# How to tell whether an AI receptionist actually booked the appointment

*Written by [Notaray team](https://notaray.com/about#notaray-team). Published: September 29, 2026. Updated: September 29, 2026.*

> **Short answer** An AI receptionist booked the appointment when your calendar holds an event with the caller’s name, the service, the date and the time, and the caller heard those details read back. A voice saying “you are all set” is not proof. Open the calendar, find the event, then confirm the caller received a message.

An AI receptionist answers the phone, asks what the caller needs and offers a time. The part that matters to your business is what happens after the caller hears “you are booked”: whether an event now exists on your calendar. This guide explains how to check that yourself and what proof to ask a vendor for. Notaray is in development at Agnotiq, and this is the standard we are building toward.

## What counts as proof that an appointment was booked?

The calendar is the record. A spoken confirmation, a transcript that says “booked” or a dashboard tile that turns green are all claims about the calendar, not the calendar itself. Treat a booking as real when all of these hold:

- An event exists on the calendar your staff actually use, not on a test or shared calendar nobody reads.
- The start and end time match what the caller heard, in the right time zone.
- The event holds the caller’s name, a way to reach them and the service requested.
- The caller received a confirmation after the event existed, not while the system was still working.
- There is a single event for that request, not two.

If any of these is missing, the appointment is proposed, not booked. That word matters: a proposed appointment still needs a person.

## Why can an AI phone agent say a booking is done when it is not?

Speaking and booking are two separate steps. The voice model produces the sentence the caller hears. A different call goes to the calendar to create the event, and that call can fail, time out, be refused or find the slot already taken by someone else.

If the agent speaks before the calendar has answered, it is confirming an intention. A calendar service such as the Google Calendar API returns the created event when the request succeeds. When the request times out, the outcome is unknown: the event may exist or may not. A system that treats “unknown” as “done” tells callers they are booked when nothing is on the calendar. A system that retries without care can create the event twice. One safeguard is to send the same event identifier with every retry. Google’s reference says an identifier you supply must be unique on the calendar, so a repeat request need not create a second event. It also warns that a duplicate is not guaranteed to be caught when the event is created, so the system should still check the calendar.

The US Federal Trade Commission has said there is no AI exemption from the laws against unfair or deceptive practices, and in September 2024 it announced actions against companies over deceptive AI claims. A vendor’s statement that “every call ends in a booking” is a claim you are entitled to test.

## How can you check a booking yourself in five minutes?

You do not need a technical background. You need a phone, your calendar and a second phone number or a colleague.

1. Call the receptionist as a new customer and ask for a specific service on a specific day. Write down what the agent says at the end of the call, word for word.
2. Open the calendar within a minute of hanging up. Find the event. Check the time, the name, the contact details and the service.
3. Check the caller’s side. Did a text or email arrive? Does it match the calendar?
4. Ask for the same slot from a second phone at the same time as a colleague. One call should get the slot and the other should be offered a different time or a person.
5. Call back and repeat the exact request. You should hear that the appointment already exists, not get a second event.

Do this on a normal day and again with something awkward: a slot that is already taken, a time outside your hours, a service you do not offer. Guide two in this series lists eight such scenarios.

## What should a vendor show you?

Ask for a record for every booking attempt, not a summary. A useful record shows the request as the caller made it, the moment the calendar answered, the calendar’s event reference, what the caller was told and when, and the final state: confirmed, proposed or failed. Ask what happens when the calendar does not answer, and who is told.

Notaray is being built to confirm a booking after the calendar shows it, not before. Notaray is also being built so that every finished booking attempt leaves a receipt. These are design rules, and we will publish measured results, with denominators, after they exist.

## Frequently asked questions

### Is a confirmation text proof that the appointment is booked?

No. A text proves that a message was sent. Check that it was sent after the event was created, and that the calendar entry matches its details.

### What should happen if the calendar is down or the slot was just taken?

The caller should hear that the request is pending or that another time is needed, and a named person should get the details with a deadline. The caller should not hear “booked”.

### Can an AI receptionist book the same appointment twice?

It can if a retry creates a second event. Ask a vendor how a repeated request is handled and ask to see the result of a test that repeats one.

### Who should own an uncertain booking?

One person with a deadline. A shared inbox that nobody is assigned to is how “I thought you had it” happens.

## Sources

- [FTC Announces Crackdown on Deceptive AI Claims and Schemes (press release)](https://www.ftc.gov/news-events/news/press-releases/2024/09/ftc-announces-crackdown-deceptive-ai-claims-schemes), Federal Trade Commission. Document dated: September 25, 2024. Read on: September 29, 2026.
- [Events: insert (Google Calendar API reference)](https://developers.google.com/workspace/calendar/api/v3/reference/events/insert), Google for Developers. Read on: September 29, 2026.
