We Added llms.txt to This Site. It Probably Does Nothing Yet.

This site has a file at /llms.txt. It is a tidy markdown index of everything we publish — services, industries, articles, the demo builds — generated automatically so it cannot drift from what is actually on the site.

We think it was worth adding. We do not think it is currently doing anything for our visibility, and we would rather say so than let you assume otherwise.

What llms.txt is

It is a proposal: a standard location for a markdown file that gives a language model a curated map of a site, so it can find the useful pages without wading through navigation and boilerplate. The idea was published in September 2024 by Jeremy Howard of Answer.AI. The specification lives at llmstxt.org.

The reasoning behind it is sound. Models work with limited context. A clean index of what a site contains is genuinely more useful to a machine than a rendered page full of menus, cookie banners and footers.

What it is not

It is not, as far as anyone outside the major AI labs can demonstrate, an input to how those systems retrieve or cite. As of this writing in 2026, no major provider has published documentation saying they read it, weight it, or treat it as a signal.

Compare that to the two files it is often mentioned alongside. Robots.txt dates to 1994 and its handling is documented by every major crawler operator. Sitemaps date to 2005 and both Google and Bing document what they do with them, including how to submit one. You can look up the behavior. With llms.txt you cannot, because nobody has published any.

That is not evidence it does nothing. It is the absence of evidence that it does something — which is a different claim, and the difference matters when someone is charging you for it.

Why we added it anyway

Three reasons, none of them "it will get us cited."

It costs essentially nothing. Ours is generated from the same data the site renders from and the sitemap uses. It updates itself. Nobody maintains it. A hand-written one would be a second copy of the site's structure that starts rotting the day it is written, which would be a real cost — so we did not write one by hand.

The downside is bounded. It is a public file listing public pages. It reveals nothing, slows nothing, and breaks nothing.

If adoption arrives, we are already there. Cheap options on uncertain futures are reasonable. The mistake is not buying the option; the mistake is booking the payoff before it exists.

The pattern worth recognizing

This is a good miniature of how a lot of AI search advice works right now. A plausible mechanism appears. It gets described as best practice. Somewhere in the retelling, "this might help" quietly becomes "this helps," and eventually it is a line item.

The tell is always the same: ask what would have to be true for the claim to hold, and whether anyone has checked. For llms.txt the question is simply does any retrieval system fetch this file and use it? That is answerable in principle. Until someone answers it publicly, the honest description is "unproven, cheap, harmless."

You can apply the same test to schema markup, to whatever file convention is proposed next, to any of it. Structured data, for what it is worth, has a much stronger case: it is documented, it demonstrably drives rich results in traditional search, and it makes your facts machine-readable regardless. But even there, "helps a machine parse your page" and "makes a model recommend you" are two different claims, and only the first is established.

What we would spend money on instead

If you have a budget for AI visibility and you want it spent on things with a mechanism behind them:

Content that is actually in your HTML, because a crawler that does not run JavaScript cannot cite what it cannot read. Pages that answer a real question in a way that could be quoted in a sentence. Being genuinely referenced elsewhere — mentions, listings, citations — since retrieval systems are built over the web, and the web's opinion of you is largely other pages. And accurate, current facts about your business in the places machines look, because the most common AI visibility problem we see is not absence, it is being described wrongly.

None of that is novel. That is rather the point.

Next steps

If you want the file, take ours: it is at /llms.txt and the approach is a hundred lines of code generating it from the site's own content. We are happy for anyone to copy the idea.

If you want the parts that are load-bearing, that is what our AI and search work actually is — and we will tell you which parts are proven and which are bets, in those words.

Ask us what is worth doing.