Configuring Response Rules for InSite Agents

Response rules let you shape how your agent responds in specific situations — whether that means delivering a particular message, including a link, attaching a form, or directing visitors away from topics you don't want the agent to handle. This article covers how rules work and how to configure them effectively.

How response rules work

Your InSite agent can only respond based on two things: the content synced into its knowledge sources, and the guidance you define in response rules. It cannot access your website directly, pull live page content, or connect to any other services. This means the agent's knowledge reflects your site as it was at the last sync — by default, every 30 days.

When a visitor sends a message, the agent checks whether any of your response rules apply. Each rule has one or more triggers, and the agent uses AI to evaluate whether the visitor's message matches a trigger — so matching is flexible and intelligent, not just literal.

If a rule is triggered, the agent responds using that rule's guidance instead of searching your knowledge source for an answer. This gives you precise control over specific situation without limiting what the agent can do everywhere else.

Creating a response rule

Navigate to the Responses tab in the agent composer and add a new rule. Each rule has the following parts:

Triggers

Add one or more triggers to define when the rule should fire. Each trigger is a text field where you can enter either a specific phrase or a natural-language description of what the visitor might be asking.

For example, both of these are valid triggers for the same rule:

• book a demo - matches on that specific phrase
• visitor is asking to speak with someone or schedule a meeting - lets the agent recognize the intent regardless of how the visitor phrases it

You can use specific terms when you want precise matching, or descriptions when you want the agent to recognize a broader range of phrasings.

Response guidance 

Write the guidance the agent should follow when the rule fires. This works in two ways:

• Exact response: Wrap your text in quotation marks and the agent will deliver what is written, without interpretation.
• Guidance Write a description of how to respond and the agent will generate a response using that guidance, shaped by your voice and tone setting.

Use exact responses when wording matters. Use guidance when you want the response to feel natural and conversational.

Asset (optional)

You can attach one asset to each rule: a link or a form.

• Link: Add a URL and display text. The link appears bellow the agent's response, giving visitors a direct next step.
• Form: Select an Act-On form to display inline in the chat. See Forms and Lead Capture for details on configuring forms.

Using rules to redirect visitors

Response rules are also useful for topics you don't want the agent to handle directly — such as billing questions, support issues or anything better handled by a person.

To redirect a visitor, write response guidance that acknowledges the question and points them to the right place. Add a link asset to make it easy for them to follow through. For example: 

Trigger: visitor is asking about their bill or invoice

Response guidance: Let the visitor know that billing questions are handled by our support team and invite them to reach out directly.

Asset: Link to your support contact page.

The agent will not search the knowledge source when a rule fires. so the visitor gets a consistent, controlled response rather than the agent attempting to answer from content it may not have.

Conflicting triggers

Rules are listed alphabetically by name in the composer. This affects how they appear in the UI but does not set a priority order — the agent is responsible for determining which rule applies when a visitors message could match more than one.

Because of this, it is worth naming rules clearly and keeping triggers specific enough that overlap is unlikely. If two rules have very similar triggers, review them to make sure the guidance in each is distinct enough that either response would be appropriate if the agent chose it.