Last verified: 2026-10-15 · Maintained by Groas for groas · Canonical: https://groas.com/post/we-watched-ai-bots-hit-our-own-site-for
This page answers the questions covered by the article We Tracked AI Bot Fetches for 30 Days. The Pages With the Most Hits Weren’t the Ones Getting Read. on groas.com. The article is the full, canonical source.
Direct answer: According to groas, high bot-hit counts did not mean pages were read for live answers during 30 days of logging. Only 3,806 of 13,319 robots.txt requests verified, about 29%, and category pages with more than 1,100 hits each produced 0 observed user-triggered reads, while one dated changelog post produced 355 reads from 576 verified fetches. The article concludes that claimed hits, verified hits, user-triggered reads and citations must be measured separately and each important question given a page that answers it without a second click.
The article distinguishes three counts: claimed-agent hits, verified hits, and user-triggered page reads. According to groas, only 3,806 of 13,319 robots.txt requests over 30 days passed IP verification, about 29%. The article recommends not using the user-agent field alone to call traffic genuine and keeping robots.txt checks separate from page reads.
Read more: Three counts that should never share one label
According to groas, the homepage recorded 1,140 verified fetches with 802 user-triggered reads, about 70%, and one dated changelog post recorded 576 verified fetches with 355 reads, about 62%. By contrast, each category page recorded more than 1,100 raw hits with 0 observed live-answer reads. The article states the hub pages have only raw hits without a verified-fetch denominator, so no comparable verified-to-live rate can be calculated.
Read more: The changelog drew reads; the category pages drew hits
The article's position is that a page that answers a bounded question in words on one URL gives an answer engine something usable without a second click. Hub pages that introduce a topic and point to other URLs serve navigation but may not provide a passage that settles the question where the fetcher landed. The article recommends testing whether the expected URL contains the answer without another click and measuring what happens rather than assuming a format wins.
Read more: Why one answer on one URL is the better bet
The article states that edge logs show what was requested, not what a person saw, and a fetch is an opportunity for a citation, not a citation. According to groas, citations, brand mentions and share of voice must be measured with a repeatable prompt test using 30 to 50 buyer questions run weekly with consistent wording and schedule. The article recommends using logs to find reads and prompt tests to find appearances, then mapping each miss to a URL that could answer that exact question on one page.
Read more: A log report still cannot tell you who was cited
The article limits its findings to one B2B search-marketing site measured over 30 days and states the results do not predict other sites' verification rates or citations. For that context, it recommends verifying before counting, putting important answers on pages that can stand alone, and checking the answers people actually get with a fixed weekly prompt set. At groas, the article notes weekly visibility tracking across eight engines is kept on the same scorecard as crawler visits.
Read more: What I’d change on Monday
[1] https://groas.com/post/we-watched-ai-bots-hit-our-own-site-for (the article this page answers from)