Feedback

Your feedback on suggestions and sources is one of the most effective ways to improve Agent Assist over time. Every rating you provide feeds back into the system, influencing how future suggestions are ranked and which knowledge base content is surfaced.

Why Feedback Matters

Agent Assist relies on a RAG pipeline that retrieves document chunks and generates answers from them. Not all chunks are equally useful -- some may be outdated, poorly written, or irrelevant to certain questions. Feedback helps the system learn which content works and which does not.

Specifically, your feedback:

• Improves RAG quality. Positive ratings reinforce high-quality answers and sources. Negative ratings signal that the retrieval or generation needs adjustment.
• Helps suppress bad chunks. When enough agents mark a particular source as unhelpful, the system can automatically reduce its ranking or suppress it from future results.
• Feeds analytics. Administrators see aggregated feedback data in dashboards, helping them identify knowledge gaps, outdated content, and areas where the knowledge base needs improvement.

Two Types of Feedback

Agent Assist supports feedback at two levels:

Type	What You Rate	Effect
Answer Feedback	The overall AI-generated answer on a suggestion card.	Signals whether the generated response was useful for the conversation.
Source Feedback	An individual source citation within a suggestion.	Signals whether a specific document chunk was relevant and accurate.

Both types use thumbs-up and thumbs-down controls. Thumbs-down opens a modal for you to select a reason and optionally leave a comment.

TIP

Rating both the answer and its individual sources gives the system the most complete signal. An answer can be poor even when the sources are good (generation issue), or sources can be poor even when the answer seems acceptable (retrieval issue).

Sub-Pages

Page	What You Will Learn
Answer Feedback	How to rate overall answer quality, select reason codes, and add comments.
Source Feedback	How to rate individual sources, how feedback affects chunk scoring, and the connection to KB helpfulness reports.