Removing x-default entirely: a deliberate negative test

Removing x-default entirely: a deliberate negative test

Hypothesis: x-default is often cargo-culted; we wanted to measure what actually breaks without it. Methodology: deliberate negative experiment. One property with a clean 8-market cluster. We removed x-default from half the clusters (~1,200 URLs) and kept it on the other half as control. 9 weeks. We pre-registered the prediction that unmatched-locale serving would degrade.

What we found:
— For matched locales (a French searcher in France, etc.), removing x-default changed nothing measurable — those are resolved by the specific language-region tags, not the default. As expected.
— For unmatched locales (e.g., a Portuguese searcher with no pt variant), the no-x-default group served a less predictable variant: Google fell back to its own guess, and correct-fallback consistency dropped from ~88% to ~64%.
— Clicks from unmatched-locale countries fell ~12% in the test group; the control held flat.

Caveats: 'unmatched locale' traffic was only ~8% of total here, so the sitewide impact was small. On a site whose audience closely matches its language coverage, removing x-default might be near-harmless — which is the uncomfortable nuance. The 88%-to-64% figure is the real finding; the click number is downstream and noisier.

Conclusion: x-default only matters for the long tail of unmatched locales. Its value scales with how mismatched your traffic is to your coverage — not a universal must-have, but cheap insurance.