When canonical and hreflang fought: a 4,000-URL conflict resolved

When canonical and hreflang fought: a 4,000-URL conflict resolved

Hypothesis: cross-language canonical tags (every localized page canonicalizing to the English original) were causing Google to drop the localized versions from the index entirely. Methodology: one platform, 4,000 localized URLs across 6 languages, where a CMS default pointed every rel=canonical at the source-language URL. 11-week observation.

Findings:
— Pre-fix index inspection on a 50-URL sample: 44 of 50 localized pages were 'Duplicate, Google chose different canonical' — the English page. Localized content was effectively invisible.
— We changed each page to self-canonical, kept hreflang reciprocal, and resubmitted. Within 3 weeks, 47 of the same 50 were 'Submitted and indexed.'
— Outcome: non-English indexed pages rose from ~2,300 to ~3,800; non-English clicks rose 73% over the window (off a suppressed base, so the percentage is flattering).

Caveats: the base was artificially low because of the bug, so a 73% recovery says more about how broken it was than how good the fix is. We can't claim a healthy site would see anything like this. Also, hreflang requires self-canonical to function — the original setup was self-contradictory, so this was correcting an error, not optimizing.

Conclusion: hreflang and cross-language canonicals are mutually exclusive. The data is a cautionary tale, not a growth tactic.