AEO & GEO
llms.txt for Ecommerce: Who Reads It, What to Include, and How to Add It on Shopify
What llms.txt is, which AI companies have said they use it (and who has said they do not), a sample ecommerce file, and how to add it on Shopify.
Gaurav Rawat
llms.txt is a Markdown file at your site root that lists your most useful pages for AI systems. Jeremy Howard proposed it in September 2024. For a store, it works as a short, curated index of collections, policies and key products. No major AI search vendor documents reading it today, so treat it as a cheap, small addition.
What is llms.txt?
llms.txt is a plain Markdown file served at /llms.txt. The proposal comes from Jeremy Howard and is published at llmstxt.org, dated September 3, 2024. The idea is simple: give a language model a clean, human-written table of contents instead of making it parse navigation, scripts and banners.
The format is strict about very little. The only required element is an H1 with the site name. After that you can add a blockquote summary, optional plain Markdown notes, and H2 sections that hold lists of links, each with an optional short description. An "Optional" section marks links a model can skip when it needs less context.
It is not a blocking or permissions file. That is the job of robots.txt. llms.txt only points to content, and it cannot stop anyone from reading anything.
Which AI companies read llms.txt today?
Short answer: no major AI search company has published a statement saying its answer engine reads llms.txt to decide what to cite or recommend. Several have said or shown the opposite. Here is what the primary sources say as of November 3, 2026.
Source | What it says | Where |
|---|---|---|
Google Search | Says you do not need new machine-readable files, AI text files, markup or Markdown to appear in Google Search, and that files you keep for other services will not affect Google Search visibility | |
Google (John Mueller) | Called llms.txt "purely speculative for now" in a Reddit comment, as reported on June 2, 2026 | |
Chrome Lighthouse | Includes an llms.txt audit in its agentic browsing category. Docs call it an "emerging convention" and providing the file "optional at the moment"; a 404 does not fail the audit | |
Shopify | Serves /llms.txt on every store by default and lets merchants override it, from May 29, 2026 | |
OpenAI, Anthropic, Perplexity | Each publishes an llms.txt for its own developer docs. We found no published statement from any of the three that its consumer answer engine reads llms.txt on other sites |
Publishing a file on your own docs is not the same as reading other sites' files. That distinction is the one most llms.txt articles blur.
What does server-log data show?
Ahrefs checked 137,210 domains with traffic in May 2026 and published the results on June 15, 2026. About 28% (38,360) had a valid llms.txt. Of those, 97% received zero requests for the file during the month. AI retrieval bots were a small slice of the requests the remaining files did get.
Ahrefs named GPTBot (OpenAI's training crawler) as the largest AI fetcher, and Claude Code, Anthropic's coding tool, as the second-largest named AI tool. It noted that Googlebot fetched the file about 900 times but attributed that to normal crawling. Its own verdict: the file is "largely decoration" for AI search visibility, and the cons currently outweigh the pros. See the Ahrefs study.
Two cautions from the same study. Requests are a ceiling on use, since fetching a file does not mean a system read or acted on it. And the study found its clearest audience in coding assistants and similar tools, not shopping answers.
Does llms.txt matter for ecommerce?
Slightly, and mostly not yet. A shopper asking ChatGPT, Perplexity or Gemini for a recommendation is served by product pages, structured data, feeds, reviews and third-party sources. None of the sources above shows llms.txt in that chain.
Where it earns its keep is as a low-cost, tidy summary for any tool that does fetch it: a developer's coding assistant, a research script, an AI browser feature that checks for it. Ahrefs also noted that tools tend to fetch the file when pointed at it, so linking to it from your own documentation or developer pages can help.
Lever | What it does | Evidence it affects AI shopping answers | Effort |
|---|---|---|---|
Product structured data (JSON-LD) | Tells machines price, availability, variants, ratings per product | Strong and documented, see product schema mistakes | Medium |
Product feeds (for example Google Merchant Center) | Feeds product facts to shopping surfaces | Strong, one of the surfaces AI shops through, see product feed optimization | Medium |
Agentic protocols (ACP, UCP) | Lets agents check out natively | Live rails, see ACP vs UCP | High |
robots.txt | Allows or blocks crawlers | Documented control | Low |
llms.txt | Curated index of pages | No vendor statement that answer engines use it | Low |
Nudge's Catalog Analysis scores every product on how readable it is to AI, and the structured data fixes it ships sit higher in this table than any text file. If you want to see where your own gaps are before touching llms.txt, the free catalog scan reads your catalog the way an AI agent would and ranks the fixes by value. It needs only your store URL.
What should an ecommerce llms.txt contain?
Keep it short and factual. It should answer the questions a shopper's AI tool would otherwise have to dig for.
A one-paragraph brand summary in the blockquote: what you sell, to whom, where you ship.
Collections with a line on what each covers.
Top product pages, your 10 to 30 best sellers or hero products, not the whole catalog.
Policies: shipping, returns, warranty, and anything that changes a buying decision.
Sizing and fit guides, care guides, and buying guides.
Your sitemap URL, so a tool can find the long tail.
Nothing private. No contact emails, phone numbers, discount codes or internal links.
Use the same facts that appear on your pages. If the file says free returns and the returns page says store credit, you have created a conflict, not a summary.
What does a sample ecommerce llms.txt look like?
Here is a file for a fictional outdoor apparel brand. The structure follows the llmstxt.org format: H1, blockquote, then H2 link lists with an Optional section at the end.
# Harbor & Pine
> Harbor & Pine sells waxed-cotton jackets, merino base layers and rain gear for
> hiking and everyday wear. We ship to the US and UK. Prices are in USD and GBP.
Product pages are the source of truth for price, stock and variants. Policies below
apply to all orders.
## Collections
- [Men's jackets](https://harborandpine.example/collections/mens-jackets): waxed cotton, rain shells and insulated layers
- [Women's jackets](https://harborandpine.example/collections/womens-jackets): same range, women's fits
- [Base layers](https://harborandpine.example/collections/base-layers): merino tops and bottoms in three weights
## Best-selling products
- [Ridgeline Waxed Jacket](https://harborandpine.example/products/ridgeline-waxed-jacket): waxed cotton, six sizes
- [Tideline Rain Shell](https://harborandpine.example/products/tideline-rain-shell): packable waterproof shell
- [Merino 200 Crew](https://harborandpine.example/products/merino-200-crew): mid-weight base layer
## Policies
- [Shipping](https://harborandpine.example/policies/shipping-policy): delivery times and costs by region
- [Returns](https://harborandpine.example/policies/refund-policy): return window and conditions
- [Warranty](https://harborandpine.example/pages/warranty): what is covered and for how long
## Guides
- [Size and fit guide](https://harborandpine.example/pages/size-guide): measurements by product type
- [Waxed jacket care](https://harborandpine.example/pages/waxed-jacket-care): rewaxing and washing
## Optional
- [Sitemap](https://harborandpine.example/sitemap.xml): every product, collection and page
- [Journal](https://harborandpine.example/blogs/journal): trail notes and gear stories
Harbor & Pine is made up for this example, and so are its URLs.
How do you add llms.txt on Shopify?
Shopify now serves the file for you. According to the Shopify changelog (posted May 28, 2026, effective May 29), every store has a default agents.md at /agents.md, and /llms.txt and /llms-full.txt serve that same content unless you override them. Check yourstore.com/llms.txt first.
Shopify's documentation describes the default as covering the store's UCP discovery and MCP endpoints, browsing URLs for products, collections and search, published policies, and guidance for shopping agents. It omits contact details.
To write your own, go to Online Store, then Themes, then Edit code, and add one of these files in the Templates folder:
templates/llms.txt.liquidcontrols only/llms.txt.templates/llms-full.txt.liquidcontrols only/llms-full.txt.templates/agents.md.liquidcontrols/agents.mdand is the fallback for the other two.
Three details from Shopify's template documentation matter in practice:
Overriding replaces the default. The default's UCP and MCP endpoint information disappears unless you carry it into your template. The template can read an
agentsobject that includesucp_discovery_url,mcp_endpoint_urlandsitemap_url, so use those values rather than hardcoding them.Store data is limited. Only the
requestandagentsobjects are available. Objects such asshop,collectionsandpagesrender blank, so collection and product links must be written by hand.One location only. The file is served at the bare primary domain, with no locale or Markets prefix and no localized version.
Older guides tell you to fake the file with a URL redirect under Navigation. Community reports say that workaround stopped working once Shopify began serving the path natively. Use the template instead.
Because the default already points shopping agents toward UCP and MCP, the question for most Shopify brands is not "should I create one?" but "is the default enough, and what would I lose by replacing it?" For many stores, the answer is to leave the default alone and spend the effort on product data. The Shopify Shop MCP post covers the endpoint side.
How do you know whether AI systems fetch your pages?
Do not guess from a text file. Look at what actually requests your store. Nudge's Agent Analytics classifies AI traffic hitting your store by request type (crawl, citation fetch or training pull) and shows it per page, none of which appears in GA4. The pages your llms.txt lists are ordinary URLs, so you can see whether they are being pulled. Tracking has a start date per client, so counts reflect the period since tracking went live.
For the other half of the question, Nudge's Answer Engine Insights runs your topics and prompts against ChatGPT, Gemini, Perplexity and Google AI surfaces and records whether your brand and products were mentioned and which sources were cited. That tells you whether a change to your files, schema or content moved anything, which is the only test that counts.
Where a page serves an AI buyer's agent directly, Nudge's Agent Funnels build a version of the product page for agents that is fact-identical to the original, machine-readable and JSON-LD complete. That is a more reliable channel than a root-level text file, because it is tied to the product page itself. See catalog enrichment for AI shopping agents for the longer version.
One honest limit: Nudge is built for brands with catalogs of real size, 100+ SKUs. If you run a small store, a clean product schema, an accurate feed and Shopify's default llms.txt will cover most of what matters, and a platform adds little.
What are the risks of llms.txt?
A stale file is worse than none. Wrong prices, discontinued products and old policies get repeated by anything that does read it.
Prompt injection. Ahrefs flags that a text file read by AI tools can carry instructions, and recommends treating it like code: version control, restricted editing and review of anything auto-generated.
False confidence. Teams ship the file, tick the box and skip schema and feed work that has documented effects.
Private data. Shopify's docs warn against publishing contact emails or phone numbers because the file is broadly cached and served to every requester.
Should you create one?
If Shopify's default is serving and your product data is in good shape, you can reasonably do nothing. If you want one, spend an hour on it, keep it to your best pages and policies, and review it quarterly. Do it after you have fixed structured data, feeds and policy pages, not instead of them.
For the structured data side, start with the schema markup types that matter for AI citations. To see how your catalog, visibility and agent traffic fit together, book a demo.
Frequently asked questions
What is llms.txt?
llms.txt is a Markdown file at a site's root that lists useful pages for language models, with short descriptions. Jeremy Howard proposed it in September 2024 at llmstxt.org. The only required element is an H1 with the site name. It informs; it does not block or grant access.
Do ChatGPT, Gemini and Perplexity use llms.txt?
We found no published statement from OpenAI, Google or Perplexity that its answer engine reads llms.txt on other sites. Google's Search documentation says you do not need such files to appear in Google Search. OpenAI and Perplexity publish llms.txt for their own docs, which is a different thing.
Will llms.txt improve my AI visibility?
There is no vendor statement or published study showing it does. An Ahrefs analysis of 137,210 domains found 97% of llms.txt files got no requests in May 2026, and called the file largely decoration for AI search. Product schema, feeds and strong content have stronger evidence behind them.
Does Shopify support llms.txt?
Yes. Since May 29, 2026, every store serves /llms.txt by default, using the same content as /agents.md. You can override it with templates/llms.txt.liquid in your theme. Overriding replaces the default, so carry over the UCP and MCP endpoint values if you want them kept.
What should I put in an ecommerce llms.txt?
A short brand summary, your main collections, 10 to 30 top products, shipping, returns and warranty pages, sizing and care guides, and your sitemap URL. Keep facts identical to your live pages. Leave out private contact details, discount codes and anything that changes often.
