Suppose a marketing salesperson walks into your shop tomorrow: “Search is entering a new era. Without llms.txt, AI can’t see you.” He opens his price quote, and suddenly your shoe repair shop, which has been open for twenty years, feels as though it forgot to get a boarding pass.
I lean toward putting the pen down for now. What small businesses urgently need may not be llms.txt, but service details that strangers and machines alike can verify. The file can be added later; if you haven’t put your value into words, another URL won’t make it appear on its own.
It’s a Guide Sheet, Not a Guarantee of Recommendations
llms.txt is a proposal: place a Markdown text file on a website, usually at /llms.txt, that brings together an introduction to the site, links to important pages, and necessary explanations to help large language models use the site’s content. Imagine preparing a guide sheet for a building, telling visitors: the repair counter is here; see that page for service limitations.
Its purpose differs from robots.txt, which manages crawler access rules, and it cannot serve as a replacement for a sitemap. More crucially, providing a guide does not mean an AI search service will read it; reading it does not mean it will recommend you. Between “I’m ready” and “they’ve adopted it” lies the design of their system.
So if someone equates installing the file directly with “improving your AI ranking,” I’d ask them to lay out the causal chain in between.
The Busier AI Gets, the Less You Should Rely on Pretty Words
The angle in the news material about AI agents consuming computing resources made me think of another possible future: when machines run errands for people and every read has a cost, competition among businesses may look more like “who can quickly match a need with evidence.” This is an extrapolation, not a proven ranking formula.
Someone asks: “Where in Tainan can I get an old leather handbag my mother left me repaired? The lining is sticky, and I can only drop it off on Saturday.”
Shop A’s website says “artisan dedication, top-tier craftsmanship,” and its llms.txt is neatly organized. Shop B has no such file, but its service page clearly states that it replaces linings in old leather handbags, which materials require photos first, and its Saturday drop-off hours, along with restoration examples published with permission.
If I were that customer, Shop B would already have addressed my worries. Which AI might choose it still depends on how the system obtains and interprets the information; but Shop A’s polished guide has not filled in what the customer actually wants to ask.
The assumption most easily slipped in here is this: if a machine understands the format, it will understand your value. A format can organize text; it cannot discover for you the kind of skill behind “others won’t take this job, but I’ve repaired it more than a dozen times.”
Some Shops Need It; Others Are Just Being Pushed to Need It
If you run a website with extensive documentation, service specifications, and frequently asked questions, organizing an llms.txt file makes sense: it brings scattered entry points to information together, making it easier for tools that support the proposal to use them. If the cost of creating and maintaining it is very low, it can also be treated as an experiment.
But for a small shop with only a few website pages, I lean toward first updating its opening hours, range of services, pricing conditions, and examples of past work. With a limited marketing budget, buying a new format first while leaving a three-year-old phone number in place looks a bit like installing a smart doorbell on a house with a leaking roof.
The shoe repair craftsman sits back down at his workbench and picks up a pair of hiking boots with cracked soles. He says: “With these, I’d check the midsoles first. Some won’t last long even if you glue them back together.” The marketer, who was about to enter “professional shoe repair,” pauses, then writes down the craftsman’s words exactly as he said them.