4 min read · updated 2026-10-18
What an AI receptionist can and cannot do
Short answer
An AI receptionist handles the repetitive end well: answering common questions, qualifying an enquiry, and booking from a live calendar, at any hour. It should not handle pricing exceptions, complaints, or anything it has not been given a clear answer for — those belong to a person, and the handoff is what makes the rest safe.
What it does well
The genuine value is not that it sounds human. It is that it answers at the moment someone asks, which is usually the moment they are deciding between you and the next business on the list.
- Answering the same twenty questions, in your wording, at 9pm on a Sunday.
- Working out whether an enquiry is real before it reaches you — area covered, rough timing, what they actually need.
- Offering times read from your live calendar and booking one, rather than promising a slot that does not exist.
- Taking a callback detail and passing the whole conversation to a person, so nobody has to ask the customer to repeat themselves.
What it should not do
This list matters more than the first one, because getting it wrong is what makes these systems embarrassing.
- Quote a price for anything non-standard. A confident wrong number is worse than no number.
- Handle a complaint. Someone already unhappy does not want a machine managing them.
- Interpret a policy it has not been given. If the answer is not written down, the correct behaviour is to escalate, not to reason.
- Decide anything irreversible on its own — a cancellation, a refund, a commitment you then have to honour.
The failure mode to design against
The common failure is not that the assistant cannot answer. It is that it answers anyway. Ask most general-purpose assistants something outside their knowledge and they will produce something plausible, because producing something plausible is what they do.
The fix is structural, not a matter of better wording. The assistant is given a defined set of answers, and anything outside that set triggers a handoff rather than an attempt. Designing that boundary before the conversation flow is what makes the whole thing deployable.
If you are evaluating someone to build this, ask what happens when a caller asks something unexpected. If the answer is "the AI will figure it out", keep looking.
Does it replace a receptionist?
For most small businesses, no, and framing it that way sets it up to disappoint. It covers the hours a person is not there and the questions a person finds tedious, which frees that person for the calls that actually need judgement.
That is a smaller claim than the marketing around these tools usually makes. It is also the one that survives contact with real customers.
This is what I do: AI Agents & Voice Assistants — three packages from $150.
