llms.txt is a proposed convention: a Markdown file at a site's root that points AI systems to the pages a publisher considers most useful, in a clean format free of navigation, scripts and markup. It was proposed in late 2024 and has been widely discussed since, largely by analogy to robots.txt and the XML sitemap.
The format is deliberately simple — a Markdown document with an H1 for the site name, a short blockquote summary, and linked lists of key pages grouped under headings, each with a one-line description. A companion convention, llms-full.txt, holds the full text of that content in one file. Both live at the domain root, alongside robots.txt.
The intent is to reduce the work an AI system has to do to understand a site: instead of crawling and parsing HTML pages laden with menus, cookie banners and scripts, it reads a curated list of what matters, already in the format language models handle best.
This is where most write-ups get vague, so it is worth being plain. No major AI provider has confirmed that it reads llms.txt. Google has publicly said it is not used for AI Overviews or AI Mode. OpenAI, Anthropic and Perplexity have not documented support for it either. Its adoption to date is on the publishing side — many sites now serve one — rather than the consuming side.
That asymmetry is the whole story. A convention only works when the consumers honor it, and so far they have not committed to. Anyone claiming measurable ranking or citation gains from adding the file is describing correlation, not a documented mechanism.
The case for is that it costs very little. A single static file, generated once from the site's key pages, carries no risk of harming search performance and positions the site for a convention that may yet be adopted. Documentation-heavy sites in particular report that it makes their content easier for developers to feed into AI tools manually, which is a real if modest benefit today.
The case against is opportunity cost and false confidence. The hours spent hand-curating an llms.txt are hours not spent on the things AI systems demonstrably do use — clean, well-structured, factually specific pages that retrieval can lift passages from. A brand that adds llms.txt and considers its AI strategy complete has made itself worse off than one that ignored it entirely.
Treat llms.txt as a cheap hedge, not a lever. Generate it, keep it current, and spend the actual effort on content structure, factual specificity and third-party presence — the inputs that affect AI search optimization today rather than hypothetically. On Shopify specifically, serving a file at the domain root requires either theme-level routing or a proxy, which is worth knowing before committing to the idea.
Deciding what belongs in that curated set, and making the underlying pages worth citing, is the substance of answer engine optimization. The related question of how AI agents interact with a storefront programmatically is covered under agentic commerce setup.
.png)
.png)
