Translate a large site in controlled stages

Use staged generation to keep source extraction, POM AI credits, local quotas, review capacity, and public discovery aligned.

Define a release unit

Choose one target locale and one coherent surface, such as:

1. global navigation and registered strings;
2. top landing pages;
3. their terms and localized slugs;
4. one post type or catalog section;
5. secondary archives and long-tail content.

Do not begin with every locale and every object.

Prepare the batch

1. Back up the database and export existing translations.
2. Finish the source content and configure all eligible fields.
3. Exclude objects or locales that must not enter the launch.
4. Set realistic daily word and string limits.
5. Confirm the POM AI license, domain authorization, credits, and background processing.
6. Generate missing source strings before requesting translations.

Generate and review

Start the resumable generation job and let its persisted cursor, queue, bounded batches, retry metadata, and recovery worker continue the workload. Do not start competing manual and scheduled processors.

Review mt results for terminology, placeholders, links, sensitive claims, and layout. Promote only approved text to the reviewed workflow. If processing pauses, inspect the visible status and resolve authorization, quota, provider, or invalid-source conditions before resuming.

Publish the stage

Translate slugs and navigation, verify canonical/hreflang and sitemap behavior, clear caches, and test logged-out routes. Keep unfinished objects excluded or manage their robots policy deliberately. Record completion, remaining POM AI credits, and the next release unit before expanding the batch.