Field notes
The discovery window
On July 11 we published a field note about a small-business finance data site in our network: clean technicals, original data, average position around 40, verdict of links. Four days earlier, though we did not know it yet, Google had already ended the site's trial. Daily impressions fell from 282 to 36 to single digits inside 72 hours, and stayed there. This note is the forensics, and the rule we bought with them.
By Mario Bailey · Published
Let us set the stakes honestly before the chart makes them look dramatic: the site has earned twenty clicks in its lifetime, at an average position near 40 and a 0.7% click-through rate. Nothing that pays was lost. What was lost is more useful than traffic: a clean, well-instrumented look at how Google treats a brand-new domain that shows up with a very large sitemap.
Three weeks, start to finish
The daily series as a table
| Jun 20 | 0 |
| Jun 21 | 53 |
| Jun 22 | 248 |
| Jun 23 | 235 |
| Jun 24 | 181 |
| Jun 25 | 183 |
| Jun 26 | 247 |
| Jun 27 | 118 |
| Jun 28 | 156 |
| Jun 29 | 285 |
| Jun 30 | 185 |
| Jul 1 | 134 |
| Jul 2 | 204 |
| Jul 3 | 258 |
| Jul 4 | 282 |
| Jul 5 | 36 |
| Jul 6 | 15 |
| Jul 7 | 6 |
| Jul 8 | 12 |
| Jul 9 | 5 |
| Jul 10 | 3 |
| Jul 11 | 2 |
| Jul 12 | 4 |
First impressions on June 21, the day after the sitemap was submitted. Two weeks between roughly 120 and 285 impressions a day, spread across hundreds of URLs. Then July 5: 36. The days after: 15, 6, 12, 5, 3, 2, 4. In the last healthy week about 370 URLs earned at least one impression. The week after, 30 did. Every section of the site fell together and in proportion: the entity pages, the editorial guides, the money pages, the homepage queries. When one page class dies, you look at that page class. When everything falls by the same ratio on the same day, the verdict was passed on the domain.
The window that hid it
Here is the uncomfortable part. Our July 11 note described this same site as having real momentum, clicks rising across consecutive 28-day windows. That sentence was true when the windows were computed and already false as a description of the present: the cliff was six days old. A 28-day total is a low-pass filter. It is the right instrument for judging a trend and the wrong one for noticing that the trend has ended. The daily series above was sitting in the same Search Console property the whole time, one click away, telling a different story.
The rule we now apply to our own dashboards: cumulative windows for decisions about trends, the daily series any time a number surprises you in either direction. Regime changes only exist at daily resolution.
What the index actually said
Search Console's URL Inspection tool turned the mystery into a mechanism. Three findings, in increasing order of interest.
First: the site's best pages are out of the index, not ranked lower. The guide that earned the most impressions, and the strongest of the entity pages, now return Crawled - currently not indexed. That status is not a penalty and not an error. It is Google saying: we fetched this page, we understood it, and we are choosing not to spend an index slot on it. There is no notice, nothing to fix, and nothing to appeal.
Second: the eviction happened without a recrawl. The top-earning guide was last crawled June 21 and kept earning impressions through July 4. Google did not revisit the page and change its mind about the content. It recomputed something about the domain and dropped the page from the index without looking at it again. Whatever was re-evaluated on July 5, it was not the pages. It was the site.
Third: most of the site was never in the game at all. Every live, sitemap-listed page we sampled outside that first crawl batch returns URL is unknown to Google: never crawled, never discovered, no record. Google took one bite of the site in late June, served impressions off that snapshot for two weeks, and never came back for more. The sitemap report agrees: last downloaded July 6, and the copy Google holds is the old one, from before we slimmed it.
Put together, that is what a discovery window looks like from the inside. A new domain gets a trial: liberal indexing of an early snapshot, real placement in live results, and a measured audience response. Ours came back at a 0.7% click-through rate from an average position of 40. Google graded the trial, closed the window, and reallocated the crawl budget somewhere more promising. The window was never a ranking. It was an audition, and we mistook the audition for the early innings of a climb.
What it was not
We checked the usual suspects so you do not have to take the exotic explanation on faith. Almost certainly not a manual action: those announce themselves in Search Console's manual actions report, and a domain-level one does not typically leave the homepage and section hubs indexed, which ours are. Not technical: robots is open, canonicals are correct, the duplicate host redirects, and the pages fetch clean; the inspection results themselves say indexing allowed. Not an algorithm update: the May core update finished June 2 and a spam update finished June 26, a week and a half before the cliff; possibly a contributor upstream, but nothing rolled out on July 5. And not the value-gating we shipped, which removed about 1,500 thin pages from the sitemap: that landed on July 7, two days after the verdict, and Google has not re-read the sitemap since. The cleanup was right. It was also late.
The rule we bought
The site went live with well over two thousand URLs in its sitemap, most of them programmatic entity pages built on a real public dataset, and the count passed four thousand within two weeks. Every one of those pages clears our own quality gates, and we would defend any single one on its merits. That was never the question Google was asking. Index slots are not granted per page on page quality. They are granted per domain, on trust, and a three-week-old domain has roughly none. We asked a brand-new domain to carry thousands of pages on zero authority, and the trial measured exactly that.
So the rule, which we will apply to every launch from here: a new domain does not get to submit two thousand pages. Launch with the fifty to a hundred pages that could plausibly rank on their own merits, earn the links and the engagement that build the domain's budget, and let the programmatic tail in behind the authority, not ahead of it. Publishing the tail first does not just fail to rank. It spends the one liberal-indexing trial you get proving your median page is not worth an index slot. And gating the tail after launch, as we did, is damage control on a verdict already passed. The gate has to precede the launch.
The position-40 note ended by promising that its sequel would cite it whether the thesis held or broke. This is the sequel, earlier and harsher than we expected, and the thesis held in the strictest form: authority was not just the ceiling on the site's rankings. It was the ceiling on its right to be indexed at all.
Check your own site
Four steps, ten minutes, all in Search Console:
- Switch the performance chart to daily and look at the last 60 days. Cliffs, not slopes, are what you are looking for. A cliff means a verdict; a slope means a trend.
- Inspect your five most important URLs. Crawled - currently not indexed on money pages is the single most informative status Google publishes: content understood, slot declined.
- Read the last-crawled dates. If your best pages have not been fetched in weeks while you publish daily, Google's interest in the domain, not the content, is the constraint.
- Check the sitemap's "last read" date against the version you actually serve. Google acting on a stale copy of your site is more common than most cleanup plans assume.
The diagnosis discipline The position-40 ceiling
Method: the figures on this page come from our own records, the Search Console UI and API, rounded as written. The sites are ours, and we deliberately do not name them. Field notes describe what happened on our network, not a promise of what will happen on yours.