Making a Client Site AI-Ready: What I Actually Check

Things I'm Trying

Making a Client Site AI-Ready: What I Actually Check

Every client asks about AI now, and most of them mean a chatbot. A chatbot is the least interesting thing AI can do for a website, and it is the one most likely to embarrass you. This post is about the checks I actually run when making a client site AI-ready, and where I draw the line.

Every client asks about AI now, and most of them mean a chatbot. A chatbot is the least interesting thing AI can do for a website, and it is the one most likely to embarrass you. This post is about the checks I actually run when making a client site AI-ready, and where I draw the line.

AI-ready means something specific: a site that AI tools can read, cite, and use correctly. That is mostly unglamorous work. Structured content, honest metadata, current sitemaps, and accessibility as the foundation. The chatbot comes last, if it comes at all.

What we cover

  • What AI-ready actually means
  • The checks I run
  • Accessibility is the foundation
  • Chatbots, done right, and the red lines
  • What I will not build

What AI-ready actually means

AI tools read your site the way a very fast, very literal visitor would. They parse the HTML, follow the links, read the metadata, and summarize what they find. If your content is a mess of divs with no headings, no structure, and no metadata, the summary will be a mess too.

AI-ready is not a feature you buy. It is the result of building a site properly: semantic HTML, real headings in a real order, descriptive page titles and meta descriptions, structured data where it fits, and a sitemap and robots file that are current. None of that is new. All of it matters more now, because there is a new kind of reader.

The checks I run

Here is the list, in the order I run it:

  1. Semantic structure. Real headings, real lists, real landmarks. If the page is a pile of divs, AI tools guess at the structure, and they guess wrong.
  2. Metadata that says what the page is. Titles and descriptions written for humans, which also happen to be what AI tools quote. A page with no description gets summarized from whatever text the tool finds first.
  3. Structured data where it fits. Recipes, events, FAQs, organizations. Schema that matches the content, never invented.
  4. Sitemap and robots, current. The sitemap lists what exists. The robots file says what is allowed. Both go stale quietly, so they get checked, not assumed.
  5. Content that answers the question. AI tools cite pages that answer clearly. A page that buries the answer under three paragraphs of throat-clearing does not get cited.

None of these checks require an AI tool. They require caring about the basics, which is the whole point.

Accessibility is the foundation

Here is the connection most people miss: if a human using a screen reader cannot use your site, an AI agent cannot use it either. Both of them read structure, not pixels. Both of them need headings, labels, and alt text. Both of them fail silently when the structure is missing.

That is why accessibility work is AI-readiness work. I ran my own AI accessibility audit on this site, and it scored 61 percent, with six real issues to fix. I wrote it up here: What an AI Accessibility Audit Found on My Own Site. The fixes were not exotic. They were the basics, done properly.

Handle WCAG at scaffold time, when it is cheap. Bolted on at the end, it is expensive, and it is usually incomplete.

Chatbots, done right, and the red lines

Sometimes a chatbot is the right call. When it is, the rules are simple. The bot answers only from your content, never from its own imagination. It says when it does not know. And it hands off to a human the moment the question gets complicated.

The red lines: a bot must never invent a price, a date, or a promise the organization cannot keep. It must never present a guess as a fact. And it must never be the only way to reach a human. A chatbot that traps a frustrated user is worse than no chatbot at all.

What I will not build

I will not build a chatbot as a strategy. A chatbot on top of a broken site is a faster way to give wrong answers. Fix the site first, then decide whether a bot adds anything.

I will not build AI features that collect data the organization does not need, or that make promises about accuracy nobody can keep. Grounded features, or none at all.

Final thoughts

AI-ready is mostly good web development with a new audience in mind. Structure your content, write honest metadata, keep your sitemap current, and make the site accessible. The chatbot is the last step, not the first.

The short version: semantic HTML, honest metadata, current sitemaps, and accessibility as the foundation. The chatbot comes last, and only if it earns its place.

/things-im-trying/2026-11/making-client-site-ai-ready-what-i-actually-check austin.amento.dev