Machine translation engines provide the first complete target-language draft. They are efficient for high-volume website updates and establish a consistent baseline across every changed string.
AI polishing is a separate, optional step. It receives the source, machine translation, locale, glossary and nearby context, then improves fluency while preserving protected terms, variables and markup. Keeping the stages separate makes fallback routing and cost reporting much clearer.
Automated quality checks should run after polishing. They verify placeholders, tags, numbers, terminology, length and language detection. Human reviewers then focus on flagged or commercially important content instead of reading every unchanged string.