> For the complete documentation index, see [llms.txt](https://docs.limecall.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.limecall.com/ai-receptionist/scenarios-and-call-types.md).

# Scenarios & call types

What your assistant says and does on a call.

Open **AI Receptionist → Scenarios**. This is where you decide what the assistant actually does.

## How the assistant should behave

The main instruction field. It is written in plain language, not code.

Good instructions are specific about behaviour and boundaries:

> You are the receptionist for a two-partner dental practice. Be warm and efficient. New patients should be offered a check-up appointment; existing patients should be asked what the problem is before booking. Never give clinical advice — if someone describes pain or a dental emergency, offer the emergency slot and transfer to the on-call number. Do not quote prices for treatment beyond the check-up fee.

Notice what that does: it sets a role, a tone, a default action, and two hard limits. The limits matter more than the tone.

Use **Insert prompt template** to start from a written example for your industry.

## What it should not do

State this explicitly. Models follow a clear prohibition better than an implication.

Worth naming: quoting prices it cannot verify, promising delivery dates, giving medical or legal advice, confirming appointments it has not actually booked, and claiming to be human if directly asked.

## Call types

A call type is a named kind of call, each with its own behaviour and the fields it should collect. You can define up to twelve.

For each one, set:

**Name** — what this kind of call is. "Book an appointment", "Emergency", "Complaint", "Supplier".

**Description** — how the assistant recognises it. This is how it tells a routine booking from an emergency, so be concrete about the signals.

**How it should handle this call** — the behaviour:

| Behaviour                         | What happens                 |
| --------------------------------- | ---------------------------- |
| Answer questions & take a message | Handles it and writes it up. |
| Take booking / order details      | Collects structured details. |
| Book into my calendar             | Books a real slot.           |
| Transfer the call                 | Hands to a human.            |

**Fields** — what to collect. Mark a field with **Make sure it finds out** and the assistant will keep asking until it has it, rather than letting the caller skip past.

Ask only for what changes your next action. Every required field lengthens the call.

## Emergencies

If your business has genuine emergencies, define one call type for them.

Set **What counts as an emergency** precisely. "Urgent" is not precise — callers say everything is urgent. "No heating and there is a child or someone over 70 in the property" is precise.

Set an **On-call number for emergencies** so those calls transfer immediately rather than being written up for the morning.

## Specialized terms

Under **Specialized terms** you can add words the assistant will otherwise mishear — product names, drug names, local place names, your own brand if it is an unusual spelling. Adding them measurably improves recognition.

## Custom variables

**Custom variables** let you pass in values — a campaign name, a source, a customer ID — and refer to them in the instructions. Useful when the same assistant answers several numbers and should behave slightly differently on each.

## After you change anything

Test it. Scenario edits have effects you will not predict by reading them. See [Test your assistant](/ai-receptionist/test-your-assistant.md).