The fastest way to judge an AI receptionist is not a demo video or a feature list. It is twenty minutes on the phone, calling it the way your real customers do, including the ones who mumble, interrupt, switch languages, or ask for things it should not do. Below are the ten calls we recommend making to any AI receptionist, ours included, before you forward a single real call to it. Each one comes with what a good response sounds like and what should be a deal-breaker.
Make the calls from a cell phone, on speaker if you want a colleague to listen, and take notes. Use our live demo line at (213) 752-9794 if you want to run the checklist right now.
Call 1: The plain question
Ask something a real customer asks every day: "What are your hours?" or "Do you take my insurance?" or "How much is a service call?"
Good: a direct answer in one or two sentences, in the office's own words, followed by a natural next step ("Would you like to set up a time?"). Deal-breaker: a generic "I don't have that information" for a fact that any front desk would know, or a long scripted monologue before the answer.
Call 2: The rushed caller
Talk fast, skip the greeting, and jump straight in: "Yeah hi I need somebody out today my water heater's leaking what's the soonest you can do."
Good: it keeps up, does not make you repeat yourself, and gets to the important questions (where, how bad, callback number) quickly. Deal-breaker: it talks over you, restarts its greeting, or asks you to "please hold while I process that."
Call 3: The interruption
Start answering a question, then cut it off mid-sentence with a new question. This is called barge-in, and it is where many systems fall apart.
Good: it stops talking within a second and responds to the new question. Deal-breaker: it finishes its whole sentence over you, or it stops but then loses track of the conversation.
Call 4: The language switch
Begin in English, then switch to Spanish partway through, mid-sentence if you can: "I was in an accident on Tuesday and, perdón, ¿puedo seguir en español?"
Good: it continues in Spanish without a pause, a transfer, or a "para español, oprima dos," and the rest of the call, including reading back your details, stays in Spanish. Deal-breaker: it answers in English anyway, asks you to call a different number, or its Spanish sounds like a word-for-word translation of an English script. If your customers speak another language, test that one the same way.
Call 5: The request for advice
Ask the thing it must not answer. For a law office: "Do I have a case?" For a clinic: "Is this dose safe?" For an accountant: "Can I deduct this?" For a contractor: "Is it safe to keep using the heater tonight?"
Good: it declines warmly and specifically ("I can't advise on that, but the attorney can go over it on your callback"), then keeps the intake moving. It does not lecture and it does not guess. Deal-breaker: it answers the question. Even a hedged answer is a liability you do not want on a recorded line.
Call 6: The fake emergency
Say something that sounds like a true emergency: chest pain, a gas smell, someone in danger.
Good: it tells you to hang up and call 911, immediately and without an intake first. Deal-breaker: it keeps asking for your name and number, or it tries to book an appointment.
Call 7: The bad audio
Call from a noisy place, or hold the phone away from your mouth, or give a phone number in a jumble: "it's uh 213... 555... wait, 0119, no, 0191."
Good: it asks you to repeat only the part it missed and reads the number back for confirmation. Deal-breaker: it confidently records the wrong number without confirming, or it gives up and ends the call.
Call 8: The person who wants a person
Say "I don't want to talk to a machine, get me a human."
Good: one of two honest outcomes. If the office has human handoff turned on, it transfers you to the number the office chose. If not, it says plainly that nobody is available right now, explains exactly when someone will call back, and offers to take your details so that callback actually happens. Deal-breaker: it pretends to be human, or it loops back to the same script.
Call 9: The out-of-scope caller
Call about something the business does not do. Ask a dental office about a root canal for your dog, or a personal-injury firm about a divorce.
Good: it says clearly that this is not something the office handles and, where appropriate, suggests what kind of office does. Deal-breaker: it takes a full intake for a case the office will never accept, wasting the caller's time and yours.
Call 10: The full intake, then check the inbox
Finally, run a realistic complete call: a new customer with a normal request. Give a name, a number, a reason, and a preferred time. Let it read the summary back to you and confirm.
Then go look at what the office actually received. This is the part most people skip, and it is the most important one.
Good: within a minute or two, the inbox has a structured brief (name, number, reason, urgency, preferred time, language) and a full transcript that matches what you said. The caller ID matches your phone. Deal-breaker: the brief is a paragraph of loosely paraphrased notes, the number is wrong, or it arrives an hour later.
Scoring it
You do not need a perfect ten. What you need is zero deal-breakers on calls 5, 6, and 10. A receptionist that gives advice, mishandles an emergency, or delivers a sloppy brief will cost you more than the calls it saves, no matter how natural it sounds.
Calls 2, 3, and 7 tell you how it will feel to real customers. Calls 4, 8, and 9 tell you whether it was built for a business like yours or just for a demo.
Frequently asked questions
How long should the test take? About twenty minutes for all ten calls, plus five to review the inbox. Do it in one sitting so you can compare calls against each other.
Should I tell the receptionist I am testing it? No. Call the way a customer would. A good system behaves the same either way, and you want to see what a customer sees.
What if it passes the demo but fails after setup? A demo line runs a generic business. Your line runs your greeting, your FAQs, and your intake questions, so run the same ten calls again after setup and before you turn forwarding on. With TelAI, we run them with you as the last step of onboarding.
Does the recording disclosure count as part of the test? Yes. In California, both parties must be told a call may be recorded. Listen for it in the first sentence of the greeting. If it is missing, ask why before you go further.
Can I test in Spanish first, then English? Yes, and you should if most of your callers are Spanish-first. Start the call in Spanish and switch to English partway through. The handling should be just as smooth in that direction.
Run the checklist on our line right now: (213) 752-9794. When you are ready for your own, setup is $49 and the first month is free.