If an AI Can't Read Your Site, You're Not on the Shortlist

Here is how vendor selection starts now. Someone on a leadership team opens ChatGPT, or Claude, or Gemini, or Copilot, and types a question: "Who does IT compliance advisory for tribal governments in the Southwest? Give me three firms and tell me why." The assistant answers. A shortlist now exists. Your website has not been visited.

It may never be.

The website visit used to be the first impression. Now it is often the second or third, and it only happens if you survived the first one, which took place inside an answer engine you never see and were never told about.

The shortlist happens before the click

For twenty years the playbook was search. Rank on Google, win the click, let the site do the convincing. Buyers still search. But more and more, they delegate the research itself. They ask an assistant to compare and recommend, and they trust the answer enough to act on it. The assistant reads whatever it can reach and answers with whatever it found.

If it found nothing about you, you are not in the answer. Not ranked low. Not penalized. Absent. And a buyer who never sees your name has no reason to go looking for it.

Why the assistant cannot read your site

When a business is invisible to these tools, the failure is rarely the content. Plenty of firms have good case studies and clear service pages. The failure is the plumbing underneath. Four problems do most of the damage:

The content only exists in JavaScript. Many modern site platforms assemble the page inside the visitor's browser. A human sees everything. Most AI crawlers do not execute JavaScript, so they request your page and receive a nearly empty shell with a loading script inside. Every case study you wrote, every service you describe, unreadable.

robots.txt blocks the crawlers at the door. This is the small file that tells crawlers what they are allowed to read. Many sites block AI crawlers outright, usually a default someone accepted years ago and never revisited. The block does not protect anything valuable. It just guarantees the assistant learns about you from somewhere else, or not at all.

No structured data. Assistants lean on structured markup to understand what a page actually is. This is a business, here is what it does, here is where it operates. This is an article, here is who wrote it and when. Without that markup, the machine is guessing at your identity from loose prose.

Metadata that arrives too late. Titles, descriptions, and link previews injected at runtime by the same JavaScript the crawler never runs. The page describes itself well to browsers and says nothing to machines.

None of this shows up in a design review. The site looks fine. It demos fine. It is illegible to the exact readers your next client is consulting first.

We rebuilt our own front door before offering to rebuild yours

I do not recommend work I have not done on my own house.

When we rebuilt desotollc.com, we treated AI legibility as a build requirement. We opened our robots.txt to AI crawlers, and I will admit that meant reversing blocks I had put there myself. The default felt safe at the time. It was costing us visibility.

We prerendered every page and every case study as static HTML, so a crawler that never touches JavaScript still receives the full text of who we are and what we have done. We added LocalBusiness and Article structured data so machines are told, not left to infer, what each page is. We made the metadata part of the page itself instead of something a script bolts on after the fact.

And we put Frank on every page. Frank is a Claude-powered site advisor grounded in our actual services, so the human who does arrive can ask questions the way buyers now ask them, and get answers drawn from what we really do rather than from a chatbot's imagination.

That is our own front door, rebuilt for how buyers actually approach it now.

What you can control, and what nobody can promise

There is a wave of vendors forming around this, and some of them will promise too much. This field is early. The standards are moving. Nobody can guarantee your firm a spot in an AI-generated answer, and anyone who guarantees placement is selling you something else. The assistants decide what they cite, and how they decide keeps changing.

What you can control is simpler and more durable. Whether an assistant that comes looking can read anything at all. Whether what it reads is structured, so the machine knows what it is looking at. Whether what it finds is accurate at the source, so the answer it gives about you is the answer you would give.

We call this practice AEO, answer-engine optimization, and we offer it as part of Architecture & Build Oversight: sites built so assistants can find you, read you, and cite you. It sits naturally inside build oversight because it is not a marketing layer you sprinkle on afterward. It is a set of decisions made when the site is architected, and it is cheap to get right at that stage and annoying to retrofit later.

The uncomfortable question is the place to start. When an assistant is asked about firms like yours, does it know you exist? If you are not sure, that is worth finding out before your next prospect does.

If you want to see what an AI assistant sees when it reads your site today, that is a short conversation. Reach out.