Morphos / Journal
Multilingual sites that do not compete with themselves
Adding languages without hreflang does not double your reach. It halves the credibility of both versions.
The usual multilingual build is a language toggle and a machine translation. It looks like access and behaves like a duplicate: two sets of URLs with near-identical content, competing against each other for the same queries, with no signal to search engines about which one belongs to whom.
The four rules
In order of how much damage skipping them does.
What has to be true
- Every language gets its own URLs, not a query parameter — `/fr/` rather than `?lang=fr`. A parameter is treated as the same page with a variation, so the two versions cannot rank separately.
- hreflang on every page, in both directions — each language points at the others and at itself. One-directional hreflang is a common, silent failure.
- Self-referencing canonicals — the French page canonically points at the French page. Pointing every language at the English original is the single most destructive multilingual mistake.
- Translated, not generated — machine translation for a hospitality site produces sentences that are technically correct and commercially wrong. It is visible to a native reader within one line.
What each choice costs
| Approach | Cost | What it produces |
|---|---|---|
| Language toggle, machine translation | Low | Two versions competing; native readers notice immediately. |
| Separate URLs, machine translation | Medium | Ranks, but converts poorly — the writing reads as foreign. |
| Separate URLs, human translation | High | Each market gets a page that reads as native, and ranks separately. |
| Transcreation (rewritten per market) | Highest | The version most international hospitality brands actually need. |
How many languages
The languages your guests actually book in — not the ones your industry assumes. A property can usually name four or five countries that produce most of its bookings. Starting there and adding on evidence beats launching eleven languages nobody has checked.
The part that actually breaks
The failure is rarely the translation. It is the plumbing around it: hreflang pointing at pages that no longer exist, a language switcher that drops the visitor on a homepage instead of the page they were reading, translated navigation with English URLs, and a sitemap that lists one language and hides the rest.
Each of those is a small thing that undoes the work of the translation. A visitor who switches to French and lands on the English homepage has been told, fairly clearly, that the French version is an afterthought.
The test before launch is simple: switch language on a deep page, then switch back, and see whether you end up where you started. If not, the plumbing is not finished.