How does Google determine the canonical URL among a cluster of similar pages?

What is a URL cluster and why does Google need to choose just one?

When multiple pages on your site (or other sites) have nearly identical content, Google groups them into a cluster. This occurs with HTTP/HTTPS variants, URLs with or without www, tracking parameters, separate mobile versions, or poorly configured paginated pages.

The engine will not index all these variants. It selects a representative URL (canonical) that will be displayed in search results. The other URLs in the cluster are grouped under this main URL — consolidating ranking signals and preventing dilution.

What signals does the algorithm consider for this choice?

Google mentions a machine learning system that analyzes various signals. Site security is mentioned first: HTTPS is favored over HTTP. Secure dependencies (likely external resources loaded over HTTPS) also play a role.

The "configurability of the page" is less clear — it likely refers to the stability of the URL, the absence of chain redirects, the cleanliness of parameters, and the consistency of canonical tags. The ultimate goal remains clear: to avoid sending users to a poor experience (broken, slow, or insecure pages).

Does this logic also apply to duplicated pages across different sites?

Yes, without a doubt. When content is syndicated or copied across multiple domains, Google forms an inter-domain cluster. It then chooses the original source or the most authoritative version based on signals like content age, domain popularity, and backlinks pointing to each version.

This is why a site that scrapes your content won't necessarily steal your rankings — unless its authority far surpasses yours or if your own site exhibits failing technical signals.

• Google groups similar URLs into clusters and selects a representative URL
• The algorithm favors HTTPS, secure dependencies, and pages that provide a good experience
• Your canonical tags are recommendations, not absolute directives
• Clusters can form between different domains (syndication, scraping)
• Technical stability and security are decisive criteria in this choice

Does this statement align with what we observe in the field?

Overall, yes. It has been known for years that Google does not always respect the canonicals we indicate. There are numerous scenarios: a mobile AMP version selected while pointing to the desktop, a URL with parameters chosen instead of the clean version, or even an HTTP page indexed despite the redirect to HTTPS.

Machine learning explains this autonomy: Google makes its own calculations and sometimes determines that your choice is not optimal. The issue is that this logic remains a black box. We do not know precisely how much weight each signal carries, nor how the algorithm arbitrates between an explicit canonical tag and its own preferences.

What grey areas persist in this explanation?

The notion of "configurability of the page" remains terribly vague. Does it include the presence of coherent hreflang tags? The structure of URLs (with or without a trailing slash)? Loading speed? The presence of poorly managed dynamic content? [To verify] — Google provides no exploitable details.

Similarly, no hierarchy among the signals is specified. If an HTTPS page is slow and poorly configured, and an HTTP version is fast and clean, which one prevails? We assume that security takes precedence, but without certainty. This opacity complicates audits when Google ignores your canonical directives.

In what cases does this logic pose problems for SEOs?

The first problematic case concerns multilingual or multi-regional sites. If Google arbitrarily decides that a .com version is more relevant than a .fr version for a French user, you lose control over user experience. Hreflang tags are meant to manage this, but if the clustering algorithm ignores them, you are stuck.

The second case: migrating from HTTP to HTTPS. Even with perfect 301 redirects and canonicals pointing to HTTPS, some sites see Google continue to index HTTP URLs for weeks. Machine learning can be slow to re-evaluate a long-established cluster.

How can you ensure Google chooses the right canonical URL?

The first rule is absolute consistency among all your signals. Your canonical tag, 301 redirects, XML sitemap, and internal links must all point to the same URL version. If your canonical states HTTPS but your internal links point to HTTP, Google receives conflicting signals.

Next, secure your entire site. Switch to HTTPS everywhere, including external resources (images, scripts, CSS). An HTTPS page that loads HTTP dependencies sends a mixed security signal that Google may penalize in its choice of canonical.

What mistakes should you absolutely avoid?

Never allow multiple accessible versions of the same page to coexist. If you have migrated to HTTPS, all HTTP URLs must 301 redirect to HTTPS. No duplicated content should be accessible via both protocols.

Avoid chains of redirects. If A redirects to B which redirects to C, Google might choose B as canonical instead of C. Redirect A directly to C. Likewise, do not place a canonical tag on a page that is itself a redirect — this sends an inconsistent signal.

How to check if your canonical URLs are being respected?

Use Google Search Console to identify indexed URLs versus those that you have declared as canonical. The "URL Inspection" tool tells you which URL Google considers representative and why it made that choice.

Also, monitor your server logs. If Googlebot continues to crawl URLs that you thought were consolidated, then the clustering is not functioning as intended. This may reveal orphaned internal links or poorly cleaned sitemaps.

• Ensure that HTTPS is enabled throughout your site and its external resources
• Verify that canonical, 301 redirects, sitemap, and internal links point to the same URL version
• Eliminate all chains of redirects and temporary redirects (302)
• Check in Search Console which URL Google has selected as representative
• Analyze logs to spot outdated URLs still being crawled by Googlebot
• Clean your XML sitemap of all non-canonical URLs