Our page-one CTR went from 0.48% to 1.37% in a week because one URL crossed position 10. The other 44 page-one URLs read 0.53%, and that bucket has sat between 0.26% and 0.68% for six weekly windows

Method, stated before the numbers

Search Console's Search Analytics API, property askedabout.com, search type web, dataState all, dimension page, for six consecutive 28-day windows whose last day is a Sunday: 2026-07-26, 08-02, 08-09, 08-16, 08-23 and 08-30. Every day in every window was complete at read time (2026-09-01). "Page one" means a URL whose window-average position is 10.0 or better; the "10 to 11 band" is the same test one position wider. CTR is clicks divided by impressions on the whole bucket, not an average of per-URL rates. The "one URL" is our post on pulling generative-AI data from the Search Console API; it is removed in the second series because it earned 7 of the newest window's 11 page-one clicks on its own, and a bucket whose CTR is decided by one row is not telling you about the bucket. This is week three of a series: the 2026-08-22 read (51 URLs, 881 impressions, 4 clicks, 0.45%) and the 2026-08-30 Googlebot re-fetch join. The script and both input files are committed beside the post; no figure below was typed from memory.

Six windows of the page-one bucket

The bucket grew from 18 URLs and 160 impressions to 45 URLs and 804 impressions over the six windows, and its click count read 1, 1, 4, 4, 4, 11. The last step, 4 to 11, is the whole story of the CTR jump, and it is one row. The Search Console API post sat at position 11 in the window ending 2026-08-16 (26 impressions, 5 clicks) and 10.2 in the window ending 2026-08-23 (30, 5): outside the page-one test both times, already earning clicks. In the newest window it averaged 9.2 with 55 impressions and 7 clicks, so it entered the bucket and brought its clicks with it. Nothing about the other 44 page-one URLs changed: 4 clicks became 4 clicks on 749 impressions.

The band just outside page one

The 10 to 11 band held 6 clicks in each of the windows ending 08-16 and 08-23; 5 and 5 of those were the Search Console API post. Once it crossed, the band read 3 URLs, 9 impressions, 0 clicks: google-platform-properties-instagram-tiktok-x-youtube-ai-citations (10.2, 6 impressions); do-ai-engines-cite-homepages-92-percent-of-citations-are-deep-pages (10.5, 2 impressions); does-ai-cite-one-page-or-many-from-the-same-site-82-percent-of-cited-domains-contribute-a-single-url (11, 1 impressions). Position 10.0 is a bookkeeping line drawn through a continuous distribution, and this week the line moved across a URL rather than the URL earning anything new.

The five page-one URLs that earned a click in the newest window

40 of the 45 page-one URLs earned no click at all in 28 days. The four single-click URLs sit at positions 4.6, 2, 6.8, 7.5, on 76 impressions between them.

Where the site's impressions sit by position

Page one holds 23.89% of the site's 3365 web impressions in the window and 73.33% of its 15 clicks. Positions 71 to 90 hold 1453 impressions and 0 clicks across 15 URLs; 1209 of those impressions sit on five /compare and two /guide pages, classes our yield instrument scores impression-only (no new page opens in them), and the remaining 244 on /library, one /data page and five /news posts.

/news inside the window: posts published before it and posts published during it

63 of the 116 /news posts live at the window's close were published inside it. Our yield instrument has recorded /news's average position sliding for five days as those posts filled the window, and the obvious reading is that new, unranked URLs drag the mean. The weekly split does not support that reading on its own: in the final week the in-window cohort carried 261 impressions at weighted position 29.2, and the pre-window cohort carried 97 at 45.6, having read 10 in the first week. The older posts slid further than the new ones. Impression-weighted position on a few dozen impressions a day is a noisy number, and a handful of deep-position impressions on an old post can move it by ten places; this table is the denominator for next week's read, not a conclusion.

What this means if you run a small site

Two things follow. First, a CTR read on a page-one bucket this size is a read on one or two URLs; report the bucket with and without its largest row, or the number is decoration. Second, the six-window series without the one URL (0.63%, 0.26%, 0.68%, 0.52%, 0.48%, 0.53%) is flat inside noise while the bucket tripled in impressions, which is the third consecutive week our data favours the hypothesis that these page-one queries are zero-click by nature rather than snippets a better title could win (the 2026-08-30 join added that Googlebot had not re-fetched 30 of the 46 page-one URLs in 7 days, so most title changes could not have been measured anyway). Our own check works from the other side of this: it asks the AI engines the buying questions a customer would ask and reads whether a business is named at all, which does not depend on a Google click.

Pre-registered read for 2026-09-08

• (A) On the 28-day window ending 2026-09-06, the page-one bucket without the one URL reads a CTR between 0.2% and 0.8% again. Falsified if it reads 1.0% or more, or 0 clicks on 500 or more impressions.
• (B) The Search Console API post keeps an average position of 10.0 or better and earns at least 1 click on that window. Falsified if either fails.
• (C) In that window's final week, the /news posts published before the window again read a worse weighted position than the posts published inside it. Falsified if the older posts read better.
• Scoring: the same script on the same API calls with the window shifted one week; the 2026-09-01 artifact is reproduced exactly as the first check.