You filter the Performance report by a page URL you know ranks, and the chart is empty. Or you export every page row, join it to your CMS URL list, and a third of your pages have no match. Nothing is broken. Search Console keys page-level data to the URL Google selected as canonical for the duplicate cluster — not the URL the user requested, not the URL you declared in rel=canonical, and not necessarily a URL you would have chosen.
This single rule explains a surprising share of "missing data" tickets. It also quietly corrupts any analysis that joins Search Console rows to another dataset on URL equality.
Why a page that ranks can show zero impressions
Google groups URLs it considers duplicates into a cluster and picks one canonical URL to index and serve. Google's Search Console documentation states that performance data is assigned to the canonical URL. So when a user searches, sees your result, and clicks, the impression and click are recorded against the canonical — even if the URL that was crawled, the URL in the SERP, and the URL in the user's address bar differ from each other after redirects.
The consequence: a non-canonical variant has no performance data at all. Not low data. None. Filtering the Page report by https://example.com/shoes?color=red returns nothing, because every impression that URL ever helped generate was booked to https://example.com/shoes.
The reverse also happens. If Google picked a canonical you did not intend — a parameterized version, an uppercase variant, a print view — that URL appears in your Page report with real traffic, and your CMS has never heard of it.
How to find the URL Google actually chose
The URL Inspection API returns both canonicals for an inspected URL:
{
"inspectionResult": {
"indexStatusResult": {
"verdict": "PASS",
"coverageState": "Duplicate, Google chose different canonical than user",
"userCanonical": "https://example.com/shoes",
"googleCanonical": "https://example.com/shoes?color=red"
}
}
}
userCanonical is your declaration. googleCanonical is the selection that drives both indexing and reporting. When they disagree, your reporting join will fail for that URL.
Google documents a per-property quota on the URL Inspection API — on the order of a couple of thousand inspections per day, with a per-minute cap — so you cannot inspect a large site exhaustively every week. You can inspect a stratified sample per template, and you can inspect the full list of URLs that matter for revenue. Both are usually enough to build the mapping table described below.
What this breaks in day-to-day reporting
| Symptom | Likely cause | Check |
|---|---|---|
| A known-good page shows zero rows | Google canonicalized it into another URL | googleCanonical via URL Inspection |
| Page rows exist for URLs not in your sitemap | Google chose a variant as canonical | Compare Page report rows against sitemap URLs |
| Traffic "disappears" at migration cutover | New canonical sits outside this URL-prefix property | Switch to a Domain property |
| Country pages show implausibly lopsided data | Near-duplicate country variants folded into one canonical | Inspect two country URLs, compare canonicals |
| Folder-level totals don't match template logic | Canonical lives in a different folder than the variant | Group by canonical, not by requested path |
The property-type interaction deserves emphasis. A URL-prefix property only reports URLs under that exact prefix, including scheme and host. If Google canonicalizes http://example.com/x to https://www.example.com/x, the data lives in the https-www property and the other three properties show a clean, convincing, entirely artificial decline. A Domain property covers all subdomains and both schemes and sidesteps the split, which is the main reason to prefer it for measurement even if you keep prefix properties for team access.
The cross-domain case: your clicks are in someone else's account
Google supports cross-domain rel=canonical, and canonical-based attribution follows it. If you syndicate content and the syndicated copy carries a canonical pointing at your original, the performance data for that copy lands in your property. If you are the one republishing with a canonical pointing at the source, it lands in theirs, and you will measure approximately nothing for pages that genuinely drove clicks.
That is documented behavior for canonical selection plus documented attribution to the canonical. What is not guaranteed is that Google honors your cross-domain declaration at all — it is a hint, and Google frequently selects a different cluster member. So a syndication partner's missing data could mean the canonical was honored, or that Google clustered the pages differently than either of you expected. Inspect, don't assume.
Building a canonical map so joins stop silently failing
The fix is a mapping table maintained alongside your URL inventory:
- Enumerate your real URLs from the CMS or a crawl, not from Search Console.
- Inspect a prioritized subset — all money pages, plus a sample of 20–50 URLs per template — and store
googleCanonical,userCanonical, andcoverageStatewith the inspection date. - Join Search Console page rows on
googleCanonical, then roll up to whatever grouping you report on: template, folder, product family. - Flag mismatches as a monitored metric. The count of URLs where
userCanonical != googleCanonicalis a better canonicalization health signal than any crawler's "canonical issues" count, because it reflects Google's decision rather than your markup. - Re-inspect on change. Canonical selection moves when you alter internal linking, redirects, hreflang, or near-duplicate content. A map refreshed quarterly will drift.
If you use the Search Console bulk export to BigQuery, the url field in the page-level table is already the canonical, which means the join problem is identical — the export does not give you the requested URL. The advantage is that you can store the canonical map as a table and do the resolution in SQL instead of in a spreadsheet.
What to do this week
- Move primary measurement to a Domain property if you are still reading URL-prefix properties as if they were the whole site.
- Pull your top 200 landing pages by business value and run URL Inspection on each. Record
googleCanonical. - Any page where Google chose a different canonical: decide whether you care. If the chosen canonical serves the same content and ranks, leave it and fix your reporting. If it is a parameterized or thin variant, fix the internal links, redirects, and sitemap signals pointing at it.
- Rewrite any dashboard that joins GSC pages to CMS URLs so the join key is the canonical, and add a row count for unmatched URLs on both sides. An unmatched count that climbs is an early warning that clustering has shifted under you.