Average position is the most confidently misread number in SEO. It falls when sites improve, rises when they shed keywords, and measures something subtler than anyone assumes: the impression-weighted average of your best slot, per query, per page, per day.

Here is what the number actually counts, the paradoxes that follow, and the decoder that turns them back into truth.

What one impression records

Every time a results page loads with your site on it, GSC records where your topmost result stood, whether anyone clicked or not, whether anyone scrolled or not. Rank 3 and rank 8 on the same page? The record says 3.

Average position is those records averaged across everything your current filter includes. Which means the metric's movement has two possible causes: your positions changed, or the mix of impressions changed. The second one causes most of the confusion.

The mix-shift paradox

Last month "rank tracker": position 4 one ranking query average position: 4.0 clicks: 400 This month "rank tracker": still position 4 NEW "rank tracking tools": position 38 average position: 21.0 "worse" clicks: 430, more than before
Nothing fell. A new query joined at position 38, the average absorbed it, and the site got better while the number got worse.

This is the default physics of a growing site: new content and expanding relevance enter the average from the bottom. A worsening sitewide average with rising clicks is a growth signature, not a decline.

The reverse holds too, and it is nastier: a sitewide average that improves while clicks fall often means you lost your long-tail impressions entirely, and the survivors just average better. Celebrating that number is celebrating shrinkage.

The paradox decoder

Avg positionClicksUsually means
WorseUpNew queries joined at low positions: growth
BetterDownLong-tail impressions lost: shrinkage in disguise
StableDownSERP reshaped or snippets weakened: a CTR problem
WorseDownReal ranking decline: now investigate
BetterUpThe one honest happy case

Only one row of five means what the raw number implies. The decoder's rule: never read average position without clicks and impressions beside it.

Reading it right, in practice

Shrink the mix before trusting the average. One query plus one page is an honest number; one query across devices and countries is already a blend, as the filter discipline covers. Sitewide averages are weather reports: fine for mood, useless for diagnosis.

Compare like windows, because the mix breathes with seasons and news cycles. And when the stable-position-falling-clicks row appears, take it to the CTR workflow: that pattern is a snippet or SERP-furniture story, not a rankings one.

One more calibration: position 1 is your topmost counted result, not necessarily the top of anyone's screen. Ads, packs and AI answers stack above it, which is why even the throne can feel drafty in the click column.

What "position 40" is trying to say

New pages living at 30 to 50 for their target queries are not failing; they are queued. Google is showing them experimentally on deep pages, and impressions there are the first sign of relevance being tested.

Watch those queries over weeks: climbing into the teens marks the page as promotable with internal links and a refresh, the striking-distance play from the reports guide. Position 40 is a waiting room, and some of your best future keywords are sitting in it now.

The one-line takeaway: average position averages your best slot per impression, so it moves whenever the mix moves: growth drags it down, shrinkage dresses it up. Decode it with clicks and impressions beside it, filter to one thing before trusting it, and let only the worse-and-falling row send you investigating.