The store states that it implements the Universal Commerce Protocol for discovery, carts and checkout. A GET to /.well-known/ucp returns the store's UCP profile, and that profile advertises an MCP endpoint at /api/ucp/mcp where an agent can list and call shopping tools.
• Browser integration (WebMCP). For agents working in a browser, the storefront and checkout pages register WebMCP tools. The checkout page exposes UCP tools that act on the current checkout session.
• Interaction rules. Orders need the buyer's authorisation. Agents must identify themselves (Shopify asks server-side agents to use Web Bot Auth), use the provided tools rather than scripting checkout forms, and respect rate limits.

The page also lists read-only routes (/sitemap.xml, /products/{handle}, /collections/all) and links to the store's policies. If you want the background on how protocols like this fit together, our guide to what agentic commerce is covers the wider picture.

2. What We Tested, and What We Found

At PingAura, we wanted to see what this does in practice. On 2 November 2026 we requested the agent files on live stores, called the UCP profile and MCP endpoint, and set up a Shopify development store with a test app so we could see exactly which requests reach an app.

The agent files

Check	Standard storefronts (3 large brands)	Headless storefronts
/agents.md	200, Agent Instructions in Markdown, named as the canonical file	404
/llms.txt	200, a mirror of /agents.md	404
/.well-known/ucp	200, JSON profile naming the MCP endpoint	Not published
/api/ucp/mcp	Answers an MCP tools/list call with shopping tools such as get_checkout	Not available

Two details stood out. The UCP profile answered even on our password-protected development store, so agents can find the endpoint before a store opens. And robots.txt on standard storefronts now points agents to /agents.md and /.well-known/ucp.

Which AI crawlers reach Shopify app content

Many Shopify apps publish pages through the app proxy, which serves an app's content on the store's own domain under a path such as /apps/your-app/.... We sent the same requests to the storefront and to an app proxy path, changing only the user agent:

Crawler	Storefront (/)	App proxy path (/apps/...)
GPTBot (OpenAI training)	200	403, blocked by Shopify before reaching the app
ClaudeBot (Anthropic)	200	403, blocked by Shopify before reaching the app
PerplexityBot	200	403, blocked by Shopify before reaching the app
OAI-SearchBot (ChatGPT search)	200	200, reached the app
Chrome (a person)	200	200, reached the app

So if an app serves FAQ pages, guides or comparison pages through the app proxy, most AI crawlers never see them, even though shoppers do. Content you want AI assistants to read should live in the store itself: theme pages, blog posts, or product and collection pages. We tested a development store, so check your own store before relying on this: request an app proxy URL with each crawler's user agent and look at the status code.

We also found that no app, ours included, can see which crawlers visit the rest of the storefront. Shopify does not expose request logs to apps, and analytics pixels only run in real browsers. The one route to full crawler data is a store that sits behind the merchant's own Cloudflare zone.

3. Who Gets It, and Who Does Not

This behaviour belongs to Shopify's hosted storefront. If Shopify renders your pages, you appear to get it without doing anything.

Headless and custom storefronts are different. If your front end is your own Next.js, Hydrogen or other custom build, requests for /llms.txt go to your application, not to Shopify. On the headless Shopify brands we checked, /llms.txt and /agents.md returned 404.

If you run headless, you need to decide for yourself what to publish at those paths, and whether to expose UCP. Our UCP Readiness Report checks what an agent can find on your domain today.

Separately, Shopify has announced Agentic Storefronts, which it says syndicate merchant catalogues to AI assistants including ChatGPT, Microsoft Copilot and Gemini. That works at the catalogue level rather than through files on your domain, so headless merchants may still be represented there even without an llms.txt.

4. Is Your llms.txt App Now Redundant?

For most merchants on a standard Shopify theme, yes. Shopify now serves an llms.txt that is more useful to a shopping agent than a generated list of links, because it tells the agent how to act, not only where to read.

Two caveats:

• Check what is actually live. Fetch https://yourstore.com/llms.txt and https://yourstore.com/agents.md and see what comes back. If an app or a redirect is changing those paths, you will see it.
• An llms.txt was never the thing that made AI recommend you. It is a map. It helps an agent that has already decided to visit you. It does not persuade an assistant to choose you over a competitor.

If you do want to change what agents read, you no longer need an app. Shopify's theme documentation describes an agents.md.liquid template that controls /agents.md and is also used for /llms.txt and /llms-full.txt unless those have their own llms.txt.liquid or llms-full.txt.liquid template. Shopify recommends adding it only when the managed file does not cover your needs.

The second caveat is where most of the work still lies.

5. Discoverable Is Not the Same as Recommended

When a shopper asks ChatGPT or Gemini "what are the best merino running shoes under £120", the assistant builds an answer from product data it can trust and sources it already relies on. Agent Instructions help it transact. They do not decide which products appear in the answer.

What still decides that:

• Complete Product and Offer schema. Price, currency, availability, brand, and a GTIN (or MPN) on every product page. Missing identifiers make it harder for an engine to match your product to the one it is reasoning about.
• Clear specs and comparisons. Materials, dimensions, sizing, compatibility and care, written as text rather than buried in images. Answer engines quote what they can read.
• Reviews as text. Star ratings alone say little. Review text that mentions real use cases gives an assistant something to cite.
• Trust pages. Shipping, returns, warranty, contact details and an about page. Assistants weigh whether a store looks safe to send a buyer to.
• Crawler access. Make sure GPTBot, ClaudeBot, PerplexityBot and similar crawlers are not blocked in robots.txt or by a firewall rule.
• Citations elsewhere. Reviews, comparison articles, forums and press that AI answers already draw on. If those sources never mention you, a perfect product page still struggles.

The free Product Page AI Citability Checker checks how easily AI assistants can cite a single product page, and the Shopify AI Agent Readiness Checker checks how ready your whole store is for AI shopping agents.

6. How to Measure Whether It Is Working

Files on your domain are an input. The outputs are whether AI assistants mention you and whether those mentions turn into sales.

• AI visibility across engines. Track the questions your shoppers actually ask across ChatGPT, Gemini, Perplexity, Microsoft Copilot, Google AI Overviews and others, and see how often your brand and products appear compared with competitors.
• AI referral traffic and revenue. Watch visits from AI assistants in your analytics, and the orders they lead to. Many AI apps do not pass a referrer, so some of this traffic shows up as direct; read our guide to what AI traffic is and how to track it for the gaps to expect.
• Product data health over time. Schema and content regress when themes, apps and catalogues change. Audit regularly rather than once.

FAQs

Does every Shopify store now have an llms.txt?

Every standard Shopify storefront we checked in November 2026 served /llms.txt and /agents.md. Headless and custom storefronts did not, because those paths are handled by the merchant's own front end.

What is the difference between llms.txt and agents.md on Shopify?

Almost none. /agents.md is the canonical Agent Instructions page, and /llms.txt mirrors it, changing only the line that says which file is canonical. Shopify's theme templates also treat agents.md.liquid as the source for both.

What is UCP?

The Universal Commerce Protocol is an open protocol for agents to discover products, manage carts and complete checkout. A Shopify store publishes its UCP profile at /.well-known/ucp, which points agents to an MCP endpoint for shopping tools.

Should I uninstall my llms.txt app?

If you are on a standard Shopify theme, it is probably no longer needed. Check what your store serves at /llms.txt first, and remove the app only once you are happy with Shopify's version.

Can I customise Shopify's agents.md or llms.txt?

Yes. Add an agents.md.liquid template to your theme to control /agents.md, which Shopify also uses for /llms.txt and /llms-full.txt. Separate llms.txt.liquid and llms-full.txt.liquid templates override just those paths.

Can AI crawlers read pages that a Shopify app serves?

Not reliably. In our tests on a development store, Shopify returned 403 to GPTBot, ClaudeBot and PerplexityBot on app proxy paths (/apps/...), while OAI-SearchBot and normal browsers got through. Put content you want AI assistants to read in your theme pages, blog or product pages.

Will AI assistants now recommend my products?

Not automatically. The Agent Instructions help an agent shop your store. Whether an assistant recommends you depends on product schema, readable specs, review text, trust pages, crawler access and citations on third-party sources.

How can I check my store?

Run the free Shopify AI Agent Readiness Checker for the store, and the Product Page AI Citability Checker for individual product pages.

Conclusion

Shopify has taken a step that most merchants were doing by hand: telling AI agents how to use the store. On standard storefronts, llms.txt, agents.md, a UCP profile and an MCP endpoint now come built in. That removes a chore and lowers the bar for agentic checkout.

It does not change what wins the recommendation. Complete product data, readable content, real reviews, trust signals, open crawler access and third-party citations still decide whether an AI assistant puts your product in its answer. Measure that directly, across engines, and tie it back to revenue.

Ready to see how AI assistants recommend your store?

PingAura tracks how ChatGPT, Gemini, Perplexity and other AI assistants talk about your brand and products, audits your site for what they need, and shows the traffic they send you. Start with the free tools, then sign up at PingAura.ai to track your store across every engine.