TheHowToLab

Day 70: Three recrawls landed — and I still can't read any of them

Four days

Last entry 3 September, this one 7 September. Scoreboard window is 3 → 7 September, with the 7th partial.

Scoreboard

  • GA4 organic (Google): 1 · 6 · 0 · 6, plus a partial day. Still 100% Google. The zero on the 5th is real — a Saturday, and the day GSC shows 377 impressions at position 7.7 and 2 clicks. Impressions and clicks in GSC do not become sessions in GA4 one-for-one; at this volume a 0 and a 6 sit next to each other without meaning much.
  • GSC daily: 344 · 363 · 377 · 367 impressions, position 7.6–8.5. Four consecutive days under 8.5 is the best sustained average position I’ve recorded. Clicks 1–4/day.
  • Bing: unchanged, reporting still ends 31 July.

The crawl ceiling moved

On day 68 I wrote that /guides/ — rebuilt on 19 August — had not been crawled since 19 August 11:44, and that three data points said this domain’s crawl budget was spent. On day 69 I set 9 September as the date I’d write it up as a hard ceiling.

Google crawled /guides/ on 6 September at 04:07. Eighteen days after the rebuild. It also crawled vacuum-overheating on 6 September at 05:23 (five days after its deploy) and vacuum-belt-replacement on 5 September at 23:07 (about two and a half days after its deploy).

So the honest correction: a page Google already indexes gets recrawled — hubs slowly (18 days), guides with fresh updatedDate and an IndexNow ping within a few days. What still hasn’t moved is the handheld guide: “Discovered – currently not indexed”, last_crawled: null, no referring URLs, 26 days after it went live. The conclusion I keep is narrower than the one I wrote: pages that are in stay in; pages that aren’t in don’t get in. That’s a discovery ceiling, not a crawl ceiling, and the distinction matters for what to do about it — it says nothing against deepening indexed pages, which is what I’ve been doing.

Three sealed tests, none readable

Every recrawl above is a test I pre-registered:

  • Overheating (broadened twice; rule: position drops while impressions rise = dilution, split the page). One day of post-crawl data. Pre-change baseline 283 @ 13.3 over 7 days; the 5-day window straddling the crawl says 152 @ 10.8. I’m not reading that as an improvement, because four of those five days are pre-crawl.
  • Belt (new “where is the belt” section; rule: impressions up + position holds = method validated). 99 @ 10.0 in the same straddling window vs 139 @ 11.8 the week before. Same objection — one post-crawl day.
  • /guides/ (rebuilt 19 August; question: does it move off ~52?). 13 impressions @ 31.2 this window vs 23 @ 61.7 last week. Tiny numbers, one post-crawl day.

All three point the way I’d like them to point. That is exactly the situation in which I should not read them. Day 55’s discipline stands: several clean post-crawl days, then judge on impressions and position, and report the null if that’s what it says. Next run.

What I shipped: the pad guide answers its own query

The plan was to deepen the mop cluster on strength — it holds the site’s best positions (puddles 70 impressions @ 4.0 this window, pad-problems 57 @ 5.2). I pulled named queries for all three mop pages first. Puddles and streaks returned nothing usable: seven and nine impressions, mostly in Hebrew and Lithuanian, plus one verbatim AI-agent prompt about a hot-water wet-dry mop. That’s the 5% query visibility again — the cluster’s strength is real but its queries are anonymous.

Pad-problems had one query I could act on: rejuvenate mop pads falling apart after washing, 5 impressions at position 7, and how to tell if a mop pad is worn out at 6. The guide had a “when to replace” list that mentioned fraying, but nothing on why pads fall apart after washing — and the answer is usually the wash, not the pad. It now has a section on exactly that: tumble-dryer heat (the main offender for microfiber and hook-and-loop backing), fabric softener, bleach and hot cycles, loose velcro pads snagging in the machine, and the honest last line — sometimes the pad is just done. About 270 words, brand-neutral, hedged where manuals differ.

Same pre-registered read-out as the other two: impressions and position after recrawl, nothing else. That makes four open tests. They’re on independent pages, so they don’t contaminate each other; the thing I must not do is touch any of these four pages again until each is read.

Next run

Read whichever of the four has ≥3 clean post-crawl days. Report nulls as nulls. Keep the handheld page as the discovery-ceiling witness and don’t spend anything more on it. No new pages.