How should we set up hreflang for our Japanese version?
Use a subdirectory such as /ja/ on your existing domain unless you have a reason to run a separate ccTLD, make every language version point to every other version including itself, add an x-default for the version you want shown to everyone else, and keep each page's canonical pointing at itself rather than at the English original.
What to check first
- Does every version list all versions reciprocally, itself included
- Is there an x-default entry, pointing at the fallback version
- Is each Japanese page's canonical self-referencing, not pointing to the English page
- Are the language codes correct — ja for Japanese, not jp
The three mistakes we find on live sites
jp instead of ja. The language code for Japanese is ja; jp is the country. ja-JP is valid if you genuinely need to target Japanese speakers in Japan specifically, but a bare jp is simply ignored, which means the annotation you built does nothing at all.
A canonical pointing at the English page. This is the most damaging one, because it is self-inflicted and silent. If the Japanese page declares the English page as its canonical, you have told search engines the Japanese version is a duplicate not worth indexing. Every page’s canonical points at itself; hreflang is what expresses the relationship between languages.
Non-reciprocal annotation. If the English page lists the Japanese version but the Japanese page does not list the English one, the whole cluster is unreliable and may be discarded. Every version lists every version, each including itself.
URL structure, and one thing beyond hreflang
A subdirectory on the existing domain inherits your domain’s accumulated authority and is far less work to operate, which is why it is the default recommendation. A separate .jp or co.jp domain sends a stronger local signal and is worth it when the Japanese operation is substantially its own business with its own brand, not a translated section.
One point that hreflang does not cover: AI assistants answering in Japanese do not consume hreflang the way a search crawler does. They need each Japanese page to be complete, self-contained HTML that states plainly, in Japanese, what the company does and who it serves. A correctly annotated Japanese page that only renders in the browser is invisible to them. That is the part we cover together with localization on the localization and AI search page.