Everything this produces fits on one sheet taped next to the phone: the script to read when an assistant calls, the text that keeps each appointment pending until the customer replies, and the table where every call gets logged. Behind it sit three decisions, and you make them sitting down with whoever answers your phone. The first time takes half an hour; after that, a minute per call.
Before you start: this guide isn't for detecting machines. You don't need to catch them. What your front desk needs is to know what to do either way, without guessing who's calling, with an answer that doesn't depend on how anyone's day is going.
What's actually known about these calls
Not much, and it's better not to make up the rest.
What had been published as of September 28, 2026 fits in a few lines. According to Reuters (September 22), Muse, Meta's assistant, places calls to U.S. businesses on its user's behalf, to book an appointment or ask about a price, and Meta has been opening the feature to its users gradually. In internal posts, a Meta employee testing it complained that his insurer wouldn't stay on the line once it heard AI. The next day, at Meta Connect, the company announced that Muse will get its own email address and that in the coming months you'll be able to ask it for an appointment from your glasses.
What isn't known: how many of these calls a business gets today, how many will come next year, or whether they'll all identify themselves as AI. Nobody has published it, and this guide doesn't assume it.
What does change right now is the kind of call that can come in. Until recently there were two: the customer, or someone calling for them, like a family member or an executive assistant. Now there's a third: a customer's AI assistant, calling on that person's behalf. Your front desk already knows how to handle somebody's assistant. This is a lot closer to that than to a scam, and that's how it gets treated.
First decision: do we take the call?
Our recommendation: yes, on one condition. It has to say who it's calling for. If it doesn't, ask. If it can't tell you, that's when you end the call, politely.
Why not hang up on reflex: if the call is what it says it is, on the other end there's someone who wanted something from your business and handed the call to a machine. Hanging up on the assistant is hanging up on that person. And they do find out. The one who doesn't is you, because a dropped call never gets written down anywhere.
If your business decides the opposite, that these calls never get handled, that works too, as long as it's a decision and not a reflex. In that case the line is: "On this line we only work with the person directly. We're glad to help if they call or message us." And it still goes on the sheet in Part 6, because in a month you'll know how many times it happened.
The opening line, if you take the call:
"Happy to help. Who are you calling for, and what does that person need?"
Second decision: what a machine gets told
The rule is short: an assistant gets told what's already public. What belongs to a person gets told only to that person.
| It gets told | It doesn't get told |
|---|---|
| Hours and address | Whether someone is or isn't your customer |
| What services you offer | Anyone's history, diagnosis or file |
| Prices you've already published | Balances, debts or payment details |
| Whether there's general availability on a date | Changes to an existing appointment |
| What someone needs to bring to an appointment | Changes to a customer's phone, email or address |
If your business handles health or financial information, the privacy rules you already follow on the phone with any stranger still govern. This guide isn't legal advice: how they apply to these calls is for whoever handles your compliance to confirm.
The line when it asks for something in the right column:
"That's something we only handle directly with the person. If they call or message us, we're glad to help."
The line is the same whether the person is your customer or not. If it changes depending on the case, the line itself answers the question.
Third decision: how it gets confirmed
This is the one that makes everything else work. An appointment an assistant asks for stays pending until the customer confirms it from their own phone.
- If the person is already your customer, you text the number you already have on file. Not the one the assistant gives you. Ask for the number every time, customer or not. If they're already a customer, the text goes to the number on file even if the assistant gives a different one, and the caller doesn't get told that.
- If they're new, you text the number the assistant gives you, and the appointment stays pending until they reply.
That way, if the request was real, the customer confirms in a minute, and if the assistant got it wrong, the customer fixes it before the appointment exists. With a new customer, that text doesn't prove who they are, only that someone at that number wants the appointment. Who they are gets confirmed the usual way, when they walk in. What it does guarantee is that your calendar doesn't fill up with whatever a machine understood.
The line to close the call:
"Perfect. We've got the request down for [day] at [time]. What number should we text [name] at to confirm it? As soon as they reply, it's booked."
The confirmation text, to copy as is:
"Hi [name], this is [your business]. We received an appointment request in your name for [day] at [time], made through an AI assistant. Should we confirm it? Reply YES to confirm, or NO if you didn't request it."
The last line is the one that protects both sides. If someone booked an appointment in another person's name, that person finds out from you, and your business is the one that comes out looking good.
If your business is in health care, that text carries the most neutral business name you have, never the treatment or the specialty: if the number the assistant gave is wrong, someone else reads it.
If the customer doesn't reply. Pick a deadline and make it the same for everyone; how long you hold the slot is up to your calendar. When it passes, one follow-up text, just one, to the same number. If there's still no reply, the slot opens back up and the log gets a "didn't confirm." An appointment nobody confirmed doesn't count as a no-show: if they never saw it, they didn't miss anything.
The full script, to tape next to the phone
Print it on one sheet and tape it where the phone gets answered. Nobody has to memorize it. It's there to be read.
The log sheet: how many times it really happened
One sheet with five columns, on paper or in a spreadsheet:
| Date | Calling for | What they asked for | Handled? | Did the customer confirm? |
|---|
Log it for one month. At the end you'll have a number nobody has published for your business today: how many times it happened to you. If it was zero, you spent thirty minutes and the script is ready for when it starts. If it was several, you know how many times it happened and what was done with each one.
And if you're the one using AI on the phone
It works the other way too. If your business answers or places calls with an AI assistant, say so in the first sentence. For example: "Hi, you've reached the virtual assistant for [your business]."
According to Reuters, Meta had contract workers place some calls that went out from Muse, without proper disclosure, and a vice president acknowledged that was a miss. If your assistant also makes outbound calls, in the U.S. an AI voice needs the prior consent of the person being called (FCC, 2024). Confirm that with your advisor before you switch it on.
What this guide leaves out
It doesn't set up your own assistant. If what you want is an AI assistant answering for your business, that's a different job and it's already written up in the 5-step AI agent strategy guide.
It isn't a security policy or legal advice. It doesn't replace the privacy rules your business already follows, and it doesn't say how they apply to your case.
Last thing
If you only do one thing from this guide, make it this: print Part 5, sit down for ten minutes with whoever answers the phone, and read it out loud together. Any line that sounds off coming from that person gets changed right there, before it goes on the wall.
The first call from an assistant will be answered by whoever is working that day. Make sure the sheet is in front of them.
Want us to go over the script with your front desk before you tape it up? Book a free call.
And if you want one practical AI thing a week, no hype: subscribe to The AI Edge.
Sources
- Meta testing a "human concierge" for its new personal AI agent, Muse, Reuters via BNN Bloomberg, September 22, 2026 - https://www.bnnbloomberg.ca/business/artificial-intelligence/2026/09/22/meta-testing-a-human-concierge-for-its-new-personal-ai-agent-muse-reuters-exclusive/
- Everything we announced at Meta Connect 2026, Meta, September 23, 2026 - https://www.meta.com/blog/meta-connect-2026-everything-we-announced/
- FCC Confirms that TCPA Applies to AI Technologies that Generate Human Voices, FCC, February 8, 2024 - https://www.fcc.gov/document/fcc-confirms-tcpa-applies-ai-technologies-generate-human-voices
