AI receptionist

Taking Accurate Messages with an AI Receptionist

What a good phone message contains, how an AI receptionist can capture names, numbers and reasons reliably, and how to check and act on them quickly.

Call editorial team · Published · 5 min read

Taking a message sounds simple. In practice it is where many phone systems fail: a name misspelled, a number with a digit wrong, a reason reduced to "call me". When that happens the callback goes nowhere, and the caller concludes that nobody cares.

A hand writing a message on a notepad next to a desk telephone
Illustrative image

An AI receptionist can take messages reliably, but only if you design the exchange. This guide covers what a useful message contains, how the agent should ask for it, and how your team should handle it afterwards. For other ways of covering missed calls, see callbacks and voicemail in an IVR.

What a good message contains

Ask yourself what the person reading it needs. They are likely to be busy, switching between tasks and possibly reading on a phone. The message must stand alone.

ItemWhy it mattersExample
NameSo the callback starts politelyAlex Morgan
NumberThe route back to the callerRead back digit by digit
ReasonLets the right person prepareAsking about a delayed order
UrgencyDecides the order of callbacksWants a reply today
Best timeAvoids missed callbacksAfter three in the afternoon
Company or referenceFinds the account quicklyOrder reference given by the caller

Do not ask for more than you need. Each extra question makes the call longer and may feel like an interrogation. Keep personal information to what the callback needs, in line with data minimisation in support.

Where voice capture goes wrong

Speech recognition is good but not perfect, and the weakest spots are the ones messages depend on. Names, email addresses and digit strings are harder than ordinary sentences.

The fix is not a better microphone. It is a better conversation: repeat what you heard and let the caller correct it. See how voice AI handles numbers and names for more detail.

A reliable way to capture a number

  1. Ask"What is the best number to call you back on?"
  2. CaptureLet the caller say it at their own pace without interrupting.
  3. Read backRepeat it in groups of digits.
  4. ConfirmAsk "Is that right?" and correct if not.
  5. RecordSave the confirmed number with the message.

Ask in the right order

Order matters. Start with the reason, because it is the easiest to say and it tells the agent whether it can help directly. Then ask for the name, then the number, then the best time. If the caller sounds urgent, put urgency first and shorten the rest.

A callback request should feel like help, not a form. The agent might say: "I'm sorry the team is busy. Can I take a message? What's the call about?" and then move through the questions one at a time. Avoid asking for everything in a single sentence, because callers forget the second item.

Summary, transcript or both

A message can arrive in several forms. Choose according to how your team works.

Message formats

Short summaryFull transcript
Reading timeA few secondsLonger, but complete
RiskDetails missed by the summaryImportant point buried in the text
Best forDaily triageChecking what exactly was said
Pair withA link to the full callA short summary on top

Most teams do best with both: a short summary at the top and the full conversation one click away. See AI call summaries for what a good summary looks like and summarisation prompts for how to shape one.

What happens after the message

A message is only useful if somebody acts on it. Agree on who reads messages, how often and what they do next.

Message handling routine

  • A named person reads messages at set times
  • Urgent messages are marked and returned first
  • Each message gets a status such as new, called, done
  • Callers who cannot be reached get a second attempt
  • Messages older than a day are flagged
  • Repeated reasons feed back into the FAQ

Track how long messages wait before a callback. That number tells you more about your service than most dashboards. For a measure, see callback completion rate.

Messages outside business hours

Out-of-hours messages need clear expectations. The agent should say when the team will return calls, not just that a message was taken. If you do not call back on weekends, say so. If your trade has emergencies, tell callers what to do, for example how to reach an emergency service. See after-hours call handling for the wider approach.

A sample message exchange

Here is how a short exchange might read for a hypothetical plumbing business. The agent has already said it cannot reach the team.

SpeakerWhat is said
AgentI'm sorry, the team is out on jobs. I can take a message. What is it about?
CallerMy kitchen tap is dripping badly.
AgentThanks. Can I have your name?
CallerPriya Nair.
AgentAnd the best number to call you on?
CallerIt starts with zero seven, then the rest in groups.
AgentLet me read that back in groups of digits. Is that right?
CallerYes.
AgentIs it urgent, or is later today fine?
CallerLater today is fine.
AgentThanks, Priya. I've noted a dripping kitchen tap, a callback later today. The team will ring you back.

Notice that the agent does not promise a visit or a price. It captures the facts and the expected response. The wording of the last line sets expectations without overpromising.

Checking message quality

Do not assume the messages are good. Sample them. Take ten messages from the week and read them as if you had to call back each person without the recording.

If three of the ten fail, change the agent's instructions and test again. Quality sampling for calls works the same way; see QA sampling for calls for ways to do it with limited time.

Where messages should land

Decide in advance where messages arrive and who can see them. A message that lands in one person's mailbox disappears when that person is away. A shared place, with a clear owner each day, keeps work moving across holidays and sick days. If your team already handles email and chat together, consider putting phone messages in the same view so the customer's history is in one place. See the unified inbox for the idea.

Make sure that people who see messages are those who should. Messages contain personal information, so limit access to staff who need it and review who has it from time to time.

Messages in Call

In Call, an AI agent can take a caller's details and the call is saved in call history with its summary and transcript, so a colleague can open it and read what was said. Queues can have their own voicemail, and menu options can ask for a callback, so a caller who cannot reach anyone has a defined route. Callback requests show up as work for the team, and the AI call history lets you read each conversation. Where the line offers it, you can use the call-back feature to return a call from the history.

Frequently asked questions

Is an AI message better than voicemail?

It can be, because it asks specific questions and confirms details, while voicemail records whatever the caller chooses to say. It still needs review, as AI can mishear names and numbers.

How do I reduce wrong numbers?

Have the agent read the number back in groups and ask the caller to confirm. That one step removes most errors.

Should callers leave an email address too?

Only if a callback by phone is not enough. Email addresses are hard to capture by voice, so ask for them only when you need them.

What if the caller refuses to leave details?

Accept it. Offer another route, such as calling back later, and do not pressure them.

Who should read the messages?

Name one person for each shift, with a backup. Shared inboxes with no owner are where messages go to be forgotten.

How quickly should we call back?

Set a target that suits your business and say it in the message to callers. Whatever you promise, keep.

Capture callback requests and messages with an AI agent in Call

In the demo console you can try incoming calls in a simulation.

Related guides