AI Bots Crawled 99% of the Site, but Only 2 Pages Were Cited
Published
This article was first published in Korean by Intrix Lab on September 23, 2026.
Even when AI bots read almost an entire site, very few pages get cited. As of 2026-09-14, AI bots had crawled 167 of the 169 pages in Intrix's sitemap (99%) over the previous 90 days, but only 2 pages were cited in our CT study: the homepage (8 times) and the services page (3 times). Even the most frequently crawled pages received 0 citations: the free audit page (713 crawls), about page (658), and insights listing (614). After Profound released a tool to find "content that is indexed but not cited," we measured that share on our own site first.
How did we measure this?
We compared three sets: pages in the sitemap, pages AI bots actually visited over the previous 90 days, and pages cited in AI answers in our own citation study. A page AI has not read cannot be cited, so crawling sets an upper limit on citations.
What were the numbers?
| Set | Number of pages |
|---|---|
| Sitemap | 169 |
| Crawled by AI bots over 90 days | 167 (99%) |
| Never visited by AI bots | 2 |
| Cited in the CT study | 2 (homepage 8 times, services 3 times) |
| Crawled but received 0 citations | 166 |
The most frequently crawled pages with no citations were the free audit page at 713 crawls, about at 658, insights listing at 614, platform at 583, projects at 526, and lab at 463. We saw no relationship between crawl frequency and citations.
Why were frequently crawled pages not cited?
These data cannot establish the cause, but two things are clear. First, crawling is a condition for entering the pool of candidates, not for being selected for an answer. Second, the two cited pages answered what the company does and what it sells in a single paragraph, while the frequently crawled but uncited pages were listings or entry pages for tools. The difference appears to be whether a page has a paragraph that answers a question. Our next measurement will compare the structures of cited and uncited pages item by item.
What do we do differently as a result?
We do not treat "AI bots visit frequently" as a performance result. Bot logs measure access, while citations measure selection, so they need to be assessed separately. Defining the question each page should answer, then providing a paragraph that starts with the conclusion, numbers, and an as-of date, are conditions for moving from a citation candidate into an answer. Intrix conducts this comparison on its own site first and publishes the results here in the lab.
Limitations
The citation findings are limited to the questions in Intrix's own study, which concerned GEO companies and agencies. We did not check citations across every possible AI question. Gemini does not provide a candidate list, so we cannot count pages it "read but did not select." We used whether a page had been crawled as a proxy for indexing, rather than Search Console. These results are not evidence that any particular citation is guaranteed.
Frequently asked questions (FAQ)
Q. Can we check the same thing if our site has bot logs?
Bot logs show only half the picture. Finding out which pages actually get cited requires repeated studies of AI answers. Intrix's citation study covers that part.
Q. Are the 2 pages that were not crawled a problem?
They should be checked first. They are likely to have no internal links or to appear only in the sitemap.
Q. Do only 2 cited pages mean our GEO is performing poorly?
This finding applies to the question set in our own study. The homepage and services page were cited because those are the two pages that answer questions about the company. First, check whether the study included the questions the remaining pages are meant to answer.