How average position is calculated
For every impression, Search Console notes where the highest-placed result for your site sat in that particular results page, then averages those numbers across all the impressions in the period. It is an average of appearances, weighted by how often each query occurred — not a record of where you rank.
That weighting is the whole problem. The figure moves whenever the mix of queries changes, and the mix changes constantly. Publish a page that starts appearing far down the results for many new searches and the average worsens, although nothing existing has slipped. Lose impressions on a batch of weak, low-placed queries and the average improves, although you now get less traffic. The number moved; the site did not.
Position itself is also unstable. Results differ by device, by location, by language and by what else the search engine chose to put on the page that day. Google Ads eventually retired its own average position metric in favour of top and absolute top impression rates, precisely because a single averaged number told advertisers so little.
Why average position matters
Read narrowly, it is still useful. Filtered to one query and one country, its direction over months tells you whether a page is drifting up or down, which is a reasonable early signal before clicks change enough to notice.
It is also the only positional data that comes from Google itself rather than from a tool sampling results. That makes it a sensible cross-check when a rank tracker reports something surprising.
Where average position goes wrong
The commonest failure is putting the site-wide figure in a monthly report as though it were a ranking. It is an average across every query the site appears for, most of which nobody cares about, and it can only be interpreted alongside the impression count that produced it.
The second is setting targets on it. Once a number becomes a target, the quickest way to improve it is to stop appearing for difficult queries — which is the opposite of growth.
The third is comparing periods without checking the denominator. If impressions have doubled between the two periods being compared, the two averages describe different populations of searches and are not comparable at all.
How to act on it
Filter before you read it. One page, one country, one device, and a query filter if you can. Compare the same slice across periods and check whether impressions moved before you interpret the position at all.
When you genuinely need to know where you rank for a specific term in a specific city, use rank tracking rather than this metric — that is what it is for. And keep the reporting honest: clicks, enquiries and revenue belong at the top of a report, with position as supporting detail. If a report is built the other way round, an SEO audit is usually the faster way to find out what is really happening.