I Audited My Own Site For AI Search. Seven Pages Were Indexed.

It does not look like bad rankings. It looks like nothing.

Here is what Search Console showed for a three month window:

Indexed pages	7
Impressions	105
Clicks	2
Average position	32
Distinct queries	7

Position 32 is page three or four. Nobody goes there, so the click through rate is not a meaningful number. Two clicks is not a conversion problem, it is an absence.

The queries were worse than the volume. Seven in three months, and not one of them commercial. They read like fuzzy matches against the domain name rather than searches for what the business does. Nobody had typed anything resembling funnel consultant, webinar funnel, or marketing automation and been shown the site.

Zero appearances for the work. All appearances for the name, and even those by accident.

Because there were only seven pages, as far as a crawler was concerned.

The site had a projects page listing twenty two client projects. Each one had a "View Case Study" button. Click it and a modal opens with the detail: what was built, what it produced, the screenshots.

Fetch that page the way a crawler does and something different happens. The navigation links come back with full destinations. The footer links come back. The "View Case Study" buttons come back as plain text with nowhere to go, because the detail is rendered client side and never exists in the HTML a crawler receives.

Twenty two pieces of proof, collapsed into one URL.

That is not a content problem or a quality problem. The content was there and it was good. It was structurally unreachable, and no amount of writing more of it would have helped.

The same failure sits under most modern site builds. It looks fine in a browser, because a browser executes JavaScript. It looks empty to anything that does not.

They overlap on the technical side and they reward different things, which is the part worth understanding.

Search engines rank pages. Ranking rewards accumulated authority: links, age, domain strength. That takes years and there is no shortcut.

Answer engines assemble answers. A model composing a response needs a passage it can lift: a direct statement, attached to a clear question, attributable to a named source, and specific enough to be worth quoting.

A page can rank well and be useless to an answer engine, if it buries its answer under three paragraphs of positioning. A page can also be cited from a domain with almost no authority, if it states something specific and checkable in a form that can be extracted.

That asymmetry is the opportunity, and it has a shelf life. Extraction rewards structure and specificity, which takes weeks. Ranking rewards authority, which takes years. Right now a well structured consultant site can appear alongside sources with a thousand times its traffic. That gap will close as larger players restructure. It has not closed yet.

Four things, in this order, because each one makes the next one pointless if it fails.

Can a crawler read the site at all? Fetched as a crawler fetches it, without executing JavaScript. This is where the modal problem surfaced, and it is the failure that silently invalidates everything downstream.

Does anything specific exist to point at? Ten services lived as anchor links on the homepage rather than as pages. There was nothing for a model to cite about any individual service, because no individual service had a URL.

Are the claims quotable? "Significant results" is not citable. "$2,246,155 across twelve months, 2,108 buyers, $1,060 average cart value" is. Most of the site was in the first category despite the underlying numbers existing.

Is the entity consistent? How the business is described on the site, on LinkedIn, and anywhere else a model might read. Inconsistency prevents a stable entity forming, so three mentions get treated as three unrelated things rather than as one business.

Routing first. Every case study becomes a real URL, server rendered, that a crawler can reach without executing anything. This is the only change that unblocks the rest.

Then the pages that did not exist. Nine service pages instead of ten anchor links. Twenty two case study pages instead of one grid. Tag pages. This blog. Seven pages becomes roughly fifty one.

Then specificity everywhere. Prices stated with scope attached. Results stated as figures with periods and denominators. Where a number came from a client rather than from an account I could see, the page says so.

That last one is uncomfortable and it is the part most sites skip. One case study now states that a 20 to 25% improvement was reported by the client rather than measured by me. Another states that a six figure total has no screenshot behind it because the platform account was closed before anybody captured one.

Those admissions cost nothing with a serious reader and they are exactly what makes the rest of the numbers believable.

The site was not neglected. It was tended in the wrong places.

The design was fine. The copy was fine. What was missing was the thing I check on every client project and had never run on my own: can this actually be read, and is there anything specific here to cite.

There is a version of this in almost every consultancy. The site gets redesigned every couple of years, the positioning gets updated, and nobody fetches it the way a machine does. The work is good, the results are real, and none of it is reachable.

I was proposing a redesign when I ran the numbers. A redesign would have produced seven very attractive pages.

Three things, and the third one gives you a number.

1. Fetch your site without JavaScript.

View source rather than inspect element, or use a plain fetch. If your content is not in the raw HTML, no crawler and no AI system can see it.

2. Count your indexed pages in Search Console.

Indexing, then Pages. Compare that to the number of distinct things your business does. If it is a small number, more content will not fix it until routing does.

3. Ask ChatGPT, Perplexity or Claude the question your best client would ask before hiring you.

Whatever comes back is your baseline. Record it, because you cannot show improvement against a number you never took.