Fixing return-tag reciprocity on 9,400 product URLs

Fixing return-tag reciprocity on 9,400 product URLs

Hypothesis: non-reciprocal hreflang (page A points to B, B doesn't point back) was silently invalidating the entire cluster, not just the broken edge. Methodology: one mid-size apparel retailer, 5 markets, 9,400 products, audited with a crawler that flagged 31% of rel=alternate entries as one-way. Before/after window: 12 weeks.

Findings:
— The broken edges clustered on newer SKUs where the localization pipeline added the source page before the target existed, leaving a dangling reference.
— After making all tags reciprocal (we generated return tags from a single source-of-truth table rather than per-template), GSC's 'no return tags' error count fell from ~2,900 to 41.
— The measurable outcome: correctly-canonicalized variants in non-English markets that previously showed the US page rose to first position for branded+localized queries in 3 of 5 markets within 8 weeks. Impressions in DE and FR rose 17% and 14%.

Caveats: we found no movement in ES or IT, and we don't know why — possibly weaker pre-existing authority, possibly the queries were already correctly served. The 'invalidates the whole cluster' framing is a working model, not something we proved; what we proved is that the errors disappeared and rankings followed in most markets.

We'd want a second site before generalizing.