Improve AI-generated replies with fact invalidation

When reviewing AI-generated replies that use similar conversations as a knowledge source, admins can “invalidate” associated facts by clicking the trash can icon directly in a conversation or in workspace settings. When a fact is invalidated, Front will no longer use that fact in future AI replies.

Fact invalidation is important for:

• Refining Autopilot replies: Allows for strict control over the knowledge used in responses before activating Autopilot's auto-reply feature.

• Maintaining knowledge accuracy: Remove outdated information from past conversations, as historical records cannot be updated in the same way as knowledge base articles.

You must have Copilot or Autopilot enabled to use this feature.

Step 1

Click the gear icon, navigate to workspace settings, then select Topics in the sidebar.

Step 2

Select the Topic you’d like to manage, then click Preview auto-replies.

Step 3

In the pop-up, hover over the # sources indicator. Hover over a similar conversation (indicated with an envelope icon) to view relevant facts used to inform AI-generated responses.

Click the trash can icon to invalidate a fact.

Step 4

The fact will be displayed with strikethrough formatting, and will no longer be used when AI generates auto-replies. You can return to the fact and click the trash can icon to restore the fact at any time.

Step 1

In your inbox view, navigate to a conversation with an AI-generated message.

Step 2

Hover over the # sources indicator. In the list of related sources, hover a similar conversation (indicated with an envelope icon) to view relevant facts used to inform AI-generated responses.

Click the trash can icon to invalidate a fact.

Step 3

The fact will be displayed with strikethrough formatting, and will no longer be used when AI generates auto-replies. You can return to the fact and click the trash can icon to restore the fact at any time.

Fact invalidation does not apply to:

• Workflows using the Reply with Autopilot instructions rule action

• Drafting and refining replies when composing with Copilot

Only admins can invalidate facts. Users must have the default workspace admin role or company admin role.

For the initial launch of this feature, we are focusing on similar conversation facts, as they are the most likely to become outdated or lack broad applicability.

Fact invalidation is not currently supported for facts from connected knowledge sources such as public websites or knowledge bases, so you will see these facts without trash icons.

Not at this time, but we will be focusing on more knowledge management features in the near future.

Yes. Updates to similar conversation facts are immediately applied to newly AI-generated replies. However, these updated facts will not alter conversations that already include AI-generated replies.

We recommend invalidating facts any time you see a fact you don't want to use. However, we highly encourage a thorough review of AI replies before enabling Topic-based Autopilot to ensure that automated replies use the right knowledge.