Buyer’s checklist

How to choose an AI receptionist

Ten questions — most answerable in thirty minutes on a demo line — that separate the products that work from the ones that only demo well.

Flat engraved emblem: an audio waveform above a row of identical test squares with a small ticked checklist beside them, inside a double-rule frame

The ten questions

This checklist is a condensed, do-it-yourself version of the rubric we use in formal testing. Thirty minutes with a vendor’s demo line answers most of it.

  1. Call the demo line yourself — twice. Once as an easy caller, once as a hard one: interrupt it, mumble, change your mind mid-sentence. The gap between those two calls is the product.
  2. Ask it to reschedule something. Booking is the rehearsed path. Rescheduling and cancelling exercise lookup, verification and change — where weak products fall apart.
  3. Describe an emergency. Say the words a panicked caller would say. Anything other than immediate escalation is disqualifying for medical, dental and home services.
  4. Count the silence. Time the pause before each answer. Three seconds feels broken on a phone; ask the vendor what their real-world latency is and how they measure it.
  5. Make it take a message. Then check the message it delivers against what you said. Transcription that mangles names and numbers creates work instead of saving it.
  6. Ask what it does when it doesn’t know. The right answer involves a graceful handoff to a human. The wrong answer is confident improvisation.
  7. Demand a scheduling round-trip in YOUR system. An integration listed on a website is not an integration working in your practice-management or dispatch software. Watch a live round-trip before signing.
  8. Ask how updates are tested. Models and prompts change weekly. Who verifies that this week’s update didn’t break rescheduling? (This is exactly why continuous testing exists.)
  9. Get the failure numbers. Ask: of your last thousand calls, how many ended with the caller hanging up unserved? A vendor that can’t answer isn’t measuring.
  10. Read an independent report card. Sales demos are the vendor’s best five minutes. A report card is a run of controlled calls scored against one rubric, with the number of calls printed beside the letter and the evidence published.
Shortcut: if the vendor has a Proofground report card, questions 1–6 are already answered — with the calls behind them.
Evaluating vendors right now?
Run the same recorded, graded test calls against your shortlist — your vendors, your scenarios, priced per conversation. The calling itself takes hours; the honest variable is how quickly each vendor grants you pilot access.
See how a run works

Common questions

AI receptionist vs. answering service — which do I need?

A human answering service scales poorly and costs per minute; an AI receptionist scales infinitely but varies wildly in quality. The checklist above works on both — the grade that matters is how they handle YOUR calls.

How much should an AI receptionist cost?

Typical pricing runs from modest flat monthly fees to per-minute rates. Price predicts quality poorly in our testing — grade first, then compare price among vendors that pass.

Proofground

The independent guide to AI receptionists. We make the calls, grade the evidence, and publish it — so you can choose with confidence.

© 2026 Proofground, published by Chest LLC. Grades are based on our own recorded test calls, scored against a published rubric and independently re-reviewed. We take no vendor money for grades — how we stay independent.