# AI receptionist testing kit

By Kashif Rehman | BizThrivers | October 15, 2026
https://bizthrivers.com/resources/ai-receptionist-testing-kit/

Free to adapt. Use sample values; do not place credentials or customer records in this worksheet.

## Planning decisions

### Agent, scope and test channel
Example: inbound appointment enquiry; web test for conversation checks, actual test phone call for phone routing and transfers.
Your decision: 

### Human owner, hours and fallback
Record the approved transfer destination, availability window and fallback when no human answers.
Your decision: 

## Observed test record

### 1. Greeting and purpose
Action: Call with a normal appointment enquiry.
Expected: The agent explains its role, understands the request and moves to the next approved step.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 2. Known service question
Action: Ask a question covered by the current approved knowledge source.
Expected: The answer agrees with that source without adding unapproved claims or offers.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 3. Unknown question
Action: Ask for a detail absent from the approved source.
Expected: The agent acknowledges the gap and uses the agreed human or callback path instead of inventing an answer.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 4. Unclear request
Action: Give a short or ambiguous request, then clarify it.
Expected: The agent asks a useful clarification and follows the clarified intent.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 5. Incorrect detail and correction
Action: Use sample contact details and correct one detail during the call.
Expected: The final readback and stored test record reflect the corrected detail.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 6. Interruption or background noise
Action: Interrupt a response and repeat the request in a noisy test environment.
Expected: The agent recovers without losing the intended next step or repeating an inappropriate action.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 7. Available appointment
Action: Request a genuinely available test slot and confirm the timezone.
Expected: A booking is created only for the agreed calendar and slot; confirmation matches the observed appointment.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 8. Unavailable appointment
Action: Request a blocked or unavailable test slot.
Expected: The agent offers a valid alternative or handoff and does not claim the unavailable time was booked.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 9. Request a human — phone test
Action: Use an actual test phone call and explicitly ask to speak to a person.
Expected: The approved human-transfer path works. A successful web conversation does not prove phone transfer.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 10. No human available — phone test
Action: Test outside the transfer window or with a controlled unanswered destination.
Expected: The caller receives the agreed fallback or callback step, rather than an endless or abandoned transfer.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 11. Failed action or integration
Action: Use a controlled failure in a booking or downstream test action.
Expected: The agent avoids claiming success, records the exception and gives it an operating owner.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

### 12. Post-call record and routing
Action: Inspect the test contact, call summary, assignment and any triggered workflow.
Expected: The stored outcome and handoff match the call; unintended follow-up is identified and fixed.
Observed: 
Evidence location: 
Responsible owner / fix: 
Result: Untested / Needs work / Passed

## Release and handover
Outstanding fixes: 
Operating owner: 
Approver and decision: 
Review date: 
Where to monitor and how to pause a problem: 
Change-request route: 

## Sources checked October 15, 2026
https://help.gohighlevel.com/support/solutions/articles/155000004107-how-to-create-voice-ai-agents

Call Transfer cannot run in a HighLevel Web Call test. Verify human transfer with an actual test phone call. Agent Transfer between AI agents is a different action.

A completed log records your observations. It does not automatically verify the CRM or agent.
Independent HighLevel affiliate: I may earn a commission through referral links on BizThrivers.