Two Test Modes
The left sidebar shows two modes under Test Modes:| Mode | Interface | Best For |
|---|---|---|
| Text to Text | Chat window - type messages, read responses | Quickly testing how the agent responds to specific questions |
| Speech to Speech | Playground - speak and hear AI voice responses | Testing the full voice experience exactly as a caller would hear it |
Text to Text Mode
Text to Text opens a Chat interface where you type messages and see your agent’s responses as text. This is the fastest way to test your agent’s knowledge and behavior without making a voice call.
How to Use Text to Text
Read the greeting message
Your agent immediately fires its configured Greeting Message. This is the first thing callers hear - check that it sounds natural and matches your brand.
Accept the privacy consent
The agent asks for consent to collect and use conversation data. Click Yes to proceed. This consent screen appears for every new conversation.
Complete onboarding
The agent asks for your name and email address before the main conversation begins. Enter details and click Submit.
Start testing
The chat input at the bottom becomes active. Type messages as if you were a real customer and read the agent’s responses.
What to Test in Text to Text
| Test Scenario | What to Type | What to Check |
|---|---|---|
| Greeting | Nothing - it auto-fires | Does it match your configured greeting message? |
| Common question | Your most frequent customer question | Does the agent answer from your knowledge base accurately? |
| Out-of-scope question | Something your agent shouldn’t know | Does it fall back gracefully without making things up? |
| Contact/hours question | ”What are your business hours?” or “Where are you located?” | Does it read from Business Information correctly? |
| Escalation trigger | ”I want to speak to a human” | Does it initiate a transfer correctly? |
| Edge case | An unusual or tricky question | Does it handle it sensibly or break? |
Speech to Speech Mode
Speech to Speech opens a Playground interface where you speak to your agent and hear its voice responses in real time. This is the most accurate way to test because it replicates the exact experience a real caller will have.
How to Use Speech to Speech
Select Speech to Speech
Click Speech to Speech in the left sidebar under Test Modes. The interface changes to Playground mode.
Allow microphone access
Your browser will ask for microphone permission. Click Allow - this is required for Speech to Speech to work.
Read the greeting
Your agent speaks its greeting message aloud. Listen carefully - this is exactly what your callers will hear first.
Complete onboarding
Enter your name and email in the fields provided and click Submit to enter the main conversation.
Speak naturally
Talk to your agent as a real customer would. The conversation transcript appears in the window - your words and the agent’s responses are both displayed so you can review them.
What to Listen For in Speech to Speech
Beyond testing what the agent says, use Speech to Speech to evaluate how it sounds:| What to Check | Why It Matters |
|---|---|
| Greeting tone | Does it sound welcoming or robotic? |
| Response length | Do responses sound natural or too long when spoken? |
| Speech rate | Is it too fast or too slow for comfortable listening? |
| Voice clarity | Is the selected voice model clear and easy to understand? |
| Natural pauses | Does the conversation flow naturally or feel rushed? |
| Fallback phrasing | When the agent doesn’t know something, does the fallback message sound natural? |
Comparing Both Modes
Use both modes together for the most thorough pre-launch testing:Use Text to Text first
Quickly run through all your test scenarios in text mode. It’s faster, easier to iterate, and lets you test many questions in a short time.
Then switch to Speech to Speech
Once text responses look good, switch to Speech to Speech to verify everything sounds right when spoken aloud. Catch issues like long responses, awkward phrasing, and voice clarity.
Pre-Launch Testing Checklist
Run through this checklist before marking your agent as ready for live calls:Greeting message sounds natural
Greeting message sounds natural
The first message callers hear sets the tone. It should be warm, clear, and under 2 sentences. Test it in Speech to Speech mode.
Common questions answered correctly
Common questions answered correctly
Test your top 5 most frequent customer questions. Each answer should come from your knowledge base, not from the AI’s general knowledge.
Business information is accurate
Business information is accurate
Ask your agent its business hours, address, and contact number. Verify the answers match exactly what you entered in Train Your AI → Business Information.
Fallback behavior works correctly
Fallback behavior works correctly
Ask something your agent doesn’t know. It should acknowledge the gap, not make up an answer, and offer to connect you with a team member.
Escalation triggers correctly
Escalation triggers correctly
Say “I want to speak to a human” or “Can I talk to someone?” and verify the agent initiates the transfer correctly.
Voice and speech rate feel natural
Voice and speech rate feel natural
In Speech to Speech mode, the conversation should feel like talking to a real assistant - not too fast, not too slow, not too formal.
Ending message sounds professional
Ending message sounds professional
Ask the agent to end the conversation and verify it delivers the configured ending message naturally.
Once all items above pass - your agent is ready to go live.
Related Pages
Agents
Go back to Basic Details to make changes to your prompt, voice, or conversation messages based on test results.
Train Your AI
Update your knowledge base if test questions are being answered incorrectly or incompletely.
Prompt Engineering
Learn how to improve your system prompt based on issues you discover during testing.
Platform Analytics
After going live, use call analytics to continue improving your agent based on real conversation data.