Average position: why it misleads and how to track rankings properly

"Average position" is the most quoted number in Search Console and the one most often used to draw the wrong conclusion. It is a real measurement, but it is an average over thousands of different searches, and averages behave in ways that feel backwards. Understanding how it is calculated tells you when to trust it and how to track rankings in a way that actually supports decisions.

How the number is built

For every impression, Search Console records the topmost position any of your URLs held on that results page. If your homepage was at 3 and a blog post at 7 for the same search, that impression counts as position 3 for the property. The report then averages those positions, weighted by impressions: a query with 10,000 impressions moves the average far more than a query with 10.

A few details that matter:

• Positions count result blocks, not just blue links. Elements such as a top stories carousel or a featured snippet occupy positions. A result can be "position 1" and still sit below an ad block, since ads are not counted.
• Each impression is a different search. Location, device, language, personalisation and the time of day all change the page. The same query can put you at 2 for one user and 9 for another.
• No impression, no position. If you rank 45 for a query but nobody opens page 5, that query contributes nothing to the average.

Why the site-wide average lies

Consider a site with one strong query: 1,000 impressions at position 2. The average is 2. The site publishes useful new articles and starts appearing for 40 new long-tail queries at positions 30–50, adding 1,000 impressions in total. The old query has not moved, but the property average is now about 21.

Nothing got worse. The site gained visibility. Yet a dashboard showing "average position: 2 → 21" looks like a disaster. The reverse also happens: lose all the weak, low-ranking queries (for example after deleting thin pages) and the average "improves" while traffic falls.

This is a mix effect, and it makes the property-level average close to useless as a ranking metric. It is still a decent alarm when it moves together with clicks: if impressions are stable and the average worsens by several places, something real happened.

Page-level and query-level averages differ too

A page that ranks for 300 queries has one average that blends its best and worst queries. Two pages competing for the same query (cannibalisation) each show their own position in the page view, while the property view only keeps the better one. When the page view and the query view seem to contradict each other, this is almost always why. Filter to one query and switch to the page tab to see which URLs Google is choosing between.

Tracking rankings properly

1. Track per query, per day

The useful unit is one query on one day: "invoice template word was at 4.2 on 3 March and 8.9 on 10 March". Plotting that line for the queries that bring money or leads tells you far more than the site average. Pick a short list — 20 to 50 queries that matter commercially — and watch those.

2. Look at impressions alongside position

A query with 30 impressions a week can jump from 6 to 14 because three searches happened on mobile in another country. Before reacting to a position change, check that the query has enough impressions for the average to be stable. As a rough working rule, treat daily positions on queries with fewer than about 20 impressions a day as noisy and read the weekly trend instead.

3. Segment by device and country

Mobile and desktop rankings differ, and many sites rank very differently in each country. A site selling in the UK and the US should look at the two countries separately; a blended average can hide a drop in one market behind growth in the other. In Search Console, apply the country and device filters before reading the position column.

4. Compare windows of equal length

Average position over 28 days smooths out daily noise; over 1 day it is jumpy. When comparing, keep the same window length and the same weekdays on both sides.

Search Console versus rank trackers

Classic rank trackers send a query from one location, usually a data centre, on one device type, and record where your URL appears. That gives a clean, repeatable number that is easy to chart. It is also one synthetic search, not what your visitors see.

	Search Console average position	Location-based rank tracker
Source	Every real search that showed your site	One simulated search per check
Personalisation, location mix	Included, averaged	Excluded (fixed location)
Queries covered	Every query with impressions, including ones you never thought to track	Only the keywords you entered
Competitor positions	Not available	Usually available
Good for	What actually happened to your traffic	Watching a fixed keyword list and competitors

The two numbers will disagree, and both can be right. If a tracker says position 3 and Search Console says 6.4, the tracker's location probably sees a cleaner page than the average user does. For decisions about your own traffic, the Search Console numbers are the ground truth because they come from real searches.

Reading a position change: a checklist

1. Did impressions change at the same time? If impressions fell a lot, the average may have moved because of mix, not ranking.
2. Is the change on one device or one country only? Filter and check.
3. Did the ranking URL change? Check the page tab for the query. A swap between two of your own pages often shows as a drop.
4. Did the results page change? Search the query yourself in a private window and note new features such as AI Overviews, video blocks or "People also ask".
5. Did your site change? A deploy that broke titles, canonicals or internal links tends to move many queries at once. Run the SEO checker to rule out technical regressions.

What DeskSEO stores

DeskSEO keeps the daily position, clicks and impressions for each query and each page it pulls from your Search Console, so you can open any single query and see its own line over up to 16 months instead of relying on the blended figure. The setup takes a few minutes on the Search Console connection page.