Learning Timeline
Key Insights
The Benefits of Internal Testing
Use the 'Agent Testing' feature to ensure the agent is stable before deploying it to real customers. This prevents the AI from giving inconsistent answers or experiencing 'hallucination'.
Tips for Output Accuracy
If the AI provides good answers but doesn't follow your desired format, update the System Prompt with 'Always' or 'Strictly follow' instructions to force the model (such as Gemini or GPT) to follow your script 100%.
Prompts
System Prompt for Consistent Responses
Target:
ElevenLabs Agent System Prompt
You are a support agent. Always say goodbye this way: [Insert Your Specific Goodbye Message Here]. Do not deviate from this script.
Step by Step
How to Build and Configure a Voice Agent
- Visit the ElevenLabs website and click on 'Agent Platform' in the top menu.
- Click on the 'Agent' tab to start building your voice bot.
- Fill in the 'First Message' to determine the first sentence the AI will say when the conversation begins.
- Enter a detailed 'System Prompt' to define the agent's role, personality, and specific tasks.
- Select the 'Large Language Model' (LLM) you wish to use (e.g., Gemini or other GPT models) under the model settings.
- Add 'Variables' if you require dynamic data within the conversation.
- Click 'Save' or 'Update' to save your agent configuration.
Performing Conversation Testing (Agent Testing)
- Navigate to the newly introduced 'Agent Testing' tab.
- Select 'Next Reply Test' to test conversation responses or 'Tool Test' to test external tool functions.
- Give the test a name (e.g., 'Polite Goodbye') and describe the expected output you want.
- Go to the 'Test' tab within your specific agent profile.
- Select the test you created earlier and click 'Run Test' to view the agent's performance.
- Review the test results; if the status shows 'Failed', check the differences between the agent's response and your expectations.
- Update the 'System Prompt' with stricter instructions (e.g., "Always say goodbye this way...") to fix errors.
- Rerun the test (Run Test) until the status shows the agent is strictly following the established script.
