Use a Search Console export when a site has many landing pages. A cohort is a documented set of comparable URLs; it is not proof that the URLs have identical search demand.
Normalize and label each URL
- Keep the source export unchanged. Add a working copy with normalized path, page template, intent group, device, country, and a note for redirects or parameter variants.
- Create rules before reading the trend: which URL variants merge, what counts as a new page, and how you handle missing or anonymized queries.
Compare like with like
- Compare equal-length windows and matching weekdays; exclude the latest incomplete days. Split branded/non-branded demand only when the query data supports that split.
- Check row counts, duplicate keys, and totals against the source export. Report cohort size and coverage; do not compare averages when denominators differ without showing the denominator.
Example and expected result
- Synthetic CSV (both URLs use template
blog):url,template,month,clicks,impressions;/blog/a,blog,Jan,12,240;/blog/b,blog,Jan,8,160;/blog/a,blog,Feb,9,180;/blog/b,blog,Feb,12,200. Pivot by template and month. January: 20/400 = 5.0% CTR; February: 21/380 = 5.5%. Compare equal, complete windows. - Expected result: one row per template and period, totals matching the filtered source export, plus URLs to inspect. Remove only exact duplicate source rows first; then aggregate the remaining data to the chosen grain. Keep distinct queries, countries and search types. Exclude partial dates and keep filters identical.