The llms.txt and Schema Checklist: Making Your Site Readable to Machines

Most of the advice in my AI search series is about what you say. This post is about whether machines can read it at all.

Think of it like the trades. Beautiful finish work means nothing if the framing is wrong. What follows is the framing: the files and markup that decide whether crawlers, search engines, and AI systems can parse your site, understand your facts, and quote you with confidence. It is a checklist because that is what it should be for you, something you work through once, properly, and then mostly leave alone.

None of this requires genius. All of it requires care.

Part one: the two text files at your front door

robots.txt: make sure you are not blocking your own customers

Every crawler that visits your site checks robots.txt first for permission. Here is the thing I keep finding in audits: sites accidentally blocking the exact bots they need. Sometimes a developer blocked everything during the build and never unblocked it. Sometimes a security plugin got enthusiastic.

The AI crawlers that matter right now include GPTBot and OAI-SearchBot (OpenAI, the second one powers ChatGPT’s live search), PerplexityBot, Google-Extended (Google’s AI training), and ClaudeBot (Anthropic). If appearing in AI answers is something you want, none of these should be disallowed.

Check yours right now: visit yourdomain.com/robots.txt and read it. If you see Disallow: / under User-agent: *, or any of the names above being blocked, you have found a problem worth fixing today.

llms.txt: twenty minutes, zero risk

llms.txt is a newer idea: a plain text file at your site root written specifically for AI systems. A short description of your business, your services, your location, and links to your most important pages, in clean simple language.

Honest disclosure, because this series does not do hype: llms.txt is young and adoption by the engines is uneven. Nobody can promise you it moves the needle today. But the cost is twenty minutes, the risk is zero, and the logic is sound: it puts your clearest self-description in a standard location machines already know to check. I include one on my own site and on client sites, because cheap insurance on a growing standard is exactly the kind of bet worth making early.

Keep it factual and short. What you do, who you serve, where, since when, and links to your key pages with one-line descriptions. No slogans. Slogans are for humans, and barely.

Part two: schema, or saying your facts in a language without ambiguity

Schema markup is structured data embedded in your pages that states your facts in a format built for machines. Where a paragraph says “family-run plumbing company in Sudbury,” schema says, unambiguously, in a defined vocabulary: type LocalBusiness, name, address, telephone, areaServed, foundingDate.

The reason this matters more every year: engines that assemble answers cross-reference. Data they can read directly beats prose they have to interpret. Here are the four types that cover most small businesses, in the order to do them.

1. LocalBusiness (or Organization)

Your identity card. Name, address, phone, hours, service area, links to your profiles. One rule above everything: it must match your visible content and your Google Business Profile exactly. Schema that contradicts what your page displays does not clarify, it undermines. Consistency is the entire value.

2. Service

One for each core service you offer, saying what it is and where you offer it. This is the markup that connects “who does X” questions to your business, and most small business sites skip it entirely, which is your opening.

3. FAQPage

If a page genuinely answers customer questions, FAQ markup turns each question and answer into clean, quotable data. AI Overviews and answer engines love this format for obvious reasons, it is pre-digested. I covered why question-shaped content wins in my AI Overviews guide. Two cautions: mark up only questions that actually appear on the page, and only questions a real customer would ask. Stuffing invisible or fake FAQs is a known spam pattern and gets treated like one.

4. Article, with a real author

For blog posts and guides: headline, date, and an author who is a real person connected to a real about page. As AI-generated filler floods the web, engines increasingly discount anonymous content. A traceable human byline is quietly becoming one of the strongest trust signals you can attach to writing. It is also, not coincidentally, the one thing content farms cannot fake well.

Part three: verify, because markup fails silently

Broken schema does not show an error on your page. It just quietly does nothing. After adding or changing anything:

1. Run the page through Google’s Rich Results Test. It shows exactly what Google can read and flags errors.
2. Read your rendered page with your own eyes and confirm every fact in the markup appears in the visible text too.
3. Re-check after any site redesign or platform migration. Schema is the thing that silently falls off when themes change. I have seen years-old markup vanish in a rebuild and nobody noticed for months.

My free AI search grader checks several of these automatically, including whether your schema, robots.txt, and llms.txt are present and readable. It is the fastest way to find out which parts of this checklist your site already passes.

The point of all this

Everything above serves one goal, the same one running through this whole series and spelled out in the full AI search guide: when an engine goes looking for a business like yours, it should find clear, consistent, machine-readable evidence that you are real, competent, and exactly what the customer asked for.

The framing is not glamorous work. It is also the difference between a site that machines can vouch for and one they politely skip. If you would rather hand someone the toolbelt, I do this for a living.