Do page-one Google rankings earn clicks for a small site? 51 URLs on page one, 881 impressions, 4 clicks — 0.45% across 19 daily reads

The series, read by read

Method: one Search Analytics API call per day against the property, web search only, page dimension, complete row set (no top-N truncation), window = trailing 28 days ending two days before the read (Google's last two days are unstable). The bucket is every page whose 28-day average position is ≤10. Positions are per-page averages across all of that page's queries, so a page averaging 9 can carry impressions at 30; we use page rows anyway because query rows silently drop the roughly 57% of impressions Google anonymizes, which would make the bucket disagree with every other number from the same property. The 4% column is an assumption the series has carried since its first read, not a measured benchmark — it is there so the gap has a size. *The 08-16 read's 9 clicks was a bucket-exit artifact (a clicked page briefly inside the bucket) and reversed the next day; it stays in the table because it happened. The series is saved machine-readable beside this post.

Today's read, URL by URL

• 47 of the 51 page-one URLs earned 0 clicks, on 753 impressions between them. The two largest pages in the bucket — 142 and 105 impressions at average positions 7.9 and 7.8 — are both in that group.
• The four clicks came from four different pages, three of them small. The lost-citations post took its single click at average position 2.0 on 2 impressions — the first click at position ≤3 in the series' history. The position-≤3 slice as a whole reads 4 URLs, 7 impressions, 1 click.
• Of the 255 queries Google discloses for the property, exactly one averages a page-one position: `chatgpt://url-citation?citation-number=0` — 14 impressions, 0 clicks, position 7.6. That is a ChatGPT citation-artifact string someone pasted into Google. Every other page-one impression on this site is query-anonymous; the bucket is 42 /news posts, 3 /guide pages and 6 other pages.
• Impressions compound, clicks do not. Over the nineteen reads the bucket's impressions rose 130% and its URL count 70%; clicks moved between 3 and 5 on every clean read. Whatever Google is surfacing these pages for, it is not producing a proportional click series.

Four titles were rewritten nineteen days ago — the read so far

Four rewrites, nineteen days, 208 page-one-or-near impressions on the rewritten pages since, 2 clicks — 1.0%, against the bucket's 0.45%. Two clicks on four pages is not a read; the windows overlap the pre-rewrite period, and as of 2026-08-17 Google Search Console showed none of the four recrawled since its rewrite, so the results page was still serving the old titles. It is recorded here so that the read on 2026-09-05 cannot be told from memory. Correction 2026-08-23: this section first gave the rewrite date as 2026-08-16; the titles were changed on 2026-08-03 (the at-rewrite readings above are from that day).

One visitor wrote our URL from memory

On 2026-08-21 a visitor who had arrived at the homepage directly, read it, opened the news index and then requested `/ai-search-news/can-you-get-search-console-generative-ai-report-from-api` — a path that does not exist and was not linked from the page they were on. It is a paraphrase of `/news/can-you-get-search-console-generative-ai-data-from-the-api`, which happens to be the site's most-clicked page in the window (5 clicks on 29 impressions at average position 10.6 — just outside the page-one bucket). A typed or pasted URL that names the right post with the wrong words is the shape of a URL an assistant wrote from memory. It is 1 of 46 distinct requested paths in 30 days, so it is a datum and not a finding; we shipped a read-only sensor for the shape the same day and no redirect, because n=1 does not earn one.

Two hypotheses, and the date the read is decided

• (a) The snippets do not earn the click. Titles and descriptions fail the query; rewriting them moves CTR. If this is right, the four rewrites above should separate from the bucket within two more weeks.
• (b) The impressions are zero-click by nature. The queries these pages rank for are answered on the results page, or are not human queries at all — the one disclosed page-one query is a citation-artifact string, and the bucket's composition is news posts about AI search. If this is right, no title moves the series and the finding is the more useful one: page-one impressions on this class of page are not distribution.
• Pre-committed read: 2026-09-05. If the bucket's impressions are ≥1,000 and its 28-day clicks are still ≤5, (b) is favoured and title edits on the bucket stop. If clicks are ≥10, (a) is favoured. Published either way, with the series extended.

What cuts against the clean story

Nineteen reads of overlapping 28-day windows are not nineteen independent samples — each read shares 27 of its 28 days with the one before, so the flat click line is mostly the same clicks being counted again. The 4% column is our assumption, not a benchmark. The page-average position basis means a page inside the bucket can have taken most of its impressions from deep pages, which would make (b) look stronger than it is. And this is one site, 71 days old, whose page-one pages are almost all news posts — a site selling a product on its ranking pages could read entirely differently. What this site sells is the step before any of this: whether ChatGPT, Perplexity, Gemini and Claude name a business when its buyers ask — the free check reads that directly from the engines' citation payloads.