Canonical tag

On most websites the same content is reachable under more than one address without anyone noticing. With www in front and without. With https and with http. With a trailing slash and without. From those 3 decisions alone, one page becomes 8 different addresses that all show the same text.

To the visitor that makes no difference; to a search engine it does. It has to decide which of these addresses it takes into its directory and assesses. If it makes that decision itself, the standing the page has built up may end up spread across several variants — and every single one of them then looks weaker than the page actually is.

The canonical tag is the line with which you take that decision in advance. It sits in the head section of the source code and names the full address that is meant to count. Important here: Google treats this statement expressly as a pointer, not as an instruction. If other signals contradict it — internal links, redirects, the sitemap — the search engine may decide differently.

The pointer is needed most often in three situations. With attachments to the address, such as campaign tags from a newsletter or session numbers that make the same page appear under ever new addresses. With filter and sorting pages in directories and shops. And with texts that deliberately sit in two places, in a blog and in a topic overview, for instance.

Set wrongly, the same line does considerable damage, and it does so silently. If every page of a website points to the homepage — a fault some systems come with out of the box — the website declares its own subpages to be copies and thereby takes away its own visibility for everything except the homepage. Nothing about this is visible: the pages load normally, only no visitors arrive any more.

Just as widespread is the confusion with a redirect. A permanent redirect actually sends the visitor to a different address; they end up there and the old one is no longer reachable. The canonical tag leaves both addresses reachable and only says which one should be counted. If content really has moved, the redirect belongs there, not the pointer — otherwise you end up with a 404 error or a contradictory signal after all.

On multilingual websites one more rule applies that is often broken: each language version is a page in its own right and points to itself, not to the German version. Anyone canonicalising every language onto one takes all the translations out of the index. The way the language versions are related to one another happens through a different statement and does not replace the self-reference.

This can be checked without effort, but not in the file: what is delivered is what counts. We fetch every address individually and read out which canonical address actually appears in the result — and in the Google Search Console it says alongside which address Google itself chose. Where the two differ, that is the finding to be dealt with first.