We fetched 11 GoHighLevel agency sites. Page count did not predict which ones AI engines quote

Most competitive analysis in this market is assertion. This is not.

On 30 August 2026 we fetched eleven GoHighLevel agency and fulfilment websites and measured what was actually there. Every number below traces to a request we made. Nothing is estimated and no third-party index was used, which also means there are no backlink or authority columns, because those cannot be measured by fetching.

The sites are described by category rather than named. The findings are about a market, not about particular businesses, and several of them may become partners.

Method

For each domain we requested the homepage, robots.txt, llms.txt and the sitemap, expanding sitemap indexes to their child sitemaps to get true page counts rather than index entries. Schema was counted by extracting "@type" values from the live HTML. One domain sat behind Cloudflare and returned 403 to every request, so it is excluded from the counts rather than guessed at.

That leaves 11 measured competitor domains, ranging from 23 to 1,144 pages.

The results

What we measured	Result across 11 sites
Publish a working llms.txt	6
Return something at /llms.txt that is broken or empty	3
No llms.txt at all	2
Name any AI crawler in robots.txt	3
Use Service schema	3
Publish AggregateRating schema	4
Publish pricing anywhere on the site	2
Largest site in the sample	1,144 pages
Smallest site in the sample	23 pages

Finding 1: page count did not predict AI citation

The site AI engines quote most often for the fulfilment query in this market is the smallest one in the sample, at 23 pages. It is being cited ahead of a 1,144-page competitor and a 250-page one.

That is worth sitting with, because “publish more” is the reflexive answer to a visibility problem and this sample does not support it. Whatever is winning that query, volume is not the mechanism.

What the 23-page site does differently is visible and copyable. Its llms.txt is not a site map, it is a fact sheet: legal entity, contact, named plans with hourly rates, team size, onboarding time, refund terms. An engine asked what fulfilment costs can answer from that file without opening a page. It also publishes its pricing in a <title> tag, and its robots.txt carries a comment stating plainly that being cited by AI assistants is a distribution channel.

None of that is an accident, and none of it required 1,000 pages.

Finding 2: llms.txt adoption is mostly a mirage

Nine of eleven sites returned something at /llms.txt. Only six returned a file that would be usable by the engine requesting it.

What we got back	Count
A real, parseable file	6
An HTML soft-404 served with a 200 status	2
A 261-byte stub	1
Nothing	2

The soft-404s are the interesting failure. Both sites return a normal-looking HTML error page at /llms.txt with a 200 status code, which means an automated check that only looks at the status would record the file as present. Someone in both cases believes they have one.

Size varied by more than two orders of magnitude, from 261 bytes to 105KB. A 105KB file is not obviously better than a 5KB one; the largest in the sample reads as a dump, and the smallest usable one belongs to the site being quoted.

Finding 3: eight of eleven do not let AI crawlers in

Only three of eleven name any AI user-agent in robots.txt.

This is not the same as blocking them. A site with no rule for GPTBot is not refusing GPTBot, it simply has not considered it. But it does mean the market has largely not made a decision either way, at a point when Google’s own documentation describes crawling as the mechanism by which anything gets indexed at all.

The one site that has clearly decided is the 23-page one, and it is the one being quoted.

Finding 4: four of eleven publish rating schema, and that carries a risk

Four sites emit AggregateRating markup, one of them with three Review nodes alongside it.

We deliberately publish none, and this is worth explaining because the absence looks like a gap. Google’s review snippet guidance excludes ratings collected on another platform and republished, and self-serving reviews about the entity hosting the page are not eligible for Organization markup. Both conditions apply to genuine client reviews reproduced on your own site.

The reviews on this site are real, verbatim, and each links to the source it came from. They are just not marked up as ratings, because the markup would be the part that is not permitted, not the reviews.

What we take from it

Three things, in the order we would act on them.

A machine-readable fact sheet beats a machine-readable index. The most-cited site in the sample answers the commercial question inside llms.txt itself. A complete content map is a better file by most measures and still loses to a file that answers the question.

Check what your /llms.txt actually returns, with the status code, not by eye. Two sites in eleven are serving an error page and would pass a naive check.

Decide about AI crawlers explicitly. Eight of eleven have not, which makes it a cheap difference rather than a hard one.

Reproducing this

The method is four HTTP requests per domain and a regex over the response. If you want to check our numbers, or your own site, everything here was produced with curl and no paid tool. We are happy to be corrected by anyone who measures it and gets something different.