Plan H: AI Translation / Multilingual Export

ChatGPT / Claude

Collect brand names, product names, jargon and slogans into a source↔target glossary, ruling which stay untranslated and which get fixed translations. Consistency starts here.

💡Brand names usually stay as-is; feature words (e.g. 'save') get one canonical translation site-wide.

DeepSeek / ChatGPT / translation API

Feed 'glossary + style requirements + source' together, translating segment by segment. For bulk, use an API or script, giving each line context to avoid mis-splitting.

💡Give the model 1-2 polished reference sentences (few-shot) for much steadier style.

Native speaker / professional translator

Review cultural context, tone and marketing hooks line by line; reflow long sentences to target-language habits. Brand-voice consistency is won at this gate.

💡Give reviewers a scope ('tone / length / culture only') so style doesn't drift while editing.

Localization platform / dev handoff

Localize dates, currency, units and time zones; adapt truncation and wrapping caused by different text lengths, and restore placeholders to the right slots.

💡EN→ZH often grows ~30%, ZH→EN can shrink ~30%; leave flex in your layout.

Back-translation spot checks + UI screenshots

Back-translate samples to compare semantics; screenshot key pages for truncation, overlap and untranslated leftovers before launch.

💡Run an automated 'untranslated residue scan' before release — more reliable than eyeballing screens.