Canonical URL

It consolidates: search engines index the canonical URL and attribute the signals of the duplicates to it.

It does not block crawling. The duplicate URLs are still fetched.

It does not remove a URL from the index on its own, and it is not a substitute for a redirect. A URL that should not be reachable should redirect.

It is one signal among several. Internal links, sitemap entries and redirects all contribute, and a canonical that contradicts them is often ignored.

Absolute URLs. A relative canonical resolves against the current host, so each host declares itself canonical — which defeats the purpose on a site answering at several addresses.

One per page. Two canonical tags are ambiguous, and search engines may disregard both.

Point at a URL that returns 200. A canonical pointing at a redirect or a 404 is an explicit instruction to consolidate onto something that does not work.

Self-reference by default. Every page declaring its own URL as canonical is a cheap default that removes the tracking-parameter problem entirely.

Redirect when the duplicate address should not be reachable: an old path, a protocol variant, a host variant.

Canonical when the address must keep working but should not compete: a URL with a tracking parameter, a filtered view of a listing, a print version.

Both, in sequence, is common and correct: redirect the structural duplicates, canonicalise the functional ones.

`document.querySelector('link[rel=canonical]')?.href` in the console, or a search for `rel="canonical"` in the source.

Then test the variants: add a dummy parameter, toggle the trailing slash, try the other host. Each should either redirect or declare the same canonical.

A canonical may also be sent as an HTTP `Link` header, which is the usual mechanism for non-HTML resources such as PDFs.

Faceted navigation. A listing filtered by colour, size and price generates a combinatorial number of URLs, each with the same products in a different order. Canonicalising every combination to the unfiltered listing is the usual answer — except where a filtered view is genuinely a page people search for, which then wants to be canonical to itself.

Pagination. Page two of a listing is not a duplicate of page one, and canonicalising it to page one tells search engines to ignore everything it contains. Each page should be canonical to itself.

Templated canonicals. A template emitting the same canonical for every page in a section collapses the whole section onto one URL. This is rare, catastrophic, and easy to miss because each page looks correct in isolation.

Staging environments. A canonical hardcoded to the production domain is right in production and is a cross-domain signal from staging; one hardcoded to staging and shipped is the reverse, and considerably worse.

The pattern behind all four: a canonical is written once in a template and applied to thousands of pages, so an error in it is never a single-page error.