How we test AI receptionists
Every service gets the same fictional business, the same instructions and the same six phone calls. We record every call and publish what happened.
1. Same business, same instructions
We set up each service as the same small dental office: identical opening hours, services, prices, an after-hours emergency instruction and a rule for when to hand a caller to a person. Only the provider changes. See the exact test profile.
2. Six scripted calls
A test caller places these calls to each service's phone number. Each call has a written pass list.
1. Book an appointment
- Offers or confirms a specific day and time
- Collects the caller's name
- Collects or confirms a callback number
- Repeats the booking back before ending
2. Ask about prices
- Answers from the business information it was given
- Does not invent a price it wasn't given
- Offers a callback or next step when it doesn't know
3. After-hours emergency
- Recognizes the urgency
- Gives the emergency instruction or number it was given
- Does not give medical advice beyond its instructions
- Takes a message for the morning
4. Spanish-speaking caller
- Understands the question
- Answers in Spanish or offers a Spanish option
- Gives the correct Saturday hours
5. Interrupts and changes their mind
- Stops talking when interrupted
- Updates the booking to Friday 3pm
- Confirms the final time correctly
6. Asks for a person
- Stays polite
- Follows the transfer or message instruction it was given
- Captures the reason for the call
3. What we measure
- Pickup and response time: rings before answering and the pause before each reply, measured from the call recording.
- Task success: each item on the pass list, marked pass or fail from the transcript.
- Accuracy: any price, hour or detail it invented that wasn't in its instructions.
- Follow-through: whether the booking or message actually reached us (email, text or calendar).
4. What we publish
For each service: the recordings, the transcripts, the pass list results and the date of the test. When a provider changes its product, we test again and keep the old result with its date.
Independence
We pay for or use the standard free trial of every service we test. Providers don't see results before publication and can't pay to change them. How we make money.