The Page ChatGPT Reads Most, Google Barely Shows

01The page nobody would pick

The obvious guess for which page on the agency site an assistant reads most is one of the guides Google shows most often. The guess is wrong. In the eleven days from 6 to 16 October, ChatGPT's live agent fetched 25 different pages a total of 170 times. Its favourite article, read 40 times, is the dated record of Google's algorithm updates. On Google that same article ranks 19th of 90 pages by impressions, with 68 in three months. Google ranks pages for searches. ChatGPT fetches pages for answers. They are not the same list.

The top of Google, unread

Now look at Google's own favourites. The page Google showed most, an industry page with 722 impressions in three months, was fetched by ChatGPT zero times. The second, a schema guide with 334, once. The guide about getting cited by ChatGPT, of all things, 259 impressions and zero fetches. Most of those impressions come from deep positions, page five and beyond, so they are not the pages an assistant retrieves when it searches. The algorithm record sits at an average position of 16.3, higher than any of them, but for a narrow set of searches.

02What the read page has that the shown pages lack

It answers a question about now

A live fetch happens when an assistant decides its stored knowledge is not enough, and the clearest case is a question about the present: is there a Google update rolling out, when did the last one end. The algorithm record answers exactly that, with dates, durations and a refresh after every confirmed update, most recently on 8 October. An assistant fetches a page live when the question is about now and the page is plainly current. That matches what I saw in August, when the same record drew far more live reads than any explainer, and it fits the dating advice in the queries no human typed.

Bing sees it more, but Bing is not the whole story

ChatGPT's search leans on Bing, and on Bing the algorithm record leads with 995 impressions in about six months, while Bing's crawler spent 49 of its 378 page fetches on it in the same eleven days. But the second page on Bing, a Perplexity guide with 752 impressions, was fetched by ChatGPT once. Visibility in the index ChatGPT searches is necessary. It is not sufficient. Being in the index gets you considered; being current gets you fetched.

The rest of ChatGPT's reading is about the company itself: the homepage 53 times and the careers page 36. Those are fetches for questions about who we are, not about search, and they show something Google's list never will, which is what people ask an assistant about you by name. The careers number matters more than it looks, and it gets its own entry on the 20th.

03Keep two scoreboards

Measure each list where it lives

Google's list lives in Search Console. ChatGPT's lives in your access log: successful fetches by its live agent, by page, by week, the counting rules from the referrers that never clicked. Put them side by side once a month. Pages that lead one list and are absent from the other tell you where to work. A page that Google shows and no assistant reads is a ranking, not a source.

Write one page that is current on purpose

Every site in a moving field should have at least one page whose job is to be right about the present. A dated record, a changelog, a status page with sources. Refresh it on a rule, state the refresh date at the top, and keep old entries as history. It is the opposite of the twelve articles on the same site that Google never showed once in three months; Bing's crawler fetched them 42 times in eleven days and ChatGPT once. Crawled, stored, unused, the gap from indexed, not quoted, now with a second page to compare it to.

Tomorrow: what happened to these counts the day the site started caching its pages at the edge.