llms.txt is a proposed plain-text file, placed at the root of a website, that lists the pages a large language model should read to understand the site. For an online store it can point AI assistants to category guides, shipping and returns policies and sizing information. It is cheap to publish, but no major AI provider has confirmed it uses the file for ranking.
Where did llms.txt come from?
The format was proposed in 2024 by Jeremy Howard of Answer.AI. The idea is simple: web pages are full of navigation, scripts and ads, and a model with a limited context window benefits from a curated map written in Markdown. The file sits at /llms.txt, much like robots.txt sits at /robots.txt.
The proposal describes a heading with the site name, a short summary, and sections of links with one-line descriptions. An optional companion, llms-full.txt, can contain the full text of key pages in one document.
Does any AI platform actually read it?
This is the honest question, and the answer at the time of writing is unclear. Some developer tools and documentation platforms read llms.txt, and several companies publish one. Google has said publicly that it does not use llms.txt for Search or AI Overviews. OpenAI, Anthropic and Perplexity have not documented a formal commitment to it for their consumer search products.
So treat the file as a low-cost signal and a useful internal exercise, not as a lever that moves rankings. If an assistant fetches it, good. If not, the work of deciding which pages matter still pays off elsewhere.
What should a store put in llms.txt?
Shoppers ask assistants practical questions: does this brand ship to my country, how do returns work, which size should I buy. The file should point to the pages that answer those questions accurately.
- About and contact: who runs the store, where it is based, how to reach support.
- Shipping: countries served, typical delivery windows, carriers.
- Returns and warranty: the policy page, not a marketing summary.
- Sizing and fit guides, including how measurements are taken.
- Main category pages with one-line descriptions of what they contain.
- Buying guides and comparisons that explain the range.
Avoid listing thousands of product URLs. The file is a map, not a sitemap. Product feeds already serve that job through Google Merchant Center and similar channels.
What does an example look like?
A Shopify store selling handmade ceramics might publish a short file: a heading with the brand name, a two-sentence summary saying it sells stoneware tableware shipped from Istanbul to the EU and UK, then sections titled “Policies”, “Guides” and “Collections”, each with three to six links and a line of description. The whole file might be under a hundred lines.
On Shopify you cannot drop arbitrary files at the root by default, so stores usually create a page or use an app or a proxy route that serves the text at /llms.txt. On WooCommerce, the file can be uploaded directly to the web root or generated by a plugin.
How Alien Road applies llms.txt for store clients
In a typical engagement llms.txt is one small part of a wider AI visibility audit. Discovery starts with questions shoppers actually ask, taken from support tickets, site search logs and Search Console. We then check which pages answer them and whether those pages are accurate and crawlable.
Setup means writing the file by hand with the client’s team, because the descriptions must be factual. We serve it with a text/plain content type, link the policy pages that matter, and confirm it loads without a redirect. Where the platform allows, we also generate llms-full.txt for policies and guides.
Measurement is limited, and we say so. We check server logs for requests to /llms.txt by user agent, and we test a set of shopper questions in ChatGPT, Perplexity and Gemini before and after. Reporting lists which assistants cited the store and which pages they used, without claiming the file caused it.
What else matters more than the file?
If an assistant cannot find your returns policy, llms.txt will not save you. The basics still decide whether a store is cited.
- Pages are crawlable and not blocked for the bots you want, such as OAI-SearchBot or PerplexityBot, in robots.txt.
- Policies are written as plain HTML text, not inside images or pop-ups.
- Product structured data is valid, including price, availability and reviews where they are genuine.
- The brand name, address and contact details are consistent across the site and marketplaces.
Common mistakes
Stores sometimes copy a template and forget to update it, so the file lists pages that 404. Others paste marketing copy into the descriptions, which reads poorly when quoted. A third mistake is listing staging URLs or filtered category parameters. Keep the file short, factual and reviewed whenever policies change.
How do you check whether assistants use the file?
Server logs are the only direct evidence. Filter requests to /llms.txt and look at the user agent strings. Known AI crawlers identify themselves, for example GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot. If none of them request the file over a few months, that tells you something useful about its current value.
Indirect checks help too. Ask assistants the shopper questions you mapped and note which URLs they cite. If they quote your returns page correctly, the page is doing its job, with or without the file.
Keep a short review routine
- Monthly: confirm the file loads and every link returns 200.
- When policies change: update descriptions on the same day.
- Quarterly: compare cited URLs in assistants with the file’s list.
If you sell through marketplaces as well, such as Trendyol, Amazon or Etsy, keep the same policies and brand facts there. Assistants often cite marketplace listings, and a mismatch between them and your own site creates confusing answers.
Where to start this week
Write down the ten questions your customer service team answers most. Find the URL that answers each one, check that it is accurate, and draft a llms.txt with those links grouped under clear headings. Publish it, request it in a browser to confirm it loads as plain text, and set a reminder to review it whenever shipping or returns terms change.
Keep reading: GEO – AIO – AEO · GEO