The greeting is the only part of a phone agent every single caller experiences, which makes it tempting to over-invest in. It's also, by a wide margin, the part with the least effect on whether the system actually works for you. Both of those things are true at once, and holding them together is the point of this piece.
The short version: spend a little time on the greeting because it's cheap and it sets expectations. Then spend the rest of your evaluation energy on comprehension and the POS write, because that's where systems actually differ.
What the first eight seconds are for
A greeting does three jobs and should do them fast:
Confirm they dialed correctly. Say the restaurant's name clearly. Nothing else matters if the caller isn't sure they reached you.
Set the interaction model. The caller needs to know they can just talk. A greeting that ends with an open invitation — "what can I get for you?" — does this implicitly, and it's the main thing separating a voice agent from a phone tree in the caller's mind. If they're waiting for a menu of options, you've lost the advantage.
Get out of the way. Every second of greeting is a second the caller isn't saying what they wanted to say.
That's it. Hours, promotions, and today's special all feel like natural greeting content, and they're almost always a mistake there. The caller who wanted your hours will ask. The caller who wanted to order now has to wait through information they didn't want.
What's genuinely worth customizing
In rough order of impact:
The restaurant's name, pronounced correctly. Sounds trivial. Isn't, if your name is a family surname, a non-English word, or something people routinely mispronounce. Get this right and verify it on a live call.
Location disambiguation, if you have multiple. "Thanks for calling Marco's on Fifth" saves a whole exchange for a caller who dialed the wrong store, which matters more for multi-location groups than almost any other greeting choice.
Register. A fine-dining room and a wing shop should not sound the same. This isn't about personality quirks; it's about matching the formality your guests already expect. A greeting that's too casual for the room reads as sloppy, and one that's too formal for a counter-service spot reads as fake.
Time-of-day variation, lightly. If you're closed and taking orders for tomorrow, saying so up front prevents a confusing conversation. This overlaps with holiday and hours overrides, which is a bigger topic than greetings alone.
What's decoration
Elaborate personality. Catchphrases. A "brand voice document" for a phone agent. These feel like they should matter and mostly don't, for a reason worth stating directly: your caller is hungry and busy. They are not evaluating your brand. They want to order food and hang up. Personality that adds friction to that transaction is negative value no matter how well it's written.
There's also a failure mode where heavy personality tuning makes the agent worse at its job — a system spending words on charm is spending time, and time is the thing callers actually notice. The response-speed budget is real, as we cover in sub-second latency, and charm competes for it.
The disclosure question
Whether to say "this is an automated assistant" is a real decision, not an obvious one, and reasonable operators land differently.
The case for disclosing: it's honest, it sets expectations so a caller doesn't feel tricked when they clock it mid-call, and it gives them permission to speak more clearly. It also removes the awkward discovery moment that some people react badly to.
The case against: it can prime a caller to be skeptical or to ask for a human reflexively, before they've experienced that the thing works.
What's not up for debate is the underlying reality: some of your regulars will notice. Modern systems sound natural enough that many callers won't immediately clock it, but "many" is not "all," and building your plan on the assumption that nobody will ever know is building on sand. We look at that more directly in can callers tell it's AI.
Don't let voice quality do your evaluating
Here's the trap the greeting sets. It's the most demo-able part of the product, it's the first thing you hear, and it forms an impression that colors everything after. A warm, natural greeting makes a system feel competent before it's demonstrated any competence.
Voice quality has largely converged across serious vendors. What hasn't converged is whether the agent understands "half pepperoni, light sauce on the other half" against your actual menu, and whether that lands in your POS as a ticket the kitchen can make. That's the integration depth argument, and the greeting is precisely where it's easiest to forget it.
A practical discipline: when you demo a system, listen to the greeting, form your impression, then deliberately set that impression aside and order something complicated. Judge on the second part.
A greeting checklist
- Restaurant name, clearly, first.
- Location, if you have more than one.
- An open invitation to speak.
- Under about five seconds.
- Nothing promotional.
- Verified on a live call, from a real phone, not a recording the vendor sent you.
That last line matters. A greeting sounds different over a cell connection than in a polished sample file, and the cell connection is what your customers will hear.
The bottom line
Customize your greeting enough that it says your name correctly, matches your room, and hands the floor to the caller in under five seconds. Then stop, and put your evaluation effort into the parts that decide whether Friday night goes well. The greeting is the most visible part of a phone agent and one of the least consequential — which is exactly why it's so easy to over-weight.