What does a canonical tag look like?
One link element with the full URL of the version you want:
<link rel="canonical" href="https://www.example.com/blue-widgets">
It only counts in the <head>, and only once per page: with two, Google ignores both, often because an SEO plugin adds its own. A PDF has no head, so it gets the same line as an HTTP header.
In return you get a say in which URL people see in Google, the links to the copies count for one page, and Google crawls the duplicates less often. Google is honest about the stakes, though: a site "will likely do just fine" without one.
Does Google always follow it?
No. Google weighs your canonical against its other signals and picks the URL itself. A redirect and the canonical tag are each a strong signal, a sitemap a weak one. An https URL beats its http twin, and your internal links count too. When they all point the same way, they add up.
So Google sets your canonical aside when the signals disagree, say the sitemap lists one URL and the canonical names another. It also drops it when the pages aren't really copies, or when the canonical points at a page that doesn't exist or has noindex. A language version should point at a page in its own language.
When does a canonical do more harm than good?
When it points away from a page you want in Google, because a canonical asks Google to show the other URL instead.
- Page 2 of a list pointing to page 1. Whatever is only on page 2 and later stops being indexed, and Google puts it plainly: "Don't use the first page of a paginated sequence as the canonical page".
- A category page pointing to its featured article. The category drops out of the results. Let it point to itself.
- Thousands of filter URLs. For a shop's endless filter combinations a canonical only "may, over time, decrease the crawl volume"; to keep crawlers out, Google rates a robots.txt disallow higher.
- noindex or robots.txt instead of a canonical. A blocked URL can still be indexed without its content, and noindex takes the page out of Search completely.
- A partner's copy of your article. Here Google doesn't recommend a canonical: "The most effective solution is for partners to block indexing of your content".
Where the copies come from in the first place: duplicate content.
How do I see which canonical Google chose?
URL Inspection in Search Console shows yours and Google's side by side. Paste the URL into the bar at the top, open Page indexing and compare "User-declared canonical" with "Google-selected canonical".

My own site, corewebvitals.io, in Search Console.
The Page indexing report in Search Console (Indexing > Pages) has three statuses about canonicals:
| Status | What Google means | Do something? |
|---|---|---|
| Alternate page with proper canonical tag | Points to its indexed canonical (AMP, mobile or desktop pairs) | No, "there is nothing you need to do" |
| Duplicate without user-selected canonical | A copy without a canonical, so Google picked another URL | Only if Google picked the wrong one: add a canonical |
| Duplicate, Google chose different canonical than user | You declared this URL; Google prefers another | Find the disagreeing signal in URL Inspection |
For a whole site, or forty client sites, you need a crawl that reads every canonical. The SEO audit in MarketingTracer flags pages with "No canonical" and pages whose canonical points to another URL, with the target. It shows the canonicals you set; Google's choice stays in URL Inspection.
Does every page need a canonical to itself?
Yes. Since July 2026 Google's documentation says it outright: "Do include a rel="canonical" link on the canonical page itself." It covers the variants nobody made on purpose, such as a shared link with a tracking parameter: the variant inherits the tag and points back at the clean URL.
On corewebvitals.io, a site I run, the buttons link to the consultancy page with parameters such as ?cta=contact-me. Each of those URLs has a canonical to the clean page in its own language. The SEO audit lists 12 pages whose canonical points elsewhere, these among them. In Search Console, 10 of the page's 8,819 impressions (July 1, 2025 to October 26, 2026, all languages) went to parameter URLs: two Japanese variants. None came from July 28 on.
I want to know which button brings in the requests, so the parameter stays and the canonical does the rest. A shop with thousands of filter URLs is another story, though: I wouldn't do it like that there. Google's advice is still to link to the canonical URL rather than a duplicate URL inside your own site.
Most CMSs and SEO plugins print the self-referencing canonical for you. Check that it prints the clean URL and drops the parameters.