Clarks International SEO Migration - Struggling to Land on its Feet

This is one of the most instructive international migrations on public record, because the strategy was reasonable and the execution was not. Consolidating a set of country code top level domains into language subdirectories on one domain is a mainstream approach: it pools authority, simplifies platform maintenance and removes the need to build every market from a cold start. Clarks did exactly that in January 2020, moving four European country domains onto clarks.eu while simultaneously replatforming onto Salesforce Commerce Cloud. Within weeks the European organic footprint had largely evaporated.

The value of the case is that Sistrix documented the failure modes precisely enough to work backwards from symptom to cause. Every one of them is a check you can run on a staging environment before launch, in under an hour, with tools you already own.

What was measured

Sistrix tracks a Visibility Index per country database, so the four markets can be read independently. The recorded movement across the migration window:

Figures as reported by Sistrix. The final column is the share of ranking keywords whose ranking URL belonged to a market other than the one being queried, which is the fingerprint of a broken hreflang and canonical setup.

Failure one: locale sniffing instead of a redirect map

The dominant fault was that legacy country URLs were not mapped one to one onto their new equivalents. Instead the new platform applied browser locale detection at the edge, and Sistrix observed chains running to four or five hops that repeatedly terminated on the /en/ version rather than the market path.

This is fatal for a crawler. Googlebot crawls predominantly from United States infrastructure and does not send a meaningful Accept-Language preference, so a locale sniffing rule routes it to English every time. The Spanish and French subdirectories that were supposed to inherit twenty years of authority were, from Googlebot's perspective, barely reachable.

# What a locale sniffing chain looks like to a crawler
curl -sIL -A 'Mozilla/5.0 (compatible; Googlebot/2.1)' https://www.clarks.es/zapatos \
  | grep -Ei '^(HTTP|location)'

HTTP/2 301
location: https://www.clarks.eu/
HTTP/2 302
location: https://www.clarks.eu/en/
HTTP/2 302
location: https://www.clarks.eu/en/shoes

# What it should have been
HTTP/2 301
location: https://www.clarks.eu/es/zapatos

The correct pattern is a static rewrite map generated from a full pre migration crawl, applied before any personalisation layer runs. Locale detection, if you want it at all, belongs behind a client side banner that suggests a switch rather than a server redirect that forces one. Google's own internationalisation guidance has warned against automatic redirection on language for years, precisely because of this crawl effect.

Failure two: canonical loops and parameter sprawl

The Salesforce Commerce Cloud move introduced a second class of problem. Sistrix noted duplicate product URLs generated per colour variant, unmanaged parameters, and internal pages that pointed a canonical tag at a URL which then redirected back to itself. A canonical that resolves into a redirect loop is worse than no canonical: the crawler cannot resolve a preferred URL for the cluster at all, so it either picks one arbitrarily or drops the cluster.

Every canonical target on a migrated site should return 200 directly. This is a two line audit in any crawler, and it is the single highest yield post launch check there is.

# Screaming Frog: Reports > Canonicals > Canonicals Linking To Non 200
# Or against an exported list of canonical targets:
while read u; do
  code=$(curl -s -o /dev/null -w '%{http_code}' "$u")
  [ "$code" != "200" ] && echo "$code $u"
done < canonical_targets.txt

Failure three: hreflang pointing at non canonical URLs

Hreflang annotations only take effect when every URL in the cluster is self canonical and every member returns the annotation reciprocally. Sistrix found hreflang tags pointing at URLs that were not canonical, and pages missing self referential canonicals entirely. When that happens Google discards the cluster and falls back to picking whichever version it considers strongest, which is exactly the 61% to 68% cross market ranking figure in the table above.

The practical diagnostic is simple: if a large share of your Spanish queries are being answered by an English or French URL, the hreflang cluster is not being honoured. Check reciprocity, check self canonicals, and check that every annotated URL returns 200 rather than redirecting.

Failure four: backlink attrition

Sistrix recorded clarks.es losing 2,416 links across 362 referring domains and clarks.de losing 2,757 links from 1,075 referring domains. External links are the one asset you cannot rebuild on a schedule, and they are the reason a redirect map has to be exhaustive rather than approximate. A link pointing at a URL that now needs five hops to resolve passes a fraction of what it did, and a link pointing at a URL that resolves to a soft 404 passes nothing.

Before any migration, export the top thousand linked URLs from your backlink tool of choice and treat that list as a mandatory redirect target set, separate from and in addition to your crawl based map. Verify each one resolves in a single hop to a live, indexable page on the day of launch.

The pre launch checklist this case argues for

Run all six against staging, then re-run all six within two hours of go live. The Clarks case shows what happens when the second run is skipped.

What this case does not prove

It does not prove that ccTLDs outperform subdirectories, and it does not prove that Salesforce Commerce Cloud is bad for SEO. The confounding variables here are severe: a domain consolidation, a platform replatform and a URL structure change all shipped together. When three large changes land on the same day you lose the ability to attribute the outcome to any one of them, which is itself the migration lesson. Sequence large changes where the business will tolerate it, and instrument each phase separately.

Sistrix's own assessment was that the faults looked correctable and that recovery should not be especially difficult once the redirects and canonicals were fixed, though no significant recovery had been observed by the article's publication on 30 April 2020.

Related reading on this site

Two adjacent cases sharpen the picture. The Patient.co.uk domain migration shows what happens when the destination domain itself is the problem rather than the redirects, and the subdomain strategy analysis covers the architectural question of where content should live in the first place. For the crawl directives that decide whether any of this gets read at all, see the complete robots.txt reference.

FAQ

Source: https://www.sistrix.com/blog/clarks-international-seo-migration-to-run-in-head-first-or-to-land-on-your-feet/