The Average Position Halved. Now I Know Which Half Was True.

I wrote a piece in September about this site's average Google Search Console position falling from the 14-20 range to 7.8-9.4, right as a dozen new articles went live. I had two explanations and no way to tell them apart: either the existing pages were genuinely ranking better, or a pile of new pages entering the impression mix at decent starting positions was just pulling the site-wide average down without any single page actually moving. One data point can't distinguish a real trend from an averaging artifact. I said so at the time and left it open.

I now have the second data point.

Two comparable 28-day windows

Google Search Console's exports are more useful compared against each other than read in isolation, and I'd been exporting the same "last 28 days" report every week or two without always lining the windows up cleanly. These two do line up: 2026-08-08 through 09-04, then 2026-09-07 through 10-04, back to back with only a 3-day gap. Same metric, same window length, genuinely comparable.

Site-wide, the averages are: 79 clicks and 4,518 impressions in the first window, 78 clicks and 3,995 impressions in the second — essentially flat traffic. CTR went from 1.75% to 1.95%. And the headline number, the one the original piece couldn't explain: average position went from 14.87 to 9.24. Still halved, on a second, independent measurement. So the improvement itself is real. The question was always which half of it was a real ranking gain.

Splitting the average by what's actually inside it

An average position is a sum divided by a count, and both the sum and the count can move. To tell a ranking gain apart from an impression-mix shift, I matched every query that shows up in both exports' Forespørgsler.csv (query-level) files and asked what happened to each one individually, rather than trusting the one aggregate number either export reports on its own.

46 queries appear in both windows. For those 46, position barely moved at all: median change +0.1, 15 queries better, 19 worse, 12 within half a position either way. That's noise, not a trend. If the site's existing rankings had genuinely improved, this is exactly the set of queries that should show it, and it doesn't.

93 queries that appeared in the first window are gone from the second. Those are the ones that explain the whole thing: their median position in the first export was 20 — deep, low-visibility rankings that Search Console doesn't have enough volume to keep reporting on or that simply dropped out of visibility entirely. And a query ranking at position 20 contributes its full weight to an average the same as a query ranking at position 2. Drop sixty queries that were only ever ranking in the twenties, and the average over what's left goes up without a single page climbing anywhere.

That's the mechanism directly: the share of total impressions coming from pages ranked worse than position 20 fell from 17.1% to 3.6% between the two windows. Impressions didn't move from a bad position to a good one on the same page — they moved from being mostly absent on the deep end of the ranking curve.

The page-level check doesn't survive, either

Before settling on that, I ran the same comparison one level up, on Sider.csv (per-page position) instead of per-query. On the surface this looks like it argues the opposite: of the 104 pages present in both exports, 89 improved, with a median swing of -7.5 positions. That's a big, consistent-looking gain, and if I'd stopped there, I'd have written the wrong conclusion.

But a page's average position is itself an average — over every query that page happens to rank for, weighted by that query's own impressions. It's exactly the same arithmetic as the site-wide number, one level down. If the queries dropping out of a page's own ranking mix were its worst-performing ones, the page's average improves for the identical reason the site's average improves, without the page's actual ranking on any query changing at all. The per-page number looked like independent confirmation. It wasn't independent; it was the same artifact, recursing.

What's actually true, and what I still don't know

The real, supported conclusion: this site's visible average position improved because a large number of barely-visible, deep-ranked queries stopped contributing to the average, not because any of the 46 queries this site was already reliably ranking for climbed higher. The CTR increase (1.75% to 1.95%) is consistent with that story too — losing impressions concentrated at position 20+, where almost nobody clicks anyway, raises the average click-through rate on what's left even with flat raw clicks.

What I can't say from this data is why those 93 queries stopped showing. Search Console doesn't report the specific reason a query falls out of a window — rounding below its impression threshold, a genuine ranking drop past what it tracks, or query-level noise from a small site are all consistent with what I measured, and I don't have a third export that would let me tell those apart either. The honest version of this piece is narrower than the one I'd have written off the aggregate number alone: the halving is real, it's explained, and it isn't the "my new content is ranking well" story I'd have liked it to be.