Do pages you never request indexing for get indexed on their own? 2 of 9 went in on single organic Googlebot fetches in 3 days, 0 of the 13 pages Googlebot never fetched moved, and all 10 clicked pages were re-fetched within 72 hours

What was pre-registered, and what was read

On 2026-08-26 we published the 27-hour reading of a Request-indexing test: 10 of 11 clicked URLs indexed, 0 of 9 unclicked. The decision rule had been written down before the data: read again on 2026-08-28, count M of 11 clicked and N of 9 unclicked, and treat M of 3 or more with N of 0 as "the button works". The rule also said what an N of 1 or more would mean: the comparison weakens, because pages were going in without the button. Today N is 2, so this post is the weakened reading, stated as counts, not a new claim about the button.

Everything below was read between 16:38 and 16:43 UTC on 2026-08-28 by two instruments: Search Console's URL-Inspection API for index state and last-crawl time, and our own crawler ledger for every request whose user agent names Googlebot (probe rows and spoofed rows excluded, as in every ledger read we publish). The 14-URL denominator, the 11 clicked URLs and the 9 unclicked ones are read from files written before today, not chosen this afternoon. The cut script and artifact are in our repository, `content/news/request_indexing_followup_cut.mjs` and `request-indexing-followup-cut-2026-08-28.json`, with 14 self-checks that assert every number in this post, including that the analyst's "Googlebot doubled" figure reproduces from the raw rows.

The read, as counts

The two that went in without a click

Both are posts from the week before the test, both read "Discovered - currently not indexed" on the 08-25 sweep, and both had 0 Googlebot requests on the ledger from 2026-08-18 to the 08-26 read. Each then drew exactly one Googlebot request on 2026-08-27, from a US address, and Search Console now reports that request as the last crawl, to the second. There was no click: the scheduled task that pressed the button on 08-25 has no directory on the machine it ran from, there is no sweep file after 08-25, and the ledger shows no click cadence on either day.

One fetch each, then indexed. That is the same shape as the clicked cohort: on this site, the first Googlebot content fetch of an unindexed page has been the fetch it went in on, 12 times out of 12 (10 clicked, 2 organic). The difference the button made was when the fetch happened, not whether a fetch was enough.

What Googlebot did on the day they went in

Googlebot's requests to this site, by UTC day, from the ledger. 2026-08-28 is the partial day to 16:22 UTC. "Paths" counts distinct URLs including robots.txt and sitemap.xml.

Our analyst's standing 24-hour window (09:37 UTC to 09:37 UTC) read 53 rows over 32 paths for 08-27 to 08-28, against 27 over 15 the day before, and the raw rows reproduce that exactly. Where the doubled day went: 15 robots.txt requests, 3 sitemap requests, 10 requests to 9 already-indexed /news posts, 8 organic re-fetches of clicked pages, 7 to /guide pages, 5 to /compare pages, the 2 fetches that indexed the untreated pages, and 1 each to the homepage, a /for page and a /data page. Requests to the 14 pages that are not indexed: 0 of 53.

The ten clicked pages, three days on

All 10 are still indexed. Every one of them has since drawn at least one organic Googlebot request, between 27 hours and 69 hours after its click fetch, and one of them two. Search Console's last-crawl field did not move on any of them: it still reports the 08-25 click second. We do not know whether the inspection field lags, or whether these later requests are a kind of fetch Google does not count as the crawl of record. It is a difference between the two instruments, and we report both.

What one reading cannot separate

Two stories fit these numbers, and the data does not pick one. In the first, the clicks on 08-25 put 10 pages into Google's crawl schedule, Googlebot returned to all 10 within three days, and while it was on the site it also picked up two neighbouring posts that link to and from them. In the second, Googlebot had a bigger day for reasons of its own, 49 requests over 29 paths against a daily median of 26 for the prior week, and the two untreated pages were simply the ones its schedule reached; the re-fetches of the clicked pages were part of the same larger day. The control for both stories is the same: the 13 pages Googlebot has not requested at all. Under either story, they did not move, and a page that has never been fetched cannot be indexed.

The fourteen Googlebot has never fetched

Thirteen of these have 0 Googlebot requests since the ledger started watching them on 2026-08-18; the fourteenth is the clicked URL that drew no fetch and never went in. Every one of the fourteen was fetched in the same seven days by at least three of OpenAI's, Perplexity's, Anthropic's, Amazon's, Microsoft's and Yandex's crawlers. "Discovered" and "unknown to Google" are reported as read at 16:38 UTC; the 08-26 post showed those two labels swapping between API reads four minutes apart, so we count them together as not indexed and do not narrate the label.

The one variable we changed today, and its read date

The button is recorded as something that worked once. The task that pressed it ran from a scheduler that no longer exists on the machine, and we are not going to plan a publishing pipeline around a button with a daily quota that a person or a script has to press. So today, before this post, our engineer shipped the internal-linking change that had been prepared as one variable: the /news index is now grouped by month with dated sections, and every post carries `rel="next"` and `rel="prev"` links to its neighbours. We checked the neighbour links on the live 08-26 post before writing this. The read is pre-registered for 2026-09-11 on the same 14 URLs named above, and the rule is written: if at least 7 of the 14 are indexed by then, the hub stays; if 0 of the 14 are, internal linking is not what is holding these pages out, and no further link-shape work ships on this stall. Anything between is reported as inconclusive. The instrument is one command, `ops/verify/hub_read.mjs`, and its verdict line on 09-11 will be quoted as printed.

Why this matters if you care about AI visibility

• Being fetched once is the whole game for a small site's Google indexation. Twelve of twelve pages that Googlebot fetched went in on that fetch, clicked or not; fourteen pages it has not fetched are out, however they are labelled. If you track "Discovered - currently not indexed", track Googlebot requests in your own logs beside it; the label tells you less than the absence of a fetch does.
• The AI engines are not waiting for Google. Every one of the 14 unindexed pages was read by OpenAI's, Perplexity's or Anthropic's crawler inside the week, several by ChatGPT-User, which fires only when a ChatGPT user is actually shown the page. Whether an answer engine can cite you and whether Google has indexed you are separate questions with separate evidence, and the second does not gate the first.
• Do not read a lever off one good day. Two unrequested pages going in on the day Googlebot's requests doubled looks like a signal and might be noise; we have said which reading would settle it (09-11, 14 named URLs, a written bar) rather than declaring one. If a vendor tells you their change got your pages indexed, ask what the untouched pages did the same week.

If the question you actually have is whether the engines currently cite your business, before the question of which index it sits in, the free 60-second check asks four of them the way a customer would and shows you the sources they used.

Limits

• Nine unclicked and 11 clicked pages on one site. Two organic index events is a count of two; it separates "never happens" from "happens", nothing finer.
• Search Engine Roundtable reported on the morning of 2026-08-28 that site owners were seeing indexing or serving delays, with Google's Search Status Dashboard showing no incident; the dashboard's incident feed, read by us at 16:30 UTC, lists nothing after the August 2026 spam update, whose rollout completed 2026-08-21. Our two index events are dated 08-27, before those reports, and we saw no change on the 14 either way. We note it because a reader comparing their own 08-28 numbers should know the day was contested.
• The ledger trusts the user agent and does not reverse-DNS addresses. For the two organic index events the last-crawl time reported by Search Console agrees with the ledger row to within one second, which is the strongest confirmation we can get without a log-level reverse lookup.
• The three-day organic re-fetches of the clicked pages are ledger rows that Search Console does not reflect in its last-crawl field. One of the two instruments is describing something the other is not; we do not know which, and we have not asserted that the re-fetches changed anything.
• The 09-11 read tests internal linking as one variable on a live site where posts keep shipping daily. Six of the 14 are under four days old; some may go in for reasons that have nothing to do with the hub, which is why the bar is 7 of 14 and not 1.