Estimate a run before starting

POM Translate does not show a preflight credit estimate for its pending-string generator. Estimate from the number of service requests, not simply from the number of untranslated rows.

How requests are formed

POM Translate groups unresolved strings by target locale and related content context.

• A post and its mapped metadata are kept together where possible, with no more than 25 strings or 400 source words in one request.
• Other compatible strings normally target up to 50 entries and can never exceed 200 strings or 400 source words.
• Every target locale is processed separately.
• One individual string over 400 words is not sent automatically.
• A response that is empty or structurally unusable may be retried once as smaller groups.

POM Translate does not expose a preflight conversion from these groups to credits. Use current POM AI account information and a controlled sample rather than assuming a fixed price in this guide.

Build a practical estimate

1. Open Translations → Translate contents for one target language.
2. Filter by Pending and note the objects and fields involved.
3. Include pending values under Other translations if the planned run covers the whole site.
4. Repeat the count for every additional target language.
5. Note where word limits or unrelated contexts will produce separate requests.
6. Run a small representative sample and compare the POM AI balance before and after.

For a large or irregular site, a controlled sample is more reliable than arithmetic. Translate several representative posts, terms, menus, and defaults; compare the POM AI balance before and after; then extrapolate from the number and size of those objects.

Confirm before expanding

Refresh the authoritative balance in the POM AI account. Also review this site's daily word and string limits. A sufficient account balance does not bypass a local quota, and a high local quota does not guarantee sufficient credits.

After the sample, filter Machine and Pending separately. A returned batch can still leave an individual value pending when protected markers changed or the source was edited during the request, so saved-row counts alone are not a reliable usage estimate.