llms.txt and schema: how AI assistants learn who you are

Ask an AI assistant to describe a local business and it will produce a confident paragraph. Whether that paragraph is right depends on what the business's own website told it.

Two files, two kinds of reader

A website has two kinds of readers now. People, who look at the page. And machines, which read the code behind it and try to work out what the business is, where it is, and what it sells. Google has been the main machine reader for years. AI assistants are the newer ones, and they are the ones writing paragraphs about your company when someone asks.

Schema and llms.txt are the two things we add so the machine readers do not have to guess.

What schema does

Schema, also called structured data, is a block of code on each page that labels the facts on that page in a format machines read directly. A person sees a heading, an address and a phone number. The schema says: this is a local business, this is its name, this is its street address, these are its hours, this is the service this page is about, this is the area it serves, and here are the questions and answers on this page.

Without it, a search engine reads the paragraphs and infers. Inference is where mistakes come from. An advisor's page that mentions three cities might be read as serving all of them or none. With schema, the page says which one, in a form that cannot be misread.

We put schema on every page, not just the homepage. The kinds we use most:

• The business itself: name, address, phone, hours, area served. This one goes on every page.
• The service, on a service page.
• Questions and answers, on any page that has them.
• The article, on a blog post, with its date and topic.
• The breadcrumb trail, so a machine knows where a page sits in the site.

What llms.txt does

llms.txt is a plain text file that lives at the top level of the site, the same place the robots file and the sitemap live. It is written for AI tools rather than for Google. In a few short sections it says who the business is, what it does, where it works, what makes it different, and which pages are the important ones to read for more detail.

It is a convention, not a standard every AI tool is required to follow, and as of this writing how widely it is read varies. We add it anyway because it costs almost nothing and because it gives us one place where the description of the business is complete and current. If an AI tool reads it, good. If one does not, the schema on each page carries the same facts.

Generated from the site, not typed by hand

The mistake we see most often is schema that was pasted in once and never touched. The business moved, the hours changed, a service was dropped, and the hidden label still says the old thing. Now the machine reader has two sources that disagree, and it trusts neither.

So we do not type either file. The site builds them. The business facts live in one place. The pages pull from it, the schema pulls from it, and llms.txt pulls from it. Change the phone number once and all three change. Publish a new service page and it appears in llms.txt on the next build. The custom home builder's site does this for every one of its project pages, and the advisor's site does it for every one of its posts.

What changes when an AI assistant describes you

When someone asks ChatGPT or Perplexity for a cabinet maker in a particular town, the answer is a paragraph, not a list of links. The paragraph is built from whatever those tools could read about the businesses they found. A site with clear schema and a clear llms.txt gives them clean facts to repeat: the right name, the right towns, the right services. A site without them gives them paragraphs to interpret, and interpretation is where the wrong town, the wrong service, or a competitor's phone number comes from.

We cannot control what an AI assistant says. We can control what it has to work with, and that is what these two files are for.

Both are part of every site we ship, alongside the named assistants we build that read those same pages.

Questions people ask