Configure Language Selection for Prompt Templates
Manage which languages your prompt template can generate in and how the language is selected. Configure system-managed language selection based on user and org settings, or enable instruction-driven language selection where the LLM interprets response language directly from prompt instructions.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Performance, and Unlimited Editions with the Einstein for Platform, or Einstein or Agentforce for Sales or Service add-on, or Agentforce Foundations |
In Prompt Builder, select the languages that a template supports to generate output, and control how the system determines the response language. Manage the allowed languages to make sure that the template generates responses only in languages that you tested for quality output, and that your organization supports or uses fluently.
In the Response section of Template Settings, use the Automatic toggle to control language selection behavior. Enable the toggle for system-managed language selection, or disable it so the LLM interprets the language directly from your prompt instructions.
Automatic Language Mode (Default)
In Automatic mode, the platform automatically determines the response language based on predefined logic.
If you don't specify a language when executing the template, Prompt Builder selects the response language based on this order:
- Use the response language set in the outputLanguage parameter in Apex or the REST API if supplied and the prompt template supports it.
- If the template doesn’t support the supplied outputLanguage, Use the executing user's language if the prompt template supports it.
- If the template doesn't support the user's language, use the org's default language if supported.
- If the template doesn't support the org's default language, use English if supported.
- If the prompt template doesn't support any of these languages, Prompt Builder returns an error.
When to use Automatic mode:
- Maintain consistency across users and regions.
- Enforce language compliance.
- Align responses with org-supported languages.
- Apply allowed language restrictions.
Open Response Language Mode
Open Response Language is a configuration that uses natural language instructions to determine the response language instead of system-managed rules. When you disable the Automatic toggle, the system removes language selection constraints so the large language model (LLM) interprets language instructions directly.
How it works:
- The system no longer enforces language selection.
- The LLM determines the response language based on prompt instructions and provided context.
- You can use flexible and unstructured instructions, such as "Respond in Spanish," "Use British English," or "Write this in Hindi with a formal tone."
- The model interprets intent and generates responses accordingly, without strict system constraints.
When to use Open Response Language mode:
- You need dynamic, per-request language control.
- Your use case involves multi-language or global audiences.
- You want prompt-driven behavior instead of system enforcement.
Considerations
- The system does not enforce allowed languages when Automatic mode is disabled.
- Response quality depends on how clearly you write language instructions in your prompt.
- Results may vary based on the LLM's understanding of a language or dialect.
- Don’t disable Automatic mode for strict compliance or regulated environments where language governance is required.
- outputLanguage values supplied to prompts using Open Response Language at runtime will be ignored.
Language Support
Standard and new custom templates support the default languages. Turn on Global Language Support in Einstein Setup to support other languages. However, when you use a global language, some Einstein Trust Layer features, such as data masking and toxicity scoring, are sometimes unavailable, and the system doesn’t always optimize responses for business-appropriate phrasing.
