Testing Your AI (Playground & Call Simulator)
Rehearse text and voice conversations safely before customers ever see them
On this page
Why Test First
Both test surfaces run your real AI — same knowledge, same persona, same rules — but nothing is sent to customers and nothing is saved to your pipeline. Test after every knowledge or personality change.
Test the AI (Text Playground)
AI Receptionist › Test is a chat with your own receptionist. Toggle between SMS and Web chat modes — the AI answers with that channel’s length limits and knowledge trimming, so you see exactly what a texting customer would get. Switching channels restarts the conversation.
- Sample prompts help you start; sessions cap at 20 turns — reset to start over
- Reply badges show when an answer was grounded in your knowledge or delivered word-for-word
- The What the AI knows side panel shows the knowledge text in scope for the selected channel
Test a Call (Voice Simulator)
the Phone call tab of AI Receptionist › Test (Pro) rehearses a phone call in your browser. Type your side of the call, or go hands-free with the microphone for a real listen–think–speak loop, with spoken replies in the actual receptionist voice.
- Test scenarios auto-play scripted calls (normal inquiry, emergency phrasing, request for a human) so you can verify routing without improvising
- Tool calls the AI would make — booking, SMS — are shown but marked simulated: no side effects, ever
- Per-turn diagnostics show latency, model, tokens, and cost; session stats summarize the run
- Copy the transcript as Markdown or download the session as JSON to share with your team
A Good Test Pass
Ask your top 10 real questions
Use the questions customers actually ask — pricing, availability, service area, policies.
Try to break it
Ask about services you do not offer, out-of-area jobs, and competitor comparisons. The AI should decline gracefully, not improvise.
Run the edge scenarios
Emergency phrasing and "let me talk to a person" should escalate or transfer per your rules.
Fix and repeat
Close gaps in the knowledge pack, adjust rules, and rerun the same questions until answers hold.