Your website has two audiences. Humans see the design. Machines (Google's crawler, ChatGPT's retriever, Claude's browser, Perplexity's index) see the markup. More and more of your customers meet the machine's version first, because an assistant or a search engine summarizes you before a human ever clicks. Making that version accurate is now table stakes. Here are the four layers, cheapest first.
Layer 1: semantic HTML
Machines parse structure, not pixels. A real heading hierarchy (one h1, meaningful h2s), lists marked up as lists, addresses in text rather than images, prices as text. If your key facts live inside a JavaScript widget or a JPEG, to a crawler they do not exist. This layer is free; it is just discipline.
Layer 2: structured data
Schema.org JSON-LD is you filling out the machine's form directly: business name, type, areaServed, services, hours, author, dates. Search engines use it for rich results. AI systems use it as high-confidence facts. A retriever choosing between two detailing shops will describe more accurately, and cite more readily, the one whose pages carry clean LocalBusiness and Service markup.
Layer 3: sitemaps and robots.txt
The sitemap is the map; robots.txt is the guest list. Keep the sitemap current and honest (no dead URLs, real lastmod dates), and make sure robots.txt is not silently blocking the crawlers that feed AI answers: GPTBot, ClaudeBot, Google-Extended, PerplexityBot. Blocking them is a choice; make it deliberately, not by plugin default.
Layer 4: llms.txt
The newest layer: a markdown file at /llms.txt that gives language models a curated summary of your site: who you are, what you offer, key pages, how to get in touch. Think of it as a press kit for machines. The convention is young, but the cost is an hour of writing, and the crawlers already fetch it. This site ships one; view it and steal the format.
How the layers pay off together
When someone asks an assistant "who builds restaurant apps in Nova Scotia?", the assistant searches, reads a few pages, and answers. Clean HTML means it parses you correctly. Schema means it trusts what it parsed. The sitemap means it found the right pages. llms.txt means the summary in its context window is the one you wrote. Miss all four and the machine guesses; machines guess conservatively, and conservative guesses go to your competitors.
None of this is exotic. It is an afternoon of work on a well-built site and nearly impossible on a bloated page-builder theme, which is the real argument for building it right the first time. If you want to know what the machines see when they read your site today, email me and I will send you the honest answer.