Why does Google merge your hreflang pages even with correct implementation?

Is hreflang really a priority signal for Google?

Yes, hreflang remains a strong signal in the geographic and linguistic targeting algorithm. Google actively uses it to determine which version of a page to display to a user based on their language and location. It is not just a consultative hint.

But "strong" does not mean "absolute." Google reserves the right to override your hreflang annotations if contradictory signals are detected. The most common case: two language versions have so similar content that the engine considers it pure duplication.

What happens when Google detects identical content despite hreflang?

Google then applies its own canonicalization process, merging the two URLs into a single reference version. It doesn’t matter if your hreflang tags are technically flawless: if the content is deemed identical, the engine will select a canonical URL and ignore the other in the SERPs.

In practice? You may notice that a French version never appears in France, being consistently replaced by the English or German version. Or worse: Google indexing a random URL, alternating between versions based on its successive crawls. This phenomenon creates SEO instability characteristic of poorly differentiated multilingual sites.

How does Google assess the similarity between two pages?

Google does not communicate the exact similarity threshold that triggers a merge. It is known that it compares the primary textual content, HTML structure, and likely extracted semantic entities. A simple word-for-word translation, especially between closely related languages (FR/ES/IT), can be detected as nearly identical.

Translated metadata (title, meta description) are not enough. The body text must show substantial differences: cultural adaptation, local examples, specific geographical references, rewritten sentence structure, variable length. An unrefined automated translation is a near-certain red flag.

• Hreflang is a strong signal but conditional, not an absolute directive
• Google will merge pages if their content is deemed essentially identical, even with correct hreflang
• Differentiation must focus on main content, not just on metadata or the menu
• Unrefined automated translations are particularly at risk of merging
• No official similarity threshold is communicated by Google, complicating evaluation

Does this statement align with real-world observations?

Absolutely. For years, sites with impeccable hreflang have still suffered from cannibalization between language versions. Audits regularly reveal cases where the .de version appears in France, or the .com version overwhelms all local variants. The problem is almost never the technical implementation of hreflang.

What Mueller confirms here is that Google prioritizes the coherence of its index over your technical directives. If the engine detects two nearly identical pages with different URLs, its anti-duplication reflex takes precedence over hreflang. This is logical from an algorithmic perspective: avoiding pollution of the index with duplicates.

What nuances should be applied to this recommendation?

Mueller says "unique content" but remains vague about the degree of differentiation needed. Is high-quality human translation enough? Is it necessary to rewrite 30% of the text? 50%? No official figures. [To be verified] on your own sites with A/B testing between more or less differentiated versions.

Another nuance: merging can be intermittent or partial. Google does not always treat all your URLs uniformly. You might have 80% of your multilingual pages working correctly, and 20% merging. This likely depends on the crawl budget allocated, update frequency, and internal signal consistency.

In what cases does this rule not fully apply?

For e-commerce sites with standardized product listings, substantial differentiation is nearly impossible. A Nike shoe remains a Nike shoe, whether sold in France or Belgium. The technical description doesn’t fundamentally change. Yet, these sites need separate versions to manage currencies, local stocks, and specific terms and conditions.

In these cases, Google seems to tolerate similarity more if other signals are consistent: ccTLD domain (.fr, .de), local address in the footer, local legal mentions, server geolocation. But this is never guaranteed. [To be verified]: no official documentation details these exceptions.

What steps should you take to avoid merging?

The first step: audit your multilingual pages with a content comparison tool. Compare the main text (excluding header/footer/navigation) between your FR, EN, DE versions, etc. If the similarity exceeds 70-80%, you are in the red zone. Tools like Copyscape, Siteliner, or even a basic diff can reveal the problem.

Next, enrich each version with specific local elements: examples rooted in local culture, references to country events or regulations, local customer testimonials, adapted use cases. Do not settle for word-for-word translations. Rewrite entire passages to create genuine semantic differentiation.

How can you check that Google is not merging your pages?

Use Search Console and filter by country/language. If your .fr version receives zero impressions in France while you have French organic traffic, that’s suspicious. Check which URL Google really serves with a VPN or a geographical simulation tool. Compare the canonical URL declared by Google (by inspecting the indexed page) with your hreflang.

Another test: search with the site: operator: target by domain/subdomain. If Google indexes only one version while you have three with hreflang, it means it has merged them. Also, check the server logs: if Googlebot repeatedly crawls only one version out of three, it has probably chosen it as the unique canonical.

What mistakes should you absolutely avoid?

Do not multiply language versions if you lack the resources to differentiate them. It is better to have one high-quality EN version than a site in 12 languages with identical automated translations. Google will penalize the latter approach by merging everything into a random URL.

Avoid also frequent changes to hreflang structure without content adjustment. If you fix your technical implementation without touching the underlying duplicate content, you will resolve nothing. The issue is rarely technical; it is editorial. Finally, do not rely on ccTLDs or subdomains to compensate for identical content: these signals help but do not replace textual differentiation.

• Audit textual similarity between language versions (max threshold ~70%)
• Enrich each version with specific cultural and geographical elements
• Check in Search Console that each version receives impressions in its target country
• Test with VPN/geosimulation which URL Google actually serves by region
• Compare the canonical URL chosen by Google with your hreflang directives
• Prioritize quality over quantity: better to have 3 truly differentiated versions than 10 identical automated translations