How are AI/translation credits consumed?

Initial job runs – When AI or translation jobs process selected entities for the first time.

Regenerations – Using “Regenerate with AI” or “Retry translation” also consumes credits again.

Bulk edits with AI – Any batch action that uses AI logic will use credits for every processed item.

Continuous/live jobs – Each time new items match the rule and are processed, credits are consumed.

Credits are based on tokens (chunks of text, not just words).

Longer inputs (e.g., descriptions + detailed prompts) will use more tokens.

Output length also affects credit usage — more generated text = more tokens.

AI processing for generation, extraction, or categorization all count toward your token limit.

Credits are based on the number of characters sent for translation, including spaces.

Translating large product descriptions or multiple attributes increases usage.

DeepL glossary terms don’t cost extra, but they still count toward the character total.

Regenerating multiple times because you’re fine-tuning output.

Running rules that match more items than intended — always double-check the rule builder before saving.

Continuous jobs running frequently and picking up new items without you noticing.

Translating non-essential fields because they weren’t excluded in your settings.

The job will stop processing new items and show errors for those that could not be completed.

You’ll see an error notification in the job view (e.g., “Token limit reached” for ChatGPT).

For translations, DeepL will also return an API error for insufficient characters.

You can purchase additional credits or wait until your monthly allocation renews.

Preview your rule builder results before starting a job to avoid over-processing.

Exclude unnecessary attributes in translation settings.

Use manual or AI-assisted mode only for fields that truly need it.

For AI, keep prompts concise to reduce token use while still giving clear instructions.