Does an Online Store Need an llms.txt File?

Hardly, for now. In May 2026, Ahrefs tracked 38,000 llms.txt files: 97% received no requests during the month.

On Shopify, the file already exists without you doing anything. Leave it in place, don't pay for any app, and spend that time on your product feed.

The benchmark

Ahrefs looked at the 137,210 domains using its traffic analytics tool, and a little over a quarter of them published an llms.txt.

Among the few requests received, SEO audit tools come first at 21.7%, even though they crawl the whole site rather than looking for this file in particular.

The search bots from ChatGPT, Claude and Perplexity combined account for 1.1%, or 233 calls in a month. Google states in its guide to generative search, updated in July 2026, that no new machine-readable file is required to appear there.

We fell for it too. Our August brief for this site already cited those 97% and still planned an hour of work on this file, on the grounds that "Perplexity and Claude read it."

Yet in the same data, Slackbot ranks ahead of PerplexityBot, and Claude Code, a coding agent, fetched more llms.txt files than any search bot or assistant.

At Google, John Mueller sees it, according to Ahrefs, as a crutch for coding tools that read technical documentation, and a store has little of that to publish. Our own file exists today, generated by our publishing tool, and we still don't know who reads it.

The rule

On a Shopify theme, the platform has decided for you. Since May 2026, every store responds at /agents.md, and the addresses /llms.txt and /llms-full.txt return the same content (changelog of May 28, 2026).

We opened this file in early October 2026 on seven stores, five of them French: all served the same text of about 560 words, where only the name, the domain and the links to the policies change.

You won't find a single line about what you sell. The text explains to agents how to buy from a Shopify store: install the Shop shopping module, pay with Shop Pay, use the entry points of the Universal Commerce Protocol and get the buyer's consent at checkout.

As we read it, this file works first for Shopify, which steers agents there toward its own payment tools. That takes nothing away from you, since your products reach them through the catalog and the feed.

The steps

The first two take ten minutes.

1. Check. Open https://votreboutique.fr/llms.txt. On a Shopify theme, the page responds and starts with "Agent Instructions."
2. Apps. Check whether any of yours promises to generate this file and note its monthly price, since the platform already serves it.
3. Customization. Shopify recommends keeping the managed version unless you have specific instructions to give agents. If you do, create templates/agents.md.liquid from Online Store > Themes > Edit code: this template feeds all three addresses. Allow an hour and stop there.
4. The worthwhile project. Spend the time you saved on your Merchant Center feed and your catalog.
5. Revisit. Reopen the question the day a mainstream assistant commits to reading this file, which none has done so far according to Ahrefs.

The trap

The risk comes from the customization itself.

As soon as you create your template, it replaces the version managed by Shopify, and the documentation warns that you become responsible for keeping it up to date.

On a catalog that changes often, a hand-written text goes stale fast, with dead links and prices from before the soldes, and you have to maintain it when almost nothing reads it.

If you forget to carry over the protocol addresses, you also remove from the file the part an agent needs in order to buy.

The same time spent on the feed benefits a channel whose usage can be measured. In a study published in June 2026, Profound tracked about a million shopping listings: the share of product feeds in what ChatGPT retrieves rises from 4.3% to roughly 20% in six weeks.

The sentence for your leadership

Copy this sentence, then rewrite it with what you found in the first two steps.

"Our llms.txt already exists, Shopify generates it. According to Ahrefs, 97% of these files received no visits in a month. I propose we pay nothing to produce it and spend that time on the product feed, which ChatGPT uses more and more."