hreflang: the ultimate guide

What hreflang actually does (and does not do)

hreflang is an annotation that groups alternate versions of the same content by language and, optionally, region. A typical head implementation looks like this:

<link rel="alternate" hreflang="en-gb" href="https://example.com/en-gb/pricing/" />
<link rel="alternate" hreflang="en-us" href="https://example.com/en-us/pricing/" />
<link rel="alternate" hreflang="fr" href="https://example.com/fr/tarifs/" />
<link rel="alternate" hreflang="x-default" href="https://example.com/pricing/" />

When the cluster is valid, Google keeps ranking whichever version has the strongest signals, then swaps in the version that matches the searcher's language and location. A user in Paris searching your brand sees the French URL instead of the English one that would otherwise rank. That swap is the entire feature.

Three things hreflang does not do, because I keep seeing these assumptions in briefs:

It does not boost rankings. If your French page has no authority, hreflang will not conjure any. It does not consolidate signals the way a canonical does; each version stays independently indexed. And it is not a directive. Google calls it a hint and will override it when the annotations contradict other signals, which is exactly why broken implementations fail silently.

The reciprocity requirement, drawn out

The rule that breaks the most implementations: hreflang only works when the pages agree with each other. If page A lists page B as an alternate, page B must list page A back. No return link, no swap. Google confirms return links as a hard requirement in its documentation on localized versions.

BROKEN CLUSTER: one-way annotation, Google ignores it /en/ (en) /fr/ (fr) lists /fr/ /fr/ has no return link to /en/ result: the pair is dropped, wrong version can rank in France Cluster size grows fast: 10 languages means 10 annotations on every page, 100 in total. Generate them, never hand-write them.

Three ways to implement it

Pick exactly one method per URL set. Mixing methods across the same pages is how you end up with contradictions that nobody can debug.

The rules that break most implementations

1. Return links. Covered above. Every pair must be bidirectional.

2. Self-reference. Each page includes an hreflang entry pointing at itself. Skipping it makes clusters flaky.

3. Valid codes. Language codes are ISO 639-1, region codes are ISO 3166-1 Alpha 2, in that order. The classics that burn people: en-UK is invalid (the United Kingdom is GB), en-EU is invalid (the EU is not a country), and region-only values like gb alone are invalid, language is mandatory. Google skips codes it cannot parse.

4. Absolute URLs. Full protocol and host, always. Relative hreflang URLs are a known source of quiet failures.

5. Point at real, indexable URLs. hreflang targets must return 200, be indexable, and be the canonical version of themselves. Annotations pointing at redirects, noindexed pages, or non-canonical URLs get dropped. Conflicts between hreflang and canonical tags deserve their own article, but the short version: every hreflang target should carry a self-referencing canonical.

6. x-default is a fallback, not decoration. It names the version for users who match none of your locales, usually a language selector or your global English page. It is optional but recommended, and it can coexist with a regular annotation for the same URL.

How to audit hreflang without going blind

You cannot QA hreflang by viewing source on five pages. Clusters break in the long tail, on paginated URLs, on pages your CMS treats differently, on the one locale where a plugin misfires.

Crawl with Screaming Frog. Enable hreflang in the crawl config, then work through the hreflang tab filters: missing return links, invalid language codes, non-canonical alternates, alternates that redirect or error. This one report finds 90 percent of real-world problems. Sitebulb's hreflang hints and Ahrefs' site audit do the same job if that is your stack.

Check Search Console for symptoms. Google removed the dedicated International Targeting report in 2022, so you now diagnose by side effects: run a site: query or performance report filtered by country and look for the wrong language version ranking in a market, or canonical reports showing Google picking a different language as canonical (a duplicate-content smell, usually same-language regional pages that are near-identical).

Spot-check the raw output. One page per locale per template type: view source (or curl for header implementations, or the sitemap entry) and read the actual annotations. Template bugs cluster by template.

What good looks like: every indexable page in a multilingual set carries a complete, reciprocal, self-referencing annotation set with valid codes, all targets return 200 and self-canonicalize, and country-filtered SERPs show the matching language version. Boring and symmetrical is the goal.

FAQ