usdigitalmarketingindex.com SEO
How can missing reciprocal hreflang references be detected?
Updated November 15, 2026
By extracting the alternates each page declares, laying them out as a table of who points to whom, and looking for pairs where the reference runs one way only. The rule being tested is short. Google’s documentation states: “If two pages don’t both point to each other, the tags will be ignored” (Google’s documentation on localized versions of pages). It also gives the reason: the return link stops another site from declaring itself an alternate of your page. So an annotation from page A to page B counts only if B names A back. Detection is a comparison of two lists, and it can be done by hand for a few pages or with a crawler for a site.
Extracting declared alternates
Annotations can live in three places, and the audit has to read whichever the site uses.
| Location | How to extract |
|---|---|
| HTML head | Fetch each page; read the link elements with rel alternate and an hreflang value |
| HTTP header | Fetch each URL; read the Link header |
| XML sitemap | Read each url entry and its alternate child entries |
For each page, record:
- the page’s own URL as fetched, and its final URL after any redirect;
- its HTTP status;
- its canonical tag;
- whether it carries
noindex; - every alternate it declares: the hreflang value and the target URL.
Read the served HTML, and also the rendered HTML if scripts insert the annotations. If a site uses more than one location, extract from each and compare, since the two can disagree.
The audit can only test alternates that are declared. Which pages belong in a sitemap’s alternate sets at all, including regional pages that are not translated yet, is settled in the guide to whether international XML sitemaps should include regional pages that have not yet been translated.
Detecting missing return links
A three-URL audit, with one missing edge placed deliberately. The site is hypothetical.
Declared alternates:
| Page | Declares en-GB | Declares en-US | Declares de-DE |
|---|---|---|---|
| A: /en-gb/pricing/ | A (self) | B | C |
| B: /en-us/pricing/ | A | B (self) | C |
| C: /de-de/preise/ | A | (none) | C (self) |
Reciprocity matrix, reading each pair in both directions:
| Pair | Forward | Return | Result |
|---|---|---|---|
| A and B | A names B | B names A | Reciprocal |
| A and C | A names C | C names A | Reciprocal |
| B and C | B names C | C does not name B | Missing return |
Page C omits the US version. Under the rule, the annotation between B and C is not valid, because C does not point back. The pairs A with B and A with C remain intact: the documentation notes that where a complete set cannot be maintained, “Google will still process the ones that point to each other”.
The same test, stated as a procedure:
- Build the list of all declared edges: source URL, hreflang value, target URL.
- For each edge from X to Y, look for an edge from Y to X.
- Any edge with no matching return is a missing reciprocal reference.
- Check also that each page declares itself. The documentation requires that “Each language version must list itself as well as all other language versions”.
Return links fail for reasons other than omission. The return exists in intent but does not match.
| Cause | Example |
|---|---|
| URL mismatch | A names <!–INLINECODE1–>; C's own URL is served without the trailing slash, or on a different host variant |
| Protocol mismatch | One side lists http, the other https |
| Redirecting target | A names a URL that redirects to C; C names A, and A's annotation points at the redirect |
| Non-canonical target | A names a URL whose canonical is a different URL |
| Target not indexable | C carries <!–INLINECODE2–>, or is blocked |
| Target missing | The named URL returns 404 |
| Template gap | A newer language was added to some templates and not others |
| Relative URL | The documentation requires fully qualified URLs |
So before declaring a cluster valid, verify each target as well as each edge: it returns 200 without redirecting, its canonical is itself, and it can be indexed.
Sources of evidence beyond a manual check:
- a site crawler with hreflang reporting, which produces the edge list and flags missing returns;
- the served sitemap, if annotations are there;
- spot checks with a browser’s page source.
Correcting language clusters
Fix at the source that generates the annotations.
| Finding | Correction |
|---|---|
| A version is missing from one page's set | Add it; the set should be identical on every version |
| URLs differ by slash, case, host, or protocol | Standardize on the canonical URL everywhere |
| A target redirects | Replace it with the final URL |
| A target is <!–INLINECODE3–> or removed | Remove it from every set, or restore the page |
| One template lacks a language | Update the template or the mapping table behind it |
| Sitemap and HTML disagree | Choose one location, or generate both from the same data |
| No self-reference | Add each page's own URL to its set |
The durable fix is a single mapping: one table that lists, for each piece of content, the URL of every language version. Every page’s annotations are generated from its row. A reciprocal error is then impossible within a row, and the remaining risk is a row with a wrong or dead URL.
For the example, the correction is one line added to page C:
<link rel="alternate" hreflang="en-US" href="https://www.example.com/en-us/pricing/" />
after which all three pages carry the same three annotations.
Revalidation
| Step | Method | Pass condition |
|---|---|---|
| Re-extract | Fetch all pages in the cluster again | New annotations served |
| Rebuild the matrix | Same procedure | Every pair reciprocal |
| Identical sets | Compare each page's set | The same on all versions |
| Target status | Request every listed URL | 200, no redirects |
| Canonicals | Read each | Self-referencing and equal to the listed URL |
| Indexability | Check robots rules and meta robots | Indexable |
| Caches | Fetch through the CDN, and bypassing it | Same result |
| Site-wide | Run the crawler across all templates | No missing returns reported |
| Outcome | Performance report by country over the following weeks | Each country's version receiving that country's impressions |
Schedule the audit to repeat. Clusters break when a page is deleted, a URL changes, or a language launches, and none of those events announces itself as an hreflang change. Running the check after each release, and whenever a language version is added or retired, catches the break while it is one page wide.
Valid reciprocity confirms that the annotations will be processed. It does not confirm that the codes themselves are the right ones for the markets intended, which is covered in the guide to how hreflang should distinguish English pages targeting different countries.
The test: for every annotation from one page to another, find the annotation coming back, to the same exact URL, from a page that is live, canonical, and indexable.
Add your comment to this answer