What I learned rebuilding a website for AI search

Before we rebuilt the website, I asked ChatGPT, Claude, Gemini and Perplexity 161 questions that a buyer in our market might ask, ranging from broad questions about the technology to direct “who should I buy from” questions. The company was named in under 2% of the answers, and only when the question was tied to a specific location.

The starting point

The site ran on a heavy page builder with more than a dozen plugins. It looked good to people, but it said very little to a machine, since nothing clearly stated what the company was, what the products were or which page answered which question. The terminology the company uses to describe itself wasn’t explained anywhere in a way an AI model could quote.

That left the AI engines to guess, or to spend a lot of effort working out what we actually did. The result was vague summaries, which makes sense when you consider that an AI engine doesn’t want to mislead its users, because a confident wrong answer hurts its reputation. When we did show up, we were described as a simple product for smaller installations or low-scale testing, somewhere in the middle of the supplier lists, rather than as an established company with a unique solution.

What we did

We let the crawlers in. The first thing I did was make sure nothing stopped AI crawlers from reaching the site. I registered it in Bing Webmaster Tools and Google Search Console, which I’d recommend to anyone, because it confirms that your pages are actually being indexed and gives you the numbers to follow up on. Next to the sitemap we added an llms.txt, a short plain-language briefing that tells language models who the company is and which page answers which question.

We labelled every page from its own content. Every page now carries structured data, a machine-readable label that says “this is a company” or “this is a question and its answer”. Since the label is generated from the visible text rather than typed in separately, the page and its label can never say different things, and every FAQ on the site is tagged as question-and-answer data automatically.

We wrote answers instead of more pages. We clearly defined the company and the terminology we used together with a handful of resource pages, each answering one question and putting the answer first. Where different pages competed for the same question, we merged them into one.

We built it instead of buying it. Early on I decided not to add a plugin unless it was strictly necessary. An SEO plugin gives every website the same generic output, while AI search rewards precise and consistent signals that match your content. With Claude Code, building it ourselves took afternoons rather than months, and there’s nothing extra to pay for or renew.

What changed

After the rebuild I ran the same 161 questions again. The biggest change was in how we were described. Instead of a small product for testing, the AI answers now portrayed the company as a leader in its category, and sometimes chose it as the single best match. When buyers used our own terms we were named first, and ChatGPT described the company as the strongest match. Overall AI visibility had roughly doubled, which from a base that low is a start rather than a win, and Google impressions were up 41%.

Notice the words “our own terms”. The doubling didn’t come from suddenly appearing in new kinds of questions. It came from showing up where we should have been from the start, in answers to questions that describe exactly what the company does, but where we had been left out completely because the website never stated it clearly enough for a machine to read. Once the AI engines could understand who we are, what we do and what we offer, they could connect us to the right terminology. To reach beyond that and show up in a wider range of answers, you need to be cited by others, which is the last step in my notes on AI search.

How long it takes

The rebuild itself doesn’t have to take long if you know what to do, but the results take longer.

Search engines and AI crawlers first need to read the site again, which can take weeks for a site they don’t visit often. After that, the AI engines look for the bigger picture. What do others say about the company? Is it referenced or cited anywhere? Do the same facts show up in more than one place? Those signals build up slowly through ordinary content and SEO work, and it usually takes months before they pay off.

A solid foundation is what makes it possible for a company to be found and understood correctly in the first place, and it makes everything you do afterwards count for more, because every new article or mention is read in the right context. Without it, everything you publish builds on a picture of your company that AI has had to guess.

What didn’t change

On broad questions like “who are the leading vendors”, we were still missing. AI answers still treated the company’s own statements as claims and described it as newer and less proven than the established names, and that’s one of the most important lessons from the whole project.

The technical groundwork didn’t make us the answer by itself, but it made us readable, and being trusted takes something no website can do on its own, which is other people talking about you.

What I’d tell someone starting out

Start by measuring, because without a fixed set of questions you can’t tell whether anything you do is working. Then make sure your website has the foundations that make it readable for AI. Most older websites don’t, and that’s an easy way to be left out of the answers, or to be described wrongly when you are in them. And put the money you would have spent on tools into something worth citing.

Step 01 in AI search notes: Measure where you stand before you change anything