Skip to main content
Real Human Feedback

Guide

Common AI receptionist failures, and how to hear them

Published by the Real Human Feedback team

AI receptionists fail in patterns. The same few problems come up across products and industries, and most of them never show up on a quick test call to hear the greeting.

Here are the failures that matter most to callers, why they tend to happen, and what to listen for on your own line. None of this is a verdict on any product; it is a guide to what to check.

Invented information

The AI answers a question it was never given an answer to, and sounds confident doing it: a price, a discount, a policy, a service or opening hours that are not real. Language models are built to produce a plausible reply, so without a clear instruction to say "I do not know" they often fill the gap.

Listen for it by asking questions you never covered in the setup, such as a senior discount or a warranty length, and writing down the exact words of the reply.

Missed urgency

An emergency is handled like a routine request: a leak booked for next week, a symptom taken as a message. This is usually the failure with the highest cost, and it tends to hide because urgent calls are rare in casual testing.

Listen for it with an after hours call that uses the words your real callers would use, not the keywords the setup expects.

Failed handoff to a human

A caller asks for a person and is kept in the AI conversation, transferred to a line nobody answers, or told someone will call back with no sense of when. Callers who ask for a human have usually decided already, so a poor handoff often ends the call.

Misunderstood callers

Accents, fast talkers, background noise, interruptions and changed minds all make speech harder to understand. A receptionist that works perfectly for a calm caller in a quiet room can struggle with a parent in a car.

Long pauses and talk-over

Silence before each reply makes callers wonder whether the line dropped, and many say "hello?" just as the AI starts to speak, so both talk at once. Some hang up. Pauses are easy to miss when you are the one testing, because you expect them.

Bookings that do not match

The caller hears a confirmed time, but the calendar shows something different, or nothing at all. Checking what actually landed after a call is the only way to catch this.

Messages that lose details

A callback number with one digit wrong, a misspelled name, or a summary that drops the most important part of the call. The caller did everything right and still never hears back.

What to do with what you find

Write down each failure with the caller's words and the AI's reply, and take it to your vendor or whoever set up the receptionist. Concrete examples are what they need. Our checklist covers every category, and our rubric explains how we score them.

Hear your AI receptionist the way your customers do

Tell us the number your AI receptionist answers and anything you want us to look at. Real people place the calls, and you receive a written report.