"Duplicate, Google chose a different canonical" — diagnose programmatic near-duplicates

In Search Console's Page indexing report, a chunk of one template lands in the "Duplicate, Google chose a different canonical than user" bucket — sometimes a few hundred URLs, sometimes most of a template family. Run URL Inspection on an affected page and the "User-declared canonical" and "Google-selected canonical" fields disagree: you pointed the page at itself, Google points it at a sibling that shares ~90% of its content. Critically, the page is usually still indexed — just under the sibling's URL, not yours — so traffic does not always fall off a cliff. What you lose is control: the page you optimized, linked to, and submitted in your sitemap is not the one ranking, and impressions quietly consolidate onto whichever sibling Google picked. The pattern is strongest on entity-swap templates where only a city, product, or persona noun changes between URLs, and it grows each crawl cycle as Google discovers more cluster members. First-time operators read it as a bug in their canonical tags; it is almost always a content-differentiation problem the tag cannot fix.

Entity-swap templates that produce near-identical bodies

When two URLs differ only by a swapped {city}, {product}, or {role} noun and share the same sentence frames, headings, intro, and CTA, Google's deduplication treats them as one page wearing different labels. Your self-referential canonical is only a hint; the near-duplicate body is the stronger signal, so Google picks one representative URL for the cluster and discards the rest. The fix lives in the content, not the tag.

Canonical tags are hints, overridden by ~40 site-wide signals

Google does not treat rel=canonical as a command. It weighs roughly forty signals — internal-link distribution, sitemap inclusion, redirect targets, hreflang clusters, HTTPS, URL tidiness, and content similarity among them — to choose a canonical. If your internal links, sitemap, or hreflang point at a different member of the cluster than your canonical tag does, you have sent conflicting instructions and Google resolves the conflict by trusting the aggregate, not your tag.

Inconsistent canonical signals across the same cluster

A page can self-canonicalize while its breadcrumb, related-links module, pagination, and parameterized variants all link to a slightly different URL of the same content. Faceted filters, trailing-slash and uppercase variants, and tracking parameters multiply the cluster. When the signals disagree about which URL is the real one, Google makes the call for you — and it rarely picks the one you intended, because your intended URL is the least-linked of the bunch.

Thin per-page differentiation below Google's uniqueness floor

Templates where the unique slot is a short data table or a two-sentence blurb wrapped in heavy shared boilerplate fall below the differentiation threshold Google uses to keep pages distinct. The pages may each contain unique facts, but the proportion of unique-to-boilerplate text is too low to register as separate documents. The cluster collapses to its most-linked or most-complete member, and the rest become "duplicates" of it.

1. 1

   Open Search Console → Page indexing → "Duplicate, Google chose a different canonical" and export the URL list. Segment by URL prefix to confirm whether one template is the source or the issue is scattered across the site.

2. 2

   Run URL Inspection on five affected URLs and record both the "User-declared canonical" and "Google-selected canonical" for each — the Google-selected URL is the representative of the near-duplicate cluster you are competing against.

3. 3

   Run pseolint against your sitemap and read the near-duplicate findings first: they identify which URLs share enough body content to be clustered, and which sibling each one collapses into, so you can map clusters before touching anything.

4. 4

   Read the boilerplate-ratio finding for the affected template — a high shared-boilerplate-to-unique-content ratio is the mechanical cause of the clustering, and it tells you how much genuinely unique content each page is missing.

5. 5

   Check the template-diversity finding to see whether your headings, intros, and section structure are identical across the cluster; identical structure plus a swapped noun is the exact fingerprint that defeats a self-referential canonical.

6. 6

   Audit signal consistency for one cluster end to end: confirm the canonical tag, the internal links pointing at the page, the sitemap entry, and any hreflang all reference the same URL — any disagreement is a vote you are casting against your own intended canonical.

7. 7

   Decide per cluster whether to differentiate (make each page genuinely unique) or to consolidate on purpose (pick one canonical, 301 or canonicalize the rest, and stop fighting to index pages that should not exist separately).

pseolint findings most strongly correlated with this pattern.

A home-services marketplace published 9,400 "{service} in {city}" pages from one template — same 1,100-word boilerplate, a swapped city name, and a three-row local price table per page. Within two crawl cycles, 6,700 landed in "Duplicate, Google chose a different canonical than user"; URL Inspection showed Google had folded each city page into the largest-metro version of the same service. Nothing was penalized — the service hub pages still ranked — but the long-tail city traffic the template was built for never materialized. The team split the work: for the top 400 cities by demand they replaced the boilerplate with genuinely local content (named providers, city-specific permit rules, real per-city price ranges from booking data), and for the remaining 6,300 they deliberately consolidated to service-level pages and canonicalized the thin variants away. Indexed-as-declared URLs on the rewritten set rose from 12% to 71% within about six weeks.

There is no manual-action clock here because there is no manual action — recovery is gated entirely by Google re-crawling and re-clustering your pages, which moves at crawl-budget pace. For pages you genuinely differentiate, expect Google to re-evaluate over the next one to three crawl cycles (roughly two to six weeks for a healthy host) and to start honoring your declared canonical once the bodies are distinct enough to read as separate documents; the indexed-as-declared ratio in Search Console is your leading indicator. For pages you consolidate on purpose, cleanup is faster and more predictable: once 301s or aligned canonicals are in place, the duplicate cluster shrinks within a crawl cycle or two and the surviving URL absorbs the impressions. Do not expect the "Duplicate" bucket to empty overnight — it drains URL by URL — and resist re-requesting indexing on individual pages, since the pace at which Google re-includes your differentiated pages is itself the quality verdict you are waiting on.