Skip to main content

Conversations

The Conversations settings control how Rational AI enriches and processes end-user conversations behind the scenes. From here you choose the default conversational model and configure three automatic analyses—title generation, topic detection and sentiment analysis—each of which can be enabled independently and assigned its own model and prompt. These settings apply across the platform unless a touchpoint overrides them.

Who can use this feature

Available on any plan

Anyone with can edit access can change these settings

Conversational model settings​

Set the Default conversational model—the model used for all conversations unless a different one is selected in the touchpoint settings. This is the model that generates the replies users see.

Title generation model​

When enabled, Rational AI automatically generates a short title for each conversation so it's easy to recognise later.

  • Enable title generation – Turn the feature on or off.
  • Title generation frequency – How many messages to receive before a title is generated (for example, every 1 message).
  • Messages to analyze – The number of most recent messages used to generate the title.
  • Title generation model – The model used to produce the title.
  • Title generation prompt – The instruction sent to the model, returning the title as a JSON object.

Topics detection settings​

When enabled, each conversation is classified into one of a defined set of topics—useful for reporting and routing.

  • Enable topic detection – Turn the feature on or off.
  • Topics detection model – The model used to classify the conversation.
  • Detection frequency – How many messages to receive before topics are detected.
  • Messages to analyze – The number of most recent messages used for classification.
  • Topic list – The set of allowed topics (for example, Order Status, Billing Request, Tech Support). Remove a topic with its ×, or add your own with + New.
  • Topic detection template prompt – The instruction sent to the model. The model must choose exactly one label from the topic list and return it as a JSON object.

Sentiment analysis settings​

When enabled, Rational AI estimates the sentiment of user messages.

  • Enable sentiment analysis – Turn the feature on or off.
  • Sentiment analysis model – The model used to calculate the sentiment of users' messages.

Use Save to apply your changes, or Undo to discard them.