Average position in Search Console: why it misleads
Average position blends every query and page into one number. Here is how Google calculates it, why it can drop when you improve, and how to read it safely.
Search Console shows one number labeled average position. Teams put it on dashboards and treat a drop as a problem. Often the drop means the opposite. The number blends every query and page into a single average, and that makes it easy to misread.
How Google calculates average position
Each time your site appears in a result, Search Console records a position for that impression. If several of your pages show up for the same search, it records the highest one. Position 1 is the top result. The report then averages those positions across impressions.
Two details matter. First, the average is weighted by impressions, so a query shown a thousand times counts far more than one shown ten times. Second, a result only counts once it is shown. Your page at position 40 does not count unless it was actually displayed to someone.
Google's documentation also notes that results in a carousel or other special layout can share a position. Treat the figure as a useful signal, not a precise rank.
Why it can drop when things get better
This is the trap. Say a page ranks at position 4 for ten queries. You improve it, and it starts to appear for 200 new long-tail queries at positions 25 to 60. Traffic rises. Impressions rise. Average position gets worse, because the new low-ranking impressions pull the mean down.
The reverse also happens. If Google stops showing you for your weak queries, your average improves while you lose reach. A better number can mean a smaller footprint.
A moving average position tells you the mix of queries changed. It does not tell you rankings changed.
Filter before you read it
The site-wide average is the least useful version. Narrow it down before you draw any conclusion.
- Filter to a single page, then add the query dimension to see which searches that page ranks for.
- Filter to a single query, then add the page dimension. This shows whether two of your pages compete for it.
- Split branded from non-branded queries. Brand terms usually sit at position 1 and hide the rest.
- Filter by country and device. A page can rank well on desktop in one country and poorly on mobile in another.
- Compare equal date ranges, such as 28 days against the previous 28 days, so you do not mix weekday and weekend patterns.
A page-level and query-level view is where real movement shows up. If one page at position 8 for a query you care about slips to 14 over a month, that is worth investigating. A site-wide shift from 18.2 to 19.1 usually is not.
What to track instead
Pair position with the numbers that show whether people see and click your pages.
- Clicks and impressions for a fixed set of target queries.
- Click-through rate for each page, compared against pages at a similar position.
- The count of queries that put a page in the top 10, tracked over time.
- Pages with impressions at positions 8 to 20. These are the closest wins.
The last item is the most practical. A page at position 11 to 15 for a good query is one improvement away from page one. Raising it often takes a better title, a clearer answer near the top, or a few internal links, not a rewrite.
Where AI results fit
Google has said that appearances in AI features such as AI Overviews and AI Mode count in Search Console's Performance report. As far as we know, the report does not break them out as a separate filter, and how position is assigned inside an AI answer is not well documented. Because of that, treat position data for queries that trigger AI answers with extra caution. Rules here may change, so check Google's current documentation before building reports on it.
Position also says nothing about citations in ChatGPT or Perplexity. Those need a separate tracking method.
A simple routine
- Export the last 28 days of query and page data, with branded queries removed.
- List pages with the most impressions at positions 8 to 20.
- Pick three and check whether the intent matches the page, then improve the title, the opening answer, and internal links.
- Recheck after 4 to 6 weeks using the same filters and date range.
- Ignore the site-wide average in the meantime.
The takeaway: stop reporting the single site-wide number. Report position for a fixed list of queries and pages, next to clicks and impressions, and you will see what actually changed.