llms.txt: The New File Your Website Needs.

Search engines have crawled and indexed the web for twenty-five years using a predictable contract: robots.txt tells crawlers what they can touch, sitemap.xml tells them what exists, and the rest gets figured out by parsing HTML. AI assistants work differently. When ChatGPT, Claude, or Perplexity answer a question about your business, they are not crawling your site in real time the way Googlebot does — they are often working from a cached or retrieved snapshot, trying to extract facts from pages built for human eyes, full of navigation menus, cookie banners, and marketing copy that has nothing to do with the question being asked. llms.txt is the small, plain-text fix for that mismatch.

What llms.txt actually is

It is a markdown file at the root of your domain — yourdomain.com/llms.txt — written for a language model, not a browser. No CSS, no navigation, no cookie notice. Just your business name, what you do, who you serve, your core facts, and clean links to the pages that matter most, described in a sentence each. Think of it as the elevator pitch you would give an extremely literal assistant who has ninety seconds and needs to decide whether to recommend you.

Why it exists at all

Large language models are trained on a mix of static knowledge and, increasingly, live retrieval. When a model retrieves your live page to answer a question, it has to strip out everything irrelevant and extract the facts that matter — your service area, your pricing range, your hours, what makes you different from the next search result. That extraction step is lossy. A clean llms.txt file skips the extraction problem by handing over the distilled version directly. Several AI tooling companies proposed the convention in 2024, and by 2026 it has moved from novelty to a quiet best practice among businesses actively investing in AI search.

What belongs on yours

Keep it short and factual. A one-paragraph description of the business, the service area, the core services with one line each, pricing if you are comfortable publishing a range, hours, and a handful of links to your most important pages — contact, pricing, and your two or three best service pages. Avoid marketing language. "Family-owned HVAC company serving Katy and west Houston since 2009" is more useful to a model than "Houston's premier comfort solutions provider." Specific, plain, checkable facts are what gets repeated back to a customer.

The businesses showing up in AI answers right now are not the flashiest — they are the ones that made their facts impossible to misread.

How it fits with everything else

llms.txt does not replace structured data, and it does not replace a well-written site. It sits alongside schema markup and ordinary SEO as one more signal pointed at a different reader. We cover the broader discipline in our AI search optimization work — llms.txt is one of the easiest pieces to ship, usually in under an hour once the facts are gathered, and one of the first things we add for clients once a rebuild goes live.

Checking and maintaining it

Because it is plain text, you can open it in any browser and read it top to bottom — there is no tool required to verify it looks right. The maintenance burden is light: update it whenever your hours, pricing, or service area change, the same moment you would update your Google Business Profile. Stale facts in an llms.txt file are worse than no file at all, because a wrong hour or an old phone number gets repeated by an assistant with total confidence.

Why it won't replace your actual website

llms.txt is a summary, not a substitute — it works alongside your real pages, pointing toward them rather than replacing the depth they contain. A model that wants more than a one-line description of your service still follows the link into the full page, which is exactly why the rest of your site still needs to be clear, specific, and well-structured. Think of llms.txt as a table of contents written for a very fast, very literal reader, not as the whole book.

A simple starting template

If you want to see the shape before committing to the wording, a working llms.txt usually opens with an H1 of your business name, followed by a one-line summary, then an H2 section listing your core offerings as a short bulleted list, then an H2 "Important pages" section with markdown links and a one-sentence description of each. That is the entire format — no front matter, no metadata block, nothing a model has to parse around. The simplicity is the point; anything that looks like it was built for a browser defeats the purpose.

Common mistakes we see

The most frequent one is treating llms.txt as a marketing document and filling it with the same adjective-heavy copy as the homepage. A model does not need to be persuaded that you are "the best" at anything — it needs facts it can repeat with confidence. The second mistake is letting it go stale: a file that still lists a service you dropped two years ago, or a phone number from a previous office, does active harm, because an assistant will repeat that wrong information with the same confidence it would repeat the correct version. The third is skipping it because "nobody has asked yet" — by the time a competitor's llms.txt is already feeding clean answers to customers, you are playing catch-up on a file that takes under an hour to write.

Does it actually move the needle yet?

Adoption among AI products is uneven — some retrieval systems clearly favor it, others still lean mostly on ordinary page content and structured data. That uncertainty is exactly why it is worth doing now rather than waiting for consensus: the cost of shipping one is close to zero, the downside of having a wrong or missing one compounds the longer it sits unaddressed, and early movers in any new convention tend to keep an edge even after it becomes standard practice. We treat it the way we treated schema markup in its early years — cheap insurance against a channel that is clearly growing.

Every site we build ships with one as standard. If you are running an older site and want to add llms.txt without a full rebuild, that is a short, standalone piece of work — get in touch and we can usually turn it around in a day. It is a small file with an outsized return relative to the hour it takes to write, and one of the easier wins available to any business willing to spend that hour.

Watch on YouTube: AI Search — llms.txt explainedCurated search — opens the best current videos on this topic

Written by Rizwan Prasla, Spring Web Designer — the Houston studio behind 100 demo builds with zero repeats. When you’re ready: start a project or get your number in ten seconds.