What Is a Canonical Tag? Which Address Should Google Prefer?
If the same content opens at several addresses, Google treats one of them as the representative. The canonical tag states your preference. Google reads that preference as a strong signal. It may still choose another address. That choice does not open a penalty record.
The same page, several addresses
A duplicate address is not a violation that creates a manual action. The SEO article separates that from Google's starter guide. The cost is different. Crawling is also spent on the address you do not care about. Link and measurement signals scatter across several addresses instead of one. A visitor cannot tell which address is the real page.
Even if you state no preference, Google selects the version it judges best to show users. The document says most sites will do fine without a canonical. The reason to write one is so the choice is not left to chance.
Illustrative example
Suppose a workshop's measuring page opens at three addresses. This is not a client record. https://www.example.com/olcu is the real page. https://example.com/olcu is the copy without www. https://www.example.com/olcu?kaynak=bulten is the parameter a newsletter link adds. The preference is the first address.
If the old addresses can be retired, a redirect is the strongest path. If the parameter address has to keep working, its canonical tag points at the clean address. The clean address points at itself. Links from inside the site use the clean address.
Why a preference is written
The address shown in results
You want people to reach the measuring page from the plain address, not from the one with a parameter. The canonical tells Google that preference.
Gathering the signals
Links from other sites to the parameter address are gathered onto the preferred address if that address becomes canonical. Marks that sat apart join on one page.
Simpler measurement
If one piece of content has three addresses, reading it as a single row gets harder. One address collapses the row in the report as well.
Not wasting the crawl
You want Googlebot's time on a new or updated page rather than on crawling the copy again. That is not a promise of speed. It is a preference about the order of crawling.
How strong the signal is
Google orders the methods by how strongly they can influence canonical selection. Used together, they stack, and the chance that your preferred address appears in results goes up. None of them is required.
Do not name conflicting addresses. If the sitemap lists one address and the rel="canonical" tag names another, the signal splits. Giving a different address in the HTML tag and in the HTTP header is the mistake the document says to avoid. Pick one of the two. The file's own rules are in the sitemap article. Only the strength of the signal sits here.
| Method | How strong | When |
|---|---|---|
| Redirect | Strong. It says the target should be canonical. Permanent redirects have the same effect on Google Search. The time until they are noticed can differ by method. The fastest is an HTTP redirect on the server. | When retiring a duplicate address. Not for deleting a parameter that still has to work. |
| rel="canonical" link | Strong. A link element in HTML, or an HTTP header. Using both is supported, and it causes mistakes. | When the content stays and one address should represent it. The element on an HTML page, the header on a file such as a PDF. |
| Sitemap | Weak. Addresses in the list are canonical candidates. Google decides which page is a duplicate by looking at how similar the content is. | Keeping the list plain on a large site. A sitemap alone is not as strong as the tag. |
How the tag is written
The rel="canonical" link element sits in the head of the page. If the head is not valid HTML, Google does not accept the element. Use an absolute address. Google supports a relative address and does not recommend it. If a test site is crawled by mistake, a relative address can tie the preference to that test address.
The same element also sits on the preferred page itself. That is called a self-referential canonical. The document recommends it. The duplicate pages point at the same absolute address.
Do not add hreflang, lang, media or type on that element. rel="canonical" annotations that carry those attributes are not used for canonicalization. A language version is a separate link element, written with rel="alternate" and hreflang. That cluster is in the hreflang article.
If you add it with JavaScript, the cleanest way is to write the address in the HTML source and not let the script change the element later. If you cannot write it in the source, do not put the element in the source at all. Set it only with the script. Two sources at once make it unclear which one holds.
For a non-HTML file such as PDF or Word, an HTTP header is used. Google writes that it supports this for web search results only. The header also carries an absolute address. The shape is: Link: <https://www.example.com/olcu.pdf>; rel="canonical"
Paths that look like a canonical and are not
robots.txt
It is not used to choose a canonical. Google may index a URL disallowed in robots.txt without seeing its content.
The removals tool
It hides every version of an address from search. It does not pick a representative.
noindex
Not recommended for choosing a canonical inside one site. It removes the page from search entirely. The preferred fix is a rel="canonical" tag.
The fragment after #
Google generally does not support URL fragments. A fragment is not written as a canonical.
The address Google selected
A preference is not a promise that the address in results will be that one. URL Inspection shows two fields side by side. The canonical you declared, and the canonical Google selected. The second field is read only from indexed data. The live test does not predict it. The value can lag the index by a few hours. The inspection itself is in the Search Console article.
The address Google selected does not have to be the one shown in results either. If desktop and mobile have separate addresses, Google may show the one that fits the user's device.
If the choice is not the one you expected, look first for a conflicting signal. Do internal links go to the duplicate address. Does the sitemap list another address. Do the HTTP version and the HTTPS version point at each other. The tag lives in the site and the software. A tag that looks right in the template and is missing from the live head does not count as fixed in the panel.
HTTPS and the language cluster
Even without an explicit tag, Google reads two things in favor of a canonical. First, it prefers HTTPS over HTTP on an equivalent page. The document lists the exceptions. The preference breaks if the HTTPS page has an invalid certificate, if its dependencies other than images are insecure, if it redirects the user to HTTP, or if its canonical tag points at the HTTP address. A bad certificate and a redirect from HTTPS to HTTP lead Google to prefer HTTP strongly. HSTS cannot override that strong preference. Do not write the HTTP version into the sitemap or the hreflang cluster in place of HTTPS.
Second, addresses that point at each other inside an hreflang cluster are preferred. If two language addresses point at each other and a third is missing from the cluster, the ones in the cluster move ahead as canonical candidates. The canonical of a language version should be the page in the same language. If there is no page in that language, the closest substitute language is chosen, as the document puts it. Pointing every language at a single Turkish address with canonical does not fit the cluster.
Sources
- How to specify a canonical URL with rel="canonical" and other methods
Accessed October 20, 2026
- URL Inspection tool
Accessed October 20, 2026