What Is a Japanese Localization QA Score — and How to Read It

▶ 60-Second Slide Summary Prefer the 10-slide version of this article? Tap to read in 60 seconds. View slides →

When a foreign SaaS team asks whether their Japanese localization is good, they usually get an opinion back: "it's fine," "it reads okay," "a native speaker checked it." None of those answers can be compared, tracked, or acted on. They aren't measurements.

A Japanese localization QA score solves that. It's a 0–100 composite that breaks Japanese content quality into five measurable dimensions — so instead of "it's fine," you get a number, a per-category breakdown, and a clear picture of what to fix first.

This article explains what the score measures, how to read each range, and how to use it.

Why a Score Beats an Opinion

The core problem with Japanese localization is that the people who commissioned it usually can't evaluate it themselves. That makes quality a black box: you can't manage what you can't see. A vague "looks good" from one reviewer and a vague "feels off" from another can't be reconciled, because neither is anchored to anything.

A score fixes three things at once. It makes quality comparable — page A scores 71, page B scores 88. It makes quality trackable — this page was 64 before QA, 91 after. And it makes quality prioritizable — the terminology category scored 45, so that's where the work goes first.

It also changes who can participate in the conversation. When quality is an opinion, only the person who reads Japanese has a vote. When quality is a score with a category breakdown, the product manager, the head of growth, and the person approving the budget can all reason about it together — because the score is a shared, legible object, not a private judgment.

The Five Things a Japanese QA Score Measures

A meaningful score is not one number pulled from intuition. It is a weighted composite of five categories, each scored independently:

1. Fluency & Naturalness

Does the Japanese read like a native speaker wrote it, or like a translation of English? This catches literal sentence structures, doubled modifiers, and phrasing that is grammatically correct but no Japanese professional would actually use.

2. Terminology Consistency

Is the same feature, action, and concept named the same way across the UI, help center, and marketing site? Terminology drift is invisible to a casual read but obvious to a daily user — and it is one of the most common causes of a low score.

3. Register Appropriateness

Is the politeness level correct for the context — and consistent across the product? A SaaS product can choose a polite or a plainer register, but mixing honorific, plain, and imperative forms across screens reads as disorganized.

4. Trust Signals

Are the market-specific signals correct: payment terminology (決済 vs 支払い), tax notation, the 特定商取引法 legal disclosure, formal billing language? In FinTech especially, errors here do direct damage to conversion.

5. UI & Format Integrity

Does the Japanese text fit the components it lives in? This covers truncated buttons, overflowing labels, Western punctuation carried over, and English strings left untranslated in tooltips, empty states, and emails.

A page is scored on all five categories regardless of type, but the same raw issue can land in different categories depending on context — an English string is a UI & Format problem on a dashboard and a Trust Signal problem on a checkout page. This is why the category breakdown matters more than the headline number.

How to Read the Score Ranges

The composite score maps to four bands, each with a different business meaning:

• 90–100 — Native quality. Reads as built in Japan. No trust damage; only stylistic refinements remain.
• 75–89 — Solid. Professional and trustworthy, with minor polish needed. Safe to ship; worth a light pass.
• 60–74 — At risk. Understandable, but it signals "translated, not localized." This is the range that quietly drags conversion — the most common score for AI-translated SaaS content.
• Below 60 — Actively harmful. Contains trust-breaking errors: wrong payment terms, mixed-language UI, register that reads as careless. Japanese enterprise buyers notice, and leave.

The most important threshold is 60–74. Content in this band passes a casual check — it is not "wrong" — which is exactly why it survives. But it is also where measurable conversion loss happens, because it is good enough to ship and not good enough to trust.

It is also worth being clear about what the bands do not mean. A 91 is not a license to stop — it means the remaining work is refinement rather than repair. And a 55 is not a catastrophe — it is a clear, early signal, caught before it cost you a quarter of conversions. The score's value is the same at every level: it tells you where you actually stand, in a way an opinion never can.

A Worked Example: Reading One Page's Breakdown

Consider a Japanese pricing page that comes back with a composite score of 68 — squarely in the "at risk" band. The breakdown tells the real story:

A single number — 68 — would have sent the team to re-translate the page. The breakdown shows that re-translation would be wasted effort: the prose is fine. The actual problem is terminology and trust signals. The fix is a glossary pass and four specific corrections, not a rewrite. That is what a score does — it tells you not just that there is a problem, but where.

This is also why a single number can be actively misleading without its breakdown. Two pages can both score 68 and need completely different work — one a terminology cleanup, the other a full re-translation. The composite tells you there is a problem worth attention; only the category scores tell you what kind of problem it is, and therefore what the fix costs.

What the Score Is Not

A QA score is a diagnostic tool, not a grade on the translator. A low score on AI-translated content is expected — it is the starting point, not a verdict. The score's job is to make the gap visible and specific so it can be closed efficiently.

It is also not a substitute for knowing what the page is for. A 75 on a blog post and a 75 on a checkout flow are not equivalent risks. Always read the composite together with the category breakdown and the page's commercial role.

And it is not a one-time certificate. A page that scores 92 today can drift back into the 70s within two product cycles, as new features add new strings that were never checked against the glossary. The score is a snapshot, not a permanent state — which is exactly why the most useful way to use it is repeatedly.

How to Use the Score

In practice, the score works best as a loop: measure your most important page first, fix the lowest-scoring categories, re-measure to confirm the lift, then extend the same review to the next page. Over time the score becomes a shared language — the localization team, the product team, and the people approving the budget can all point at the same number.

The teams that get the most from scoring treat it the way they treat any other product metric: they baseline it, set a target, and review it on a cadence. A score that is measured once and filed away is just a report. A score that is tracked is a quality system.

Next Steps

A Japanese Website Mini Audit produces exactly this: a 0–100 composite score for one page, a breakdown across all five categories, before/after examples, and a prioritized fix list — delivered within 3–5 business days.

Frequently Asked Questions