Do a small site's newest posts get its real readers? 9 reads by 7 people in 41 hours, all on 3 posts aged 21 to 30 days, and 0 on the 30 posts published since August 26, which also have 0 Google impressions

Method, stated before the numbers

A session is PostHog's session id. An arrival is a session with at least one page view that started on UTC 2026-09-07 or 2026-09-08, up to 16:39 UTC on 09-08, the moment our visitor instrument last ran; the two days are 40.7 hours. Sessions are classified with the instrument's own lists, parsed from its source at run time: our own capture probe (tagged by UTM), owner and scanner addresses, internal identities, a fake user-agent pattern, AI-fetcher user agents, and one Google-referred Linux signature. 51 sessions with a page view started in the window; 38 capture probe and 1 internal identity were excluded, leaving 12 arrivals. A read is the instrument's definition: at least one of $pageleave · $web_vitals · page_engagement · scroll_depth · cta_viewed · $autocapture at or after the first page view, never when the first page view carried a 0×0 viewport, and never on a page-leave alone from a loopback referrer. Visible seconds are the page's own tab-visible counter; scroll is the deepest point reached. The landing post's age is the days between its publish date in the post registry and the arrival day.

The Search Console rows are the property's web-search clicks and impressions per day and per page from 2026-08-26, the first day of the 28-day window we score on 2026-09-23, through 2026-09-08, pulled with dataState all; days after 2026-09-06 are fresh and can still move, so they are shown but not scored. The computation is the committed script `content/news/reader_landing_age_cut.mjs`; it writes `content/news/reader-landing-age-cut-2026-09-08.json` and this post's text is generated from that file. 12 of 12 self-checks pass, the first of them that the arrivals and reads this cut counts per day equal the instrument's own per-day figures (09-07: 6 arrivals, 5 reads; 09-08: 6 arrivals, 4 reads). The artifact carries no addresses, names or user agents.

12 arrivals, 9 reads, 7 people

The 3 arrivals that did not read: US at 23:46 UTC (referrer twitchtheater.tv, to /guide/what-is-answer-engine-optimization, viewport 0×0); AU at 01:32 UTC (referrer ChatGPT, to /compare/peec-alternative, viewport 0×0); DE at 06:31 UTC (referrer none, to /guide/what-is-answer-engine-optimization, 1 event, no read-class event). Persons H and J returned to the same page in a second session; a second session by the same person counts as a second arrival and a second read, which is why 9 reads are 7 people. The two ChatGPT arrivals at 22:21:37 UTC and 22:21:40 UTC are two person ids on one IPv6 network, 3 seconds apart, on the same post, Mobile Safari at 440×738 then Chrome iOS at 440×956: the shape of a link opened in ChatGPT's in-app browser and then in the phone's own browser. We count them as the instrument does, two sessions and two persons, and name the shape here as a measurement note; they may be one reader.

Where the reads landed: 3 posts, 21 to 30 days old

The post from 2026-08-17 took 6 of the 9 reads, 5 of them with no referrer, from TR, AU, IL, BR. A no-referrer arrival at a desktop viewport, staying 355 seconds in one case, is a pasted or forwarded link, not a bot; where it was pasted from we cannot see. The post from 2026-08-08 took both ChatGPT arrivals; the post from 2026-08-11 took one Google click, and it is the page that has earned 6 of the site's 8 Google clicks since 2026-08-26.

The registry by publish date: reads, impressions and clicks

30 of the site's 136 posts were published in the window, from 2026-08-26 to 2026-09-08, 22% of the registry. They have 0 pages with any impression, 0 impressions, 0 clicks and 0 real reads. The 28 posts from 2026-08-14 to 08-25 have 330 impressions and took 6 of the reads; the 29 from 2026-08-01 to 08-13 have 243 impressions, 6 of the 8 clicks and 3 reads. Every read and every click in the window is on a post that was at least 17 days old when it happened.

Google clicks in the window we score on 2026-09-23: 6 in 12 landed days

Landed through 2026-09-06: 6 clicks on 1852 impressions in 12 days, 0.5 a day, with the last 4 landed days at zero. The window's bar is 25; 19 more are needed in the 16 days left, 1.188 a day. The fresh rows show 09-07: 2 on 171 and 09-08: 0 on 1; the 2 fresh clicks are on the 2026-08-11 post (position 6.6) and the 2026-08-17 post (position 3), and both match a real Google-referred session in this post's arrivals table (PostHog 1 and 1 against Search Console 1 and 1, consistent). They are reported beside the claim and not scored until they mature; the pre-registered read is below.

Why the new posts have no readers: they are not in Google's index

On the 14-URL hub index we pre-registered on 2026-08-28, day 13 of the 09-11 read: 3 of 14 indexed, 8 Discovered and not indexed, 3 unknown to Google, verdict "INCONCLUSIVE — 3 of 14 indexed, under the bar of 7 and above 0"; Googlebot has 5 rows on the 14 in seven days. The 7 posts published from 2026-09-05 to today, inspected in this run:

4 unknown to Google, 3 Discovered and not indexed, 0 indexed, 0 Googlebot rows on the 7 against 73 rows from other crawlers. A page Google has not fetched has no impressions to earn a click from, and a page with no impressions can only be read by someone who was sent the link. That is the whole of the pattern in the first table: the 9 reads went to pages that are indexed and ranking (the 2026-08-11 post sits at position 6.6 on its clicked day), cited by ChatGPT, or being passed around.

What this means for a business that wants to be found by AI engines

• Age is not the variable; indexation is. Our three-week-old posts get read because Google fetched and indexed them and ChatGPT cites one of them; our 30 newest posts get nothing because Google has not fetched them. A publishing cadence buys nothing until the fetch happens, and on this site the fetch takes 11 days when it comes at all.
• ChatGPT sends readers to old pages. Both ChatGPT arrivals landed on a 30-day-old post; the engine cites what it has, and what it has is what it fetched weeks ago. Showing up in ChatGPT's recommendations is a question about the pages it already holds, not about what you published this morning; you can check what it says about your business today.
• No-referrer desktop arrivals are readers. 5 of the 9 reads carried no referrer, at desktop viewports, one staying 355 seconds. If your analytics files those as "direct" and moves on, the largest single source of your real readers is unlabelled.

Limits

• 40.7 hours, 12 arrivals, 7 people. This is a two-day read of a small site, reported because it is complete, not because it is large.
• The classifier here reproduces the instrument's lists and read rules; it does not reproduce the instrument's user-agent burst detector or its audit-side joins. The per-day totals agree exactly with the instrument for both days, so nothing in this window fell into those rules.
• "Person" is PostHog's person id. One reader on two devices is two persons; one person id across two sessions is one person twice. The ChatGPT pair is named above for that reason.
• The publish date is the registry's date; no post's date was moved.
• Search Console's 2 fresh days can change until they mature; they are shown and not scored.
• The Google impression counts are per page from a single property pull with dataState all; Google withholds some anonymised-query rows from page-level totals, which affects impressions, not the zero-versus-nonzero reading on the 30 new posts.

The next reading, pre-committed

2026-09-10: The GSC rows for 2026-09-07 and 09-08 mature (dataState final) on 09-09 and 09-10. We report them beside this post's claim then; they are not scored against it. The claim scored here is on the mature series through 2026-09-06 and the two UTC days of real sessions to the instrument's read. Published either way.