> ## Documentation Index
> Fetch the complete documentation index at: https://docs.assistable.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Write a conversation node

> Give each conversation step a clear task and connect the paths it can take.

Use a **Conversation** node when the assistant needs to ask a question, explain something, or respond to a visitor. Keep each node focused on one stage of the conversation.

## Configure the response

Select **Prompt** and describe the task, what information to collect, and when to move on. For example:

```text theme={null}
Ask whether the visitor needs a new appointment or help with an existing booking.
If their answer is unclear, ask one short clarifying question.
Do not confirm a booking here; this step only identifies what they need.
```

The flow's **Global Prompt** also applies. Avoid repeating or contradicting those shared instructions in every node. Insert a known variable such as `{{caller_name}}` only when a prior step or integration supplies it.

Use **Static** for a fixed message. Variables are substituted into the text. Static does not mean that the entire step avoids model work: choosing a transition can still involve the model, and a later visit to the same static node can use conversational generation.

## Connect the next step

1. Add an outgoing condition describing an observable outcome, such as “The visitor wants a new appointment”.
2. Connect that path's handle to the node that should handle the outcome.
3. Repeat for other outcomes. Provide a suitable otherwise path where the flow should always continue.
4. Test an answer that matches each path and one that matches none.

Keep conditions distinct. “Wants help” and “Wants appointment help” overlap; “Wants a new appointment” and “Wants to change an existing appointment” are easier to test.

## Shared interruptions

A **Global Node** offers a shared conversational destination, such as a request for human help. Configure its entry conditions and test it from several stages. Marking a node global does not itself connect a telephone transfer or perform another action.

## Check the result

Use **Test → Start conversation** and follow the highlighted active node. Confirm the visitor gets the question or explanation they need before the next action. If the flow moves too early, review the prompt, outgoing conditions, and otherwise path.

See [AI assistance](/v3/guides/flows/improve-prompts) for prompt improvement and [variables](/v3/guides/flows/variables) for dynamic text.

## Conversation settings reference

| Setting | Values, defaults, and effect |
| - | - |
| Name | Label used to identify the node on the canvas and in test details. |
| Prompt / Static | **Prompt** is the default for a new node. Prompt describes the response to generate; Static supplies fixed text for the first visit. Both can use supplied variables. |
| Instruction text | Combines with Global Prompt. Use `{{name}}` for a known variable; unknown placeholders remain unchanged. |
| Outgoing conditions | Natural-language descriptions of when to take each path. Add a destination connection for each condition. |
| Otherwise path | Default connected destination when no conditional transition is selected. |
| Global Node | Off for a normal new node. Makes the node available as a shared destination from other conversation stages. The start node is also treated as global. |
| When should the AI route here? | Entry condition for a global node, such as “The customer asks to change an existing booking”. |
| Few-Shot Examples | Add/remove **User says** and **Assistant responds** pairs. These are examples included before conversation history, not new knowledge sources or guaranteed replies. |
| LLM Override | Provider, model, temperature, priority processing, fallback provider, and fallback model. Empty/inherited fields use flow defaults. [All model settings](/v3/guides/flows/model-access). |

## Visible settings with limited support

The current editor also exposes the controls below. Their saved values do not currently control the flow runtime, so use the supported alternative when the behavior matters.

| Control | Choices shown | Supported approach |
| - | - | - |
| Greeting, on the start node | Same for all or Per channel; Voice/Chat and Inbound/Outbound; Wait, Generated, or Written | Configure the assistant’s actual channel greeting and test its published channel. Do not depend on these flow greeting fields to select a greeting. |
| Skip Response | Off by default | Design the connected path explicitly; use processing nodes for steps that should not speak. |
| Block Interruptions | Off by default | Configure and test the assistant’s voice interruption settings for phone behavior. |

A variable picker beside a field only inserts text. Check the [flow variable reference](/v3/guides/flows/variables) for which names are computed and which need integration data.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.