llms.txt, agents.md or Neither: What Shopify Already Publishes for You and What Is Worth Editing
A shopper asks an assistant about your store, and the assistant goes looking for a page that explains how your store works. If you opened yourstore.com/agents.md this morning, you would find that page already there, and you did not write it.
Since spring 2026, Shopify has served discovery files on every storefront. That changes the question. It is no longer "should I create an llms.txt file?" It is "should I leave the one Shopify wrote alone, edit it, or stop thinking about it?" The short answer: for most established stores, leave it alone, check it once, and spend the effort on product data and policy pages.
What does Shopify publish on your store today?
Every Shopify store now serves /agents.md as its canonical agent discovery URL, plus /llms.txt and /llms-full.txt for older crawlers. By default all three return the same content. On the Allbirds store I read on October 24, 2026, that content is a set of agent instructions: the Shop skill, Universal Commerce Protocol (UCP) and WebMCP integration, a rule that orders need the buyer's authorization, sitemap guidance and links to the store policies. The files were shorter in May. Shopify's Help Center documents the files and says you need no app to generate them.
The rollout was quiet. Craftshift reported on May 11, 2026 that the files appeared on every Shopify store it tested in late April or early May, without a press release. The Shopify Help Center page on Shopify Catalog and product discovery for agentic storefronts now describes them as standard.
The content has changed since May. Craftshift's May walkthrough quoted a short llms.txt for Allbirds with a store description, collection links, currency and contact details. When I requested www.allbirds.com/llms.txt and /agents.md on October 24, 2026, both returned the same text, and it opened with instructions for agents: a pointer to the Shop skill, the Universal Commerce Protocol (UCP) endpoints, a rule that orders need the buyer's authorization, and a Store Policies list linking the privacy policy, terms of service, refund policy and shipping policy. That matches the Help Center's statement that the files return identical content by default.
Check yours in two minutes. Open both URLs in a browser, read what an assistant would read, and note whether anything in them is wrong for your store. If your store runs on a custom or headless front end, confirm the URLs resolve rather than assume they do.
What is documented, and what is speculation?
Documented: Shopify serves the files, and Search Engine Journal reports that Google's Search guidance does not require them. Google's Lighthouse tool checks for llms.txt in its agentic browsing audit. Speculation: that any major assistant ranks or cites your store more often because the file exists. The published evidence points the other way, and it comes mostly from third parties.
What does Google say?
Search Engine Journal reported on May 20, 2026 that Google Search's new optimization guide lists llms.txt among the tactics you do not need for generative AI features. The same report says Lighthouse 13.3 added an Agentic Browsing category with an llms.txt audit, so Google's answer depends on the product. Search Engine Journal also recounts John Mueller comparing llms.txt to the keywords meta tag and Gary Illyes confirming Google was not pursuing it. Read the full report; I relied on that summary rather than Google's own guide.
What do OpenAI, Anthropic and Perplexity say?
Limy's guide, dated May 12, 2026, states that no major provider, naming OpenAI, Anthropic, Google, Meta and Mistral, has publicly committed to using llms.txt as a signal in production search or answer surfaces. Rich Dev Tools adds that Perplexity has said it retrieves the file to prioritize pages and that Anthropic publishes one on its own site, and calls neither a documented retrieval guarantee. I found no primary statement from OpenAI, Anthropic or Perplexity in the sources I opened, so treat these as secondary reports.
What do the log studies show?
| Study | What it reports | Caveat |
|---|---|---|
| Limy, May 12, 2026 | Of more than 500 million AI bot visits over 90 days, 408 requested llms.txt directly | The sample is Limy's own monitored AI bot traffic |
| Rich Dev Tools, August 25, 2026, summarizing SE Ranking's May 2026 study | About 300,000 domains: 10.13% had an llms.txt, none of the top 1,000 by traffic did, and removing the file from a citation model improved its accuracy | Secondhand summary of the SE Ranking study; read the original before quoting it |
Neither study proves the file is harmful. Both say the same narrow thing: the bots measured mostly skip it.
When does doing nothing make sense, and when does editing earn its place?
Doing nothing makes sense when the default file is accurate and your product data is in good order. Editing earns its place when the default says something wrong or leaves out something an agent needs to complete a task for your customer. It does not earn its place as a ranking tactic.
Leave it alone if:
- Your store name, description, currency and contact details read correctly in the file.
- The policy links point to current shipping and returns pages.
- You have no special instruction for agents that your product pages and policies do not already carry.
Edit it if:
- The description at the top is generic or stale, and an assistant would repeat it.
- You sell in several markets and want agents pointed to the right shipping and returns page for each.
- One of your own pages answers a recurring buyer question, such as sizing or care, and the default does not link to it.
Shopify offers a clean way to edit. The Help Center says agents.md.liquid changes the content served from all three URLs, while llms.txt.liquid or llms-full.txt.liquid change one URL only, and a URL-specific template wins over agents.md.liquid. One caution from the Help Center: do not put a proxy in front of Shopify. Shopify says it does not recommend that and cannot support it.
How do structured data, product feeds, Shopify Catalog and plain pages compare?
Product facts reach assistants through four routes, and the discovery file is the weakest of them. Shopify's Help Center tells merchants to focus on complete product information in Shopify Catalog and on SEO best practices, and says the agent discovery files "don't replace any Shopify Catalog capabilities." That ordering is the guide for where to spend an afternoon.
| Route | What it carries | Effort | What the sources say |
|---|---|---|---|
| Shopify Catalog | Title, description, options, images, price, availability, updated continuously | Medium: complete fields, check Catalog Mapping for custom data | Shopify calls it the primary method for agentic storefronts and the authoritative product feed |
| Product feeds you own | The same facts for channels such as Google | Medium, ongoing | The Help Center says eligible products are discoverable through feeds you share, for example for advertising |
| Structured data and plain readable pages | What a crawler sees on the open web | Low to medium, mostly one-time | The Help Center says products found in a Google search can be found by Gemini users too |
| agents.md or llms.txt | A summary and pointers | Low | Documented as discovery only; no source I opened shows a citation effect |
Eligibility matters here. The Help Center says agentic storefronts such as ChatGPT and Microsoft Copilot can display your products only when your store is eligible for Shopify Catalog, and that ineligible stores may still appear through ordinary crawling and indexing. Check the Agentic section of your Shopify admin before you decide anything else.
Say a store of 120 products has a good llms.txt but a third of its product descriptions are one-line titles with no materials or dimensions. An assistant asked about those products has nothing to quote, and no discovery file fixes that. Fixing the descriptions is the larger job, and it reaches every route in the table at once.
How do you keep an edited file consistent with live prices, policies, shipping and returns?
An edited discovery file stays correct only if it never repeats a fact that lives somewhere else. Link to the page that owns the fact. Pull store values from Liquid rather than typing them. The default file already works this way: it links to your policies instead of copying them, so it cannot contradict them.
Here is the failure to avoid, using an illustrative store. You write "Free shipping over $75" into agents.md.liquid in March. In September you raise the threshold to $100 on the shipping page and in checkout. The file still says $75, and an assistant that reads both now has two answers. Nothing flags the conflict.
Three habits prevent it:
- Write the file as pointers: "Shipping rates and delivery times are on this page", with a link, rather than the rates themselves.
- Use Liquid values such as the shop name and URL instead of retyped text, so a rename flows through.
- Add the file to the checklist you already run when shipping, returns or pricing change, and open /agents.md and /llms.txt to confirm they still agree with the store.
The same rule applies to prices. Shopify Catalog sends current prices and inventory on its own, per the Help Center, so a price typed into a discovery file is a second copy that only you maintain.
FAQ
Does Shopify's default agents.md hurt my store?
No source I opened reports harm from the default file. It contains your store details and links, and Shopify generates it. Read it once and confirm the description, contact details and policy links are accurate.
Will editing llms.txt make ChatGPT or Perplexity recommend my store more?
No source I opened shows that. Limy found 408 llms.txt requests in more than 500 million AI bot visits, and no major provider has publicly committed to using the file as a signal. Product data in Shopify Catalog and clear policy pages are the documented levers.
What happens if I only edit llms.txt and leave agents.md?
Shopify's Help Center says a URL-specific template defines the content for its own URL. Editing llms.txt.liquid changes /llms.txt only, while /agents.md and /llms-full.txt keep the default. That can leave two files disagreeing, so edit agents.md.liquid unless you have a reason to split them.
Does opting out of Shopify Catalog remove my store from AI answers?
Not entirely. The Help Center says products may still appear through ordinary crawling and indexing, and that blocking crawlers in robots.txt does not stop Shopify Catalog from sending product data to agentic storefronts you have activated. Setting a product to Unlisted hides it from AI channels, and also from sitemaps, search engines and your store search.
