Quality is not a single score. A polished sentence can contain the wrong claim, and a slightly awkward sentence can still communicate perfectly. Use this checklist to judge whether an AI translation is suitable for its actual purpose.
1. Completeness
- Every heading, paragraph, list and call to action is present.
- Buttons, form labels, captions and error messages are included.
- No source-language text remains unintentionally.
- Nothing has been duplicated or added without support from the source.
2. Meaning and facts
- The central meaning has not changed.
- Conditions, exceptions and negatives remain intact.
- Numbers, dates, names, currencies and units are correct.
- Relationships between products, people and actions are unambiguous.
- The translation has not invented explanations or benefits.
3. Terminology
- Product and feature names follow the approved form.
- Industry terms match the company glossary.
- The same concept is not translated several different ways without reason.
- Terms that must remain untranslated have been preserved.
4. Claims and risk
- Claims have not become stronger or more absolute.
- Qualifications and disclaimers remain connected to the relevant statement.
- Legal, safety, medical or financial wording has the required expert review.
- The localized version does not promise a capability the product lacks.
5. Brand voice
- The level of formality fits the brand and market.
- The relationship with the reader feels intentional.
- Sentence length, directness and energy are recognisable for the brand.
- The text avoids prohibited clichés, jargon or exaggerated language.
6. Market fit
- Currency, measurements, dates and contact conventions are appropriate.
- Examples and references make sense locally.
- Calls to action match what is available in that market.
- The wording fits the target market, not merely the target language.
7. Technical presentation
- Links point to the correct localized destination.
- Variables, placeholders and code fragments are intact.
- Text fits buttons, menus and other interface limits.
- Punctuation, capitalisation and formatting are consistent.
- The page renders correctly on relevant screen sizes.
8. Approval and traceability
- High-risk questions have an identified owner.
- Reviewer comments have been resolved rather than hidden.
- The final version and its approver are recorded.
- Useful terminology and corrections will be reused next time.
Using the checklist with Localizethat
Localizethat's Reviewing Assistant can surface findings about meaning, terminology, claims, voice and constraints. The checklist remains useful because a reviewer still has to decide what matters for the content, resolve the finding and approve the result. The tool supports the decision; it does not replace ownership of the decision.