AI Receptionist: What It Can and Can't Do

In short

An AI receptionist is an assistant that answers the people who contact your business, by phone, WhatsApp or website chat. It can greet them, answer common questions, collect their details, offer appointment times and pass everything to your team. It should say that it is an AI assistant, stay inside rules you set, and hand the conversation to a person whenever the question is sensitive, unclear or outside those rules. It does not replace your staff, give medical or legal advice, or make judgment calls. And in the US it should not be used to call or text people who haven't agreed to it: the law restricts automated and AI-voiced calls and texts made without consent.

What it does well

The value is simple: no inquiry goes unanswered. A clinic's front desk can't pick up during a treatment, a contractor can't answer from a roof, and nobody answers at 11 p.m. on a Sunday. That's when people call the next business on the list. In the reviews we read of AI receptionist and messaging products, the most common praise was exactly this: calls answered around the clock and no missed calls.

Well set up, an assistant can:

Answer calls and messages at any hour, in the language the customer uses.
Answer the questions you hear every day: hours, location, parking, services, what to bring, how booking works.
Collect a name, a phone number and the reason for the call, and write it straight into your CRM.
Offer free times from your calendar and confirm a booking, when the rules allow it.
Return a missed call or message with a short text, for someone who has just contacted you.
Pass a clear summary to a person on your team, so nobody asks the customer to repeat everything.

Where it breaks

The complaints are as consistent as the praise. In the reviews we read of these products, failures of the AI, the phone line or the automation came up 23 times: an assistant asking questions the caller had already answered, appointments that seemed booked but never reached the calendar, calls dropped, and business numbers stuck in carrier verification for weeks or flagged as spam. Billing surprises and slow support were even more common complaints, tied at 28 each.

None of this means the idea doesn't work. It means the setup matters more than the product. Most failures come from three places: rules that were never written down, connections to the calendar or CRM that were never tested end to end, and no clear path to a human.

What it should never do

Pretend to be a person. It introduces itself as the business's AI assistant, and says so again if asked.
Give medical, legal or financial advice, or anything that needs professional judgment.
Make promises the business hasn't approved: prices, discounts, results.
Keep a frustrated or confused caller in a loop. When in doubt, it hands over.
Contact people who haven't contacted you or agreed to hear from you.

Being upfront is not only good manners. People who know they are talking to an assistant ask simpler questions, forgive small mistakes, and ask for a person when they need one. And some US states already require bots to identify themselves in certain sales conversations.

Consent: the rule that matters most in the US

A receptionist answers people who reached out to you. That is very different from an assistant that calls or texts people on its own. US federal law, the Telephone Consumer Protection Act (TCPA), restricts calls made with an artificial or prerecorded voice, and automated texts, to people who have not given the right kind of consent. In February 2024 the Federal Communications Commission ruled that AI-generated voices count as "artificial" under that law. So an AI voice calling people who haven't agreed to it falls under the same restrictions as a robocall. Florida has its own telephone solicitation law as well. And Florida generally requires everyone on a call to agree before it is recorded, so if the assistant records or transcribes calls, callers should hear that at the start.

In practice: let the assistant answer inbound calls and messages, reply to people who just contacted you, and send reminders only to customers who agreed to receive them. Anything outbound, such as campaigns, reactivation calls or follow-ups to old lists, needs a proper consent process first. This is general information, not legal advice. Before you switch anything on, check your plan with a lawyer who knows telemarketing rules.

Start with one job, not with everything

The assistants that work are rarely the ones switched on for every task on day one. Pick the moment where you lose the most today, and start there. For many businesses that's the after-hours call, or the missed call during the day that nobody returns until tomorrow. For others it's WhatsApp messages that wait for hours. Once that one job works reliably, add the next: booking, then reminders for customers who asked for them, then a summary of the day for the team.

Each channel has its own rules, too. On the WhatsApp business platform, a business can reply freely for 24 hours after a customer's last message; after that, only pre-approved message templates can be sent, and only to people who agreed to receive them. Phone numbers used for automated texting go through carrier registration, which can take time. Plan for these before a launch date, not after.

Clinics: privacy comes first

For clinics, the assistant will hear health information whether you plan for it or not. People say why they are calling. That shapes the setup: which details the assistant collects, where they are stored, who can see them, and what never goes into a text message or an email. If a vendor handles patient information on the clinic's behalf, HIPAA generally requires a business associate agreement with that vendor. Oddly, privacy came up in none of the 29 clinic-related reviews we read, which suggests many buyers don't ask. Ask. A good default is to collect only what is needed to book and call back, and leave the clinical conversation to the staff. Our page for clinics explains how we approach it.

Test it on your own number before you decide

Don't judge an AI receptionist by a demo. Test it the way your customers will meet it:

Call it from a normal phone, at night and during the day, and ask your ten most common questions.
Ask something it shouldn't answer, such as a medical or pricing question outside its rules, and check that it hands over.
Say you want a person, and time how long that takes.
Book an appointment, then check that it really appears in the calendar and in the CRM.
Speak with an accent, interrupt it, change your mind halfway. Real callers do.
Try each language you serve: English, and Hebrew, Russian or Spanish if your customers use them.
Read the summaries it writes for your team. Are they useful at a glance?
Ask who owns the phone number, the recordings and the data, and how you would leave.

How it looks in practice

At the Organic Well clinic in Miami, calls, messages, inquiries and appointments share one workspace. Vicky, the AI assistant the clinic named, is designed to help with missed calls, after-hours messages, follow-up on inquiries and booking, within the rules the clinic configured, and the team stays in charge of everything else. You can see how that system is built in the Organic Well CRM case study.

How we set one up

At Goldren Studio we start by writing the rules with you: what the assistant answers, what it never answers, when it hands over and to whom. We connect it to your calendar and CRM, test it on real calls in every language you need, and only then point real customers to it. The phone number, the data and the accounts are in your name. More on our AI receptionist service and on AI and automation in general.

See it working before you decide

Tell Daniel how calls and messages reach your business today. He replies within one business day, and before you decide anything you'll see an assistant working on a test number, with your own questions. Write on WhatsApp or through the contact page.