A founder forwarded us a message from an agency last week: "You need an llms.txt file or ChatGPT will never see your store." The email came with a $400 setup fee attached and a deadline that implied every day without one was lost revenue.
That message is going around right now, and it is not entirely wrong, it is just about eighteen months out of date. llms.txt for ecommerce stores is one of the most confidently oversold ideas in AI search advice this year, and the platform most of these emails target, Shopify, already moved past it.
Here is what the file actually does, what changed in May 2026, and where the effort behind that agency invoice would do more for your store.
What llms.txt actually is
llms.txt is a plain text file published at the root of a domain, yourstore.com/llms.txt, written in Markdown, that gives a large language model a short map of a site: what it does, which pages matter most, and links to clean, text-only versions of key content. The specification was proposed in 2024 by Jeremy Howard as a companion to robots.txt, aimed at AI systems rather than search crawlers, on the idea that models waste time and context parsing navigation, ads, and boilerplate when a page's actual substance is a few sentences long.
For an ecommerce store, the pitch is straightforward: point the file at your product catalog, your shipping and return policy, and your FAQ, and an AI shopping assistant supposedly reads that instead of guessing from a rendered page full of upsell modules and cookie banners.
The idea makes sense on paper. The evidence that it works is thin, and the platform layer under it just changed.
What Shopify actually shipped in May 2026
If your store runs on Shopify, the agency pitch is already stale. In May 2026, Shopify made /agents.md the canonical AI discovery file for every storefront on the platform, and the older /llms.txt and /llms-full.txt routes now serve that same content by default unless a theme developer overrides them.
Shopify's own theme documentation describes agents.md as the file that tells AI agents and shopping assistants how to discover a store's commerce capabilities, including its Universal Commerce Protocol and Model Context Protocol endpoints, none of which the original llms.txt format was designed to express. Shopify generates and maintains this file automatically for most stores. A theme developer can override it with a custom agents.md.liquid, llms.txt.liquid, or llms-full.txt.liquid template, but for the vast majority of merchants, the managed default is already doing the job an agency is charging to build from scratch.
Stores on WooCommerce, BigCommerce, or a custom stack do not get this automatically, and adding a hand-written llms.txt there is cheap enough that it is not worth arguing against. Just do not expect it to move much on its own, which is the part most of these pitches leave out.
The evidence gap nobody mentions
The stronger reason to slow down before paying for this work is that the measurable benefit has not shown up yet. SERanking analyzed roughly 300,000 domains to test whether having an llms.txt file correlated with how often a domain got cited by AI models. It found no meaningful relationship. When the researchers removed llms.txt as a variable from their prediction model, the model's accuracy improved rather than declined, which is a stronger signal than a null result, it suggests the file was adding noise, not weak signal.
That single number tells the whole story better than any theory of how it should work: at the scale AI systems actually operate, llms.txt has not yet earned the trust that would make a model prioritize it over the content it can already crawl and parse directly.
None of this means the standard is dead. Adoption is still early, and a spec that a handful of major labs decide to lean on could matter more in a year than it does today. It means the file is a low-cost bet, not the load-bearing fix an urgent-sounding sales email implies.
Where the effort actually pays off right now
The things that do correlate with AI shopping visibility today are the same things that have mattered since Google Shopping existed: structured, accurate product data that a machine can parse without guessing.
That means a current Google Merchant Center feed with correct prices, availability, and GTINs, and complete Schema.org Product markup on every product page. Our breakdown of AI shopping agents and ecommerce tracking goes into why several AI shopping surfaces, including ChatGPT Shopping and Google's AI Mode, read from that exact feed and markup rather than a bespoke AI-only file, and why a stale feed is a more common reason a product goes invisible to an agent than a missing llms.txt ever will be.
It also means checking whether your analytics can see this traffic at all once it arrives. A shopper who discovers a product through an AI assistant and completes checkout through a delegated agent flow can generate revenue that never touches a page view your GA4 property recognizes. If nobody has built the reconciliation path between your order system and your analytics, that revenue is happening and your reporting does not know it.
There is a third layer most stores skip entirely: the content an AI system actually reads once it lands on a product page. A page that renders its price, availability, and shipping terms only after a JavaScript bundle loads is invisible to a model that is not executing a full browser session. The same goes for FAQ content buried inside an accordion widget that requires a click to expand, or return policy details that only exist as a PDF linked from a footer. None of that is an llms.txt problem. It is a rendering and markup problem that predates the current wave of AI search advice by a decade, and it is the reason a store can have a technically perfect discovery file and still be represented inaccurately by an AI assistant that never found the details in the first place.
What an honest 30-minute check looks like
Before paying anyone to touch a discovery file, run this yourself.
- Load
yourstore.com/agents.mdandyourstore.com/llms.txtin a plain browser tab. Note what resolves, what matches, and what 404s. - Pull ten product pages and view the page source, not the rendered page. Confirm the price, availability, GTIN, and description are present in the raw HTML or a JSON-LD block, not only injected client-side after the page loads.
- Open your Google Merchant Center feed and spot-check the same ten products against the live site. A mismatch here is a far more common cause of an AI system citing the wrong price or an out-of-stock item than anything a discovery file controls.
- Check whether GA4 has a defined channel or source for AI referral traffic. If every AI-driven session is currently landing in "Direct" or "Unassigned," that is a measurement gap worth fixing before worrying about how a model discovers the catalog in the first place.
That checklist takes less time than reading the agency's proposal, and it tells you which of the three layers, discovery file, structured data, or measurement, is actually broken before you pay to fix the one that probably is not.
The decision, not a checklist
Skip the standalone llms.txt project if you run Shopify. The platform already ships an equivalent file, and the setup fee is buying something you likely already have.
Add a basic llms.txt yourself, in an afternoon, if you run WooCommerce, BigCommerce, or a custom storefront and want to hedge a low-cost bet on a standard that might matter more later. Do not pay an agency four figures for it, and do not expect a citation lift from it alone.
Either way, spend the real budget on the foundation under all of this: a product feed and structured data an AI system can trust, and analytics that can actually see what happens after an AI-driven visit converts. That is exactly the gap Anlyto's AI readiness audit is built to find, scoring the data, process, and structured-content gaps that decide whether an AI shopping agent can represent your catalog accurately at all. If you are not sure whether the issue is AI readiness specifically or something more basic in how your store's numbers get tracked in the first place, an analytics audit is the faster way to rule that out before spending on either.
The next agency email that tells you AI cannot find your store without a $400 text file is worth a second look, and a look at your own /agents.md first.
