Enable Filtering of Conversations with Knowledge Source Usage in Ultimate (AI Agents Advanced)

Please give a quick overview of your product feature request or feedback and note who in your organization is affected by this issue [ex. agents, admins, customers, etc.] (2-3 sentences)

We would like Zendesk to add the ability to drill down into conversations where a specific Knowledge Source article was used by the AI Agent. Currently, Knowledge Source analytics only provide aggregated metrics such as Usage Rate, Automated Resolutions, Verified Resolutions, and BSAT, without any way to inspect the underlying conversations. This limitation primarily affects AI administrators, knowledge base managers, customer support leaders, and content owners responsible for maintaining and optimizing AI performance.

What problem do you see this solving? (1-2 sentences)

This functionality would enable teams to understand how individual knowledge articles perform in real customer interactions and identify content gaps, inaccuracies, or opportunities for improvement. It would also make Knowledge Sources as actionable and manageable as AI Use Cases, which already support conversation-level analysis.

When was the last time you were affected by this lack of functionality, or specific tool? What happened? How often does this problem occur and how does this impact your business? (3-4 sentences)

We encounter this limitation regularly during our AI Agent performance reviews and knowledge base optimization initiatives. For example, when an article shows a low BSAT score or a high escalation rate, we have no way to understand what happened in the conversations where that article was used. Without access to those interactions, it is difficult to determine whether the issue stems from the article content, customer intent mismatches, or AI behavior. This occurs on an ongoing basis and significantly slows down our ability to improve automation performance, customer satisfaction, and knowledge base quality.

Are you currently using a workaround to solve this problem? (If yes, please explain) (1-2 sentences)

There is currently no effective workaround. We can only review high-level metrics and attempt to manually identify related conversations through separate reporting or ticket searches, which is time-consuming and often inconclusive.

What would be your ideal solution to this problem? How would it work or function? (1-2 sentences)