Name the scope behind the number
An average position is a summary, so start by identifying what it summarizes. Save the property, selected page or page group, dates, search type and filters. A number attached to the whole site answers a different question from one attached to a particular service page and query. Do not label either one as the permanent rank of the business.
Google describes this metric using the topmost result for the selected property or page, averaged over recorded impressions. It also explains why one manual search can differ from the report. Use the metric to investigate a defined set of observations rather than to promise that every customer sees the same result.
Keep impression weights with the averages
When exporting query rows, retain impressions alongside position. A row with two impressions should not receive the same weight as a row with two thousand simply because each occupies one spreadsheet line. A plain average of the position column answers a different question from an impression-weighted calculation.
For a deliberately defined, non-overlapping set of rows at the same aggregation level, calculate the weighted value as the sum of each row's position multiplied by its impressions, divided by the sum of those impressions. Label it as the average for those exported rows. Do not assume a selected row set is the complete population behind the report headline.
Example: a worse aggregate without a worse query position
Illustrative arithmetic: query A has 90 impressions at average position 2, and query B has 10 at position 12. The combined weighted value is (90 × 2 + 10 × 12) ÷ 100 = 3. In a second hypothetical period, A has 50 impressions and B has 50, while their respective positions remain 2 and 12. The combined value becomes 7.
The aggregate moved from 3 to 7 even though neither query's position changed in this simplified example. The impression mix changed. These invented figures are a teaching calculation, not a reconstruction of a real Search Console total or evidence that a website lost rankings.
Separate stable queries from newly visible ones
Create a comparison view with the query, impressions and position for both periods. Mark rows observed in both periods separately from rows observed in only one. Investigate the stable set first when asking whether existing queries moved. Then examine how the volume and relevance of the other rows changed the overall picture.
Use 'not observed in this export' for an absent row rather than assigning it position zero or assuming it did not exist. For important findings, inspect the relevant query and landing page directly in the report. Keep device and country scope consistent before attributing the movement to content or technical changes.
Write a conclusion that survives a second reading
A useful conclusion distinguishes the aggregate observation from its decomposition: 'The overall average worsened; the reviewed stable queries were unchanged, while lower-position queries accounted for a larger share of observed impressions.' If the available rows do not support that explanation, leave it as a hypothesis and record what evidence is missing.
Read clicks and the relevance of the landing pages alongside position. Expanding into useful new queries and losing visibility on important existing queries require different responses. Keep the original export and your calculation so another reviewer can check the weights instead of reacting to a single red arrow on a dashboard.