The Pages read tab of the AI card. Your paths, ranked by how many crawler views each received.
The AI card with the "Pages read" tab selected, showing a ranked list of paths with crawler view counts — a documentation or blog site, so the paths are recognisable and the distribution is uneven. Light theme, desktop.
Reading it
It shows the top 12 paths. Only crawler views count here — human views are in Top pages on the main dashboard.
The comparison worth making is between the two lists. Set the same date range, look at both, and the interesting rows are the ones that appear in one and not the other:
A page high in Pages read and low in Top pages. The assistants are reading something people are not finding. If that page is a good answer to a common question, it is being consumed through an AI product rather than through your site — you are supplying the answer and someone else is delivering it.
A page high in Top pages and absent from Pages read. People find it and the
assistants do not. Often it is behind a login, blocked in robots.txt, or
client-rendered in a way crawlers cannot read.
A worked example
A documentation site, last 30 days.
Top pages (human):
| Path | Views |
/ | 4,182 |
/docs/getting-started | 2,914 |
/pricing | 1,663 |
/docs/api | 897 |
/blog/why-we-built-this | 412 |
Pages read (crawlers):
| Path | Crawler views |
/docs/api | 1,046 |
/docs/getting-started | 738 |
/llms.txt | 291 |
/blog/why-we-built-this | 188 |
/ | 96 |
Three things fall out of that:
/docs/api is the assistants' favourite page and only the fourth most-read by
people. Reference material is what assistants pull in to answer questions.
That page is doing more work than its human traffic suggests.
/pricing is not in the crawler list at all. Assistants fetch pages that
answer questions; a pricing page usually answers one only for people already
deciding.
/llms.txt is third. If you publish one, this is where you find out whether
anything is actually reading it.
Narrowing to one kind of crawler
Click a row in the Crawler types tab first — say, AI assistant — and the Pages read list narrows to that type. "Which pages get pulled in to answer live questions" and "which pages are being collected for training" are different lists on most sites, and they suggest different actions.
Filtering by path pattern
The page filter accepts * as a wildcard, so you can ask about a whole section:
/docs/*— everything under docs/blog/*— everything under blog
That is the way to answer "is the AI traffic concentrated in my documentation or spread across the site", which the top-12 list alone cannot tell you.
The caveat that applies to every number here
This list shows the crawler traffic that reached us. Crawlers that fetch raw HTML without running JavaScript are invisible to a browser script, so the counts are a floor rather than a total.
The ranking is more trustworthy than the counts, because the undercount applies fairly evenly across your pages. Which page the assistants prefer is a question this list answers well. How many times they read it is a question it answers low.
See How reliable is it.