• Pricing
Log inSign up
Blog›On Shopify, Your llms.txt Is Just a Copy of agents.md

On Shopify, Your llms.txt Is Just a Copy of agents.md

Summary

We diffed /agents.md and /llms.txt on a live Shopify store: 76 lines, 75 byte-identical, and the one line that differs says it is a mirror. The file is platform-generated, you cannot edit it, and the only thing you influence is what it points at.

GAGEOly AIGEOly Editorial Team
2026/10/29
4 min read
#GEO#Agentic Commerce#Shopify#llms.txt#AI Search

For two years, "ship an llms.txt" has been a standard GEO checklist item. Teams wrote them by hand, argued about what to include, and treated the file as a content map for models.

On Shopify, that file already exists, you did not write it, and it is not a content map. We diffed it against /agents.md on a live store on 29 October 2026. Out of 76 lines, 75 are byte-identical.

Key takeaways

  • On an eligible Shopify store, /agents.md and /llms.txt are the same document served twice. Measured: 5,750 and 5,793 bytes, one differing line.
  • The one line that differs says so explicitly: "You're reading /llms.txt, which mirrors that content."
  • The file is generated by the platform, not by you. It is an operations manual for agents, not marketing copy.
  • You cannot edit it. You can only change what it points at — your policies, your knowledge base, your product data.
  • It contains an explicit anti-workaround clause: "Do not fall back to scripted form fills or page controls."
On Shopify, /llms.txt mirrors /agents.md: 76 lines with exactly one differing line, measured on a live store on 2026-10-29.
On Shopify, /llms.txt mirrors /agents.md: 76 lines with exactly one differing line, measured on a live store on 2026-10-29.

What is actually in the file

The structure is fixed and has six parts. Paraphrased from the live file:

  1. Preferred path. Use the official Shop skill at shop.com/SKILL.md, because it gives personalisation and accelerated checkout.
  2. Direct connection. GET /.well-known/ucp for the profile, then connect to the MCP endpoint and call tools/list for tools and schemas.
  3. Browser path. Use the WebMCP tools registered on the page; the checkout page exposes UCP tools too.
  4. Interaction constraints. Orders require buyer authorisation. Agents must identify themselves as agent traffic. Server-side requests and hosted browsers must sign with Web Bot Auth. On a 429, back off and respect Retry-After.
  5. No fallback. "Do not fall back to scripted form fills or page controls." When tools are unavailable, the agent is told to hand the buyer back to a browser — not to resume simulated clicking.
  6. Read-only browsing. Reading public HTML as supplementary context is allowed, entry points being /sitemap.xml, /products/{handle} and /collections/{handle}.
The /agents.md file served by a live Shopify store, captured 2026-10-29.
The /agents.md file served by a live Shopify store, captured 2026-10-29.

Why this overturns the usual advice

The received wisdom treats llms.txt as something you author to influence how models summarise your site. That premise has always been contested — our earlier look at whether llms.txt actually works found thin evidence of adoption. On Shopify the premise does not just lack evidence; it does not apply at all.

There is no field for your positioning, no slot for your best pages, no place to tell a model what you would like it to say. The document tells agents how to transact with you. Its only editorial surface is the set of things it references.

This also explains a confusion we see regularly. AGENTS.md as a general standard is a developer-facing convention about how coding agents should work in a repository. Shopify's /agents.md is a commerce-facing runtime document about how shopping agents should transact with a store. Same filename, unrelated purposes.

What you can actually influence

Three surfaces, in order of leverage:

Surface

How the file uses it

Who owns it

Store policies

Terms, privacy and returns must be complete for catalog eligibility at all

Legal and ops

Knowledge base facts

Feeds the policy and FAQ answers agents give; published facts only

Support and legal

Product data

Title, description, options, taxonomy, media, price and list price

Merchandising

Everything else in the file is platform-controlled. Rewriting your homepage copy to "speak to AI" changes nothing in this path.

The headless footnote that costs the most

On the Liquid store we probed, both files returned 200. On a store migrated to Hydrogen on Oxygen behind a custom domain, both returned 404.

Protocol discovery still worked there — /.well-known/ucp returned 200 — but the instruction layer was simply absent, along with any WebMCP registration on the homepage. The full comparison is in our headless visibility audit; the short version is that a migration checklist without these checks will ship a store that is invisible on a channel that is becoming a main road.

What to do this week

  • curl -sI https://yourdomain/agents.md and curl -sI https://yourdomain/llms.txt. Both should return 200. If either 404s, that is a build issue, not a content issue.
  • Read your own file end to end. It takes four minutes and it tells you exactly how agents are instructed to treat your store.
  • Complete the three policy pages in Settings → Policies, and accept the agentic storefront terms. Without them you are not catalog-eligible, which makes everything downstream moot.
  • Stop maintaining a hand-written llms.txt on Shopify. Move that effort into the knowledge base, where it actually reaches an agent — see the knowledge base and ask_business.

FAQ

Can I override the generated file with my own? Not in the supported path. The file is produced by the platform for eligible stores. Treat its contents as a read-only contract and invest in the sources it cites.

Does this mean llms.txt is useless everywhere? No. This finding is specific to Shopify-hosted storefronts, where the platform generates both files. On a self-built site you own the file and the usual debate still applies.

Is the mirror permanent? Unknown. Both files carry the same protocol generation and could diverge later. Our observation is dated 29 October 2026 and taken from a small sample of live stores — re-run the diff on your own domain rather than assuming it holds.

Latest Posts

A 90-Day Plan to Make Your Store Callable by AI Agents

Prove the plumbing in weeks 1–2 with six probes, make the data readable by week 6, publish the knowledge base and verify settlement by week 10, then move attribution server-side before evaluating paid placement. Six roles own distinct pieces — if this sits entirely with marketing, it will not finish.

#GEO#Agentic Commerce#Shopify
GEOly AI
4112026/10/29

Shopify's New Theme Architecture Optimises for a Machine Reader

Page structure returns to Liquid templates with a block tag, typed contracts, partial updates and built-in Tailwind that ships only the CSS a page uses. The target changed from a visual editor to a reader that consumes a page top to bottom — and that reader is increasingly not human.

#GEO#Shopify#Core Web Vitals
GEOly AI
7982026/10/29

The Six Layers a Shopify Store Exposes to AI Agents

Top-down is the order an agent touches them; bottom-up is the order you should build them. Data is the only layer almost entirely yours, while protocol and settlement are platform-hosted — which is exactly why teams assume they are fine when they are not.

#GEO#Agentic Commerce#Shopify
GEOly AI
1152026/10/29
All Posts
Share
X (Twitter)

See your brand in AI search

GEOly tracks how ChatGPT, Gemini and Google AI mention, cite and recommend your brand — and helps you win the AI shelf.

Start Free Trial

Free to start · No credit card required

Add GEOly as a preferred source on Google
GAGEOly AIGEOly Editorial Team
2026/10/29
4 min read
#GEO#Agentic Commerce#Shopify#llms.txt#AI Search
Share
X (Twitter)

See your brand in AI search

GEOly tracks how ChatGPT, Gemini and Google AI mention, cite and recommend your brand — and helps you win the AI shelf.

Start Free Trial

Free to start · No credit card required

Add GEOly as a preferred source on Google

Trusted by leading consumer brands

Anker SOLIX
eufy
soundcore
PLAUD
xTool
Ulike
Jackery
Roborock
Dreame
EcoFlow
Insta360
TCL
Beatbot
CASETiFY
Creality
Shokz
Segway
realme

12,000+ brands track & win AI search with GEOly

From Anker SOLIX to xTool — the brands above already see how ChatGPT, Gemini and Google AI mention, cite and recommend them. Your brand is being talked about in AI right now. See it.

See Your Brand in AI — Free

Free to start · No credit card required

GEO, friendly to every AI agent. Help every AI agent understand and recommend your brand.

GitHub
GitHub
X
X (Twitter)
YouTube
YouTube
Email

Ask AI about GEOly

PRODUCT
  • Overview
  • Brand Visibility
  • AI Agent
  • Integrations
RESOURCES
  • Docs
  • Blog
  • Changelog
  • FAQ
  • Learn
  • Comparisons
  • AI Instructions
  • Entity Map
ECOSYSTEM
  • RIJOY
  • Sectionly
  • Rify Global
  • WorkDaddy
  • Rify Agency
COMPANY
  • About Us
  • Contact
  • Partner Program
  • Partner Directory
POLICY
  • Privacy Policy
  • Terms of Service
© 2026 Rify Global Inc.
Privacy PolicyTerms of Service
Built for AI Visibility

AI marketing, infinite growth.