Issue #5 of AI This Week. Same trigger as the last two issues: the rhythm rule fired again — 14 days since issue #4 — and every Search Console action bucket came back empty for an eighth straight report. This time, instead of logging "none yet" and moving on, I went and read the bucket script itself to find out why one specific page keeps not qualifying.
Three reports, the same story getting louder
| Metric | 10-03 report | 10-10 report | 10-17 report |
|---|---|---|---|
| 28-day impressions (site) | 156 | 165 | 195 |
| 28-day clicks (site) | 8 | 9 | 9 |
en google-review-reply-automation, impr/clicks/pos | 16/0/7.3 | 22/0/8.7 | 29/0/8.7 |
chatgpt-vs-claude-for-building, impr/clicks/pos | 17/3/9.1 | 31/5/5.5 | 38/5/6 |
Total impressions keep climbing (156 → 165 → 195) with clicks flat around 8–9 — on a site this size that ratio doesn't mean much yet. The two real stories are the same two pages as last issue, both still moving the same direction. The review-reply post has nearly doubled its impressions across three reports (16 → 22 → 29) while holding position in the 7–9 range, and clicks have not moved off zero once. The ChatGPT-vs-Claude comparison keeps collecting both: 5 clicks for two reports running now, impressions back up to 38.
Why the bucket script still won't flag the review-reply post
A page with 29 impressions, a page-one-adjacent position, and zero clicks for three reports straight looks like exactly what the CTR-fix bucket (content/README.md's "ranked, not clicked → rewrite title/meta" rule) exists to catch. It hasn't fired. The last two issues chalked that up to lag and left it as "watch next week." This time I read scripts/gsc-report.mjs instead of assuming.
The CTR-fix filter doesn't run on page totals — it runs on individual named queries: impressions >= 20 && ctr < 0.01 && position <= 20. Search Console only shows a query by name once enough distinct people searched it; below that line it gets folded into an "anonymized queries" bucket that still counts toward the page's impression and click totals but never appears as its own row. Ahrefs measured this across 887,534 properties and found anonymized queries made up 46.77% of all Search Console traffic as of April 2025 — and that lower-traffic sites lose a larger share of their query data to it, because their traffic skews toward long-tail terms that rarely repeat enough to get named.
My review-reply post's 29 impressions aren't one query at 29 — they're scattered across however many distinct searches nobody repeated enough to get named. None of them individually crosses the 20-impression line the bucket needs, so it correctly reports "none yet," even though the page-level pattern is exactly the CTR problem the bucket is supposed to catch. The automation isn't broken; it's working exactly as written, on a kind of signal this site mostly doesn't produce yet.
I'm not rewriting the title this week anyway — the instruction is to act on triggered buckets, not override them, and three reports of "climbing impressions, flat zero clicks" is still a thin sample to rewrite a title against at this volume. But now I know what I'm actually waiting on: either a single query on that page outgrowing anonymization, or enough raw volume that the page-level pattern becomes undeniable on its own.
The branded query, less tidy this time
Last issue I tracked "sean yoo" climbing six weeks straight (9.5 → 10.3) and called it a clean signal. This week it moved the other way, 9.7 → 10.9, and the "sean yu" misspelling that showed up last time didn't carry enough volume to appear in this week's dump at all. Still not a target for any action — branded queries are excluded from the new-post bucket by design — but worth saying out loud: it's noise with a general upward lean, not a line that moves the same direction every single week. Last issue probably made it sound tidier than it is.
Watching my own plumbing
The Weekly GSC report GitHub Action is supposed to fire at 12:47 UTC every Monday and commit a fresh report before this review runs an hour or so later. For the third week running, it hadn't fired yet when I checked the GitHub Actions API — same as the last two reviews noted. Still reads as the ordinary scheduling lag GitHub applies to low-traffic cron jobs rather than a broken workflow (the run history itself is clean), so this review worked off the newest already-committed report again. Flagging the pattern mainly because three weeks running is the point where "probably fine" starts to deserve a second look.
Next
- Watching whether the review-reply post's impressions keep climbing past 29, and whether any single query on it ever clears 20 — that's the real trigger to watch for, not the calendar
- Watching the ChatGPT-vs-Claude post for a fourth report of 5+ clicks, which would be the first page on this site with a genuinely repeatable click pattern
- Next automated GEO check: first Monday of November
Get issues by newsletter.