Search Console query cohorts: compare existing, new, and missing queries
Separate queries observed in both periods from newly observed and missing rows. Use a worked example to explain changing clicks, CTR, and average position.
The queries and numbers below are fictional. Cohorts describe the supplied exports, not every search or a controlled experiment. A missing query row does not establish zero traffic or a lost ranking.
Ask whether the same queries changed
A site can gain clicks while its average position gets worse. One possible reason is that it starts appearing for additional queries at lower positions. Comparing the same observed queries across two periods helps distinguish movement within an established set from a change in the mix of visible rows.
A cohort is simply a defined group. Here, retained means observed in both exports; newly observed means present only in the later export; no longer observed means present only in the earlier export. Those labels concern reporting coverage. They do not tell you when Google first discovered a page or whether a search term stopped producing impressions.
Prepare a comparison that can be reproduced
Choose two complete, equal-length windows and keep the property, search type, country, device, and page filters the same. Save those settings with both exports. Avoid combining a sitewide Queries table with a page-filtered table, or silently changing the brand definition between periods.
Join the rows on the exact query string after documenting any normalization. Check for duplicate keys before joining. If your source includes query-and-page rows, either compare that same pair in both periods or deliberately aggregate to query level; do not merge different levels of detail. Preserve an explicit missing status when a key is absent.
Google documents omitted anonymized queries and truncated query tables. Keep the property totals alongside the row analysis, with their scope labeled. A union of the exported query rows is not a complete inventory of searches.
A four-query example
These two fictional exports use identical filters. Each cell reports clicks / impressions / average position. The dash means the row is absent from that export, not an observed zero.
| Query | Earlier period | Later period | Export cohort |
|---|---|---|---|
| seo audit checklist | 20 / 200 / 8 | 30 / 300 / 8 | Retained |
| redirect audit | 10 / 100 / 12 | 15 / 150 / 12 | Retained |
| technical seo worksheet | — | 5 / 1,000 / 40 | Newly observed |
| old audit template | 5 / 50 / 6 | — | No longer observed |
Calculate the totals and the retained cohort separately
The listed earlier rows total 35 clicks and 350 impressions: a 10% CTR. The listed later rows total 50 clicks and 1,450 impressions: approximately 3.45% CTR. The observed-row totals therefore show 15 more clicks while aggregate CTR falls sharply.
The retained cohort tells a different part of the story. Its clicks increase from 30 to 45, and its impressions from 300 to 450. Both rise 50%, leaving CTR at 10%. The two queries also keep the same row-level average positions. The newly observed query contributes 1,000 impressions at position 40 and a 0.5% CTR, changing the overall mix.
For illustration, weighting the listed row positions by their impressions gives about 8.86 earlier and 30.48 later. Within the retained cohort, the same calculation gives 9.33 in both periods. These are calculations over this teaching dataset, not a reconstruction of a real property's complete Search Console average. Do not average row percentages or positions without considering their weights and aggregation scope.
Turn each cohort into a different investigation
For the retained queries, inspect landing-page relevance and whether the additional exposure brings useful visits. For the newly observed query, check the actual landing page and search intent before treating its lower CTR as a problem. It may represent relevant discovery, or it may be a poor match; this table cannot choose between those explanations.
For the no-longer-observed query, inspect a longer window, the relevant page, and the original reporting filters. Do not write that five clicks were permanently lost simply because the row is absent. The comparison describes the visible tables, with uncertainty about omitted activity.
A defensible summary is: Listed clicks rose, the retained query set kept its CTR and positions, and a newly observed low-CTR query expanded the impression mix. That is more precise than saying the whole site lost rankings or that every title needs rewriting. It still does not establish why any change occurred.
Keep the analysis useful next month
Save the cohort definition, source settings, and raw exports with the report. Record a specific action, owner, and review date for each important finding. Reuse the same method in the next comparison, while allowing that the membership of each cohort will change.
RankCrab's existing ranking-report tool can inspect one Queries export at a time; it does not automatically join two exports into these cohorts. Use the downloadable worksheet for the comparison notes and your own spreadsheet or analysis process for the join. The public tools work without an account; the paid workspace remains in prelaunch.