Problem

In Confluence Cloud, users—particularly in large tenants—experience slow and inconsistent response times with Atlassian Intelligence (AI) search.

AI searches can take 15–20 seconds or more to return results, even for identical queries repeated within the same session (these searches integrate multiple layers—permissions checks, personalization, content ranking, and the LLM).

This scenario might lead to delays, timeouts, and reduced user trust, especially when AI search results differ between users in the same tenant who have similar permissions and access.

Suggested Solution

To improve AI search performance and consistency, especially for large tenants, the following strategies are recommended:

• Enhance backend systems to improve scalability and reduce latency;
• Improve caching to minimize redundant processing for frequently repeated queries;
• Refine personalization algorithms to ensure users with similar permissions receive consistent results, unless justified by recent content changes or permission updates.
• Provide visibility into AI search performance with metrics like average response time and cache hit rates, enabling admins to identify and report issues.
• Establish additional mechanisms for ongoing performance monitoring and tuning, leveraging user feedback and automated anomaly detection.

Why This Is Important

Fast and reliable AI search is essential for productivity in large organizations relying on Confluence Cloud.

Improving performance and consistency will enhance user satisfaction, align with industry standards, and ensure users receive consistent results, boosting confidence in Atlassian Intelligence.

Workaround

Currently, there is no known workaround for this behavior. A workaround will be added here when available.

Reference from on our Trust Center: 

Do Atlassian Intelligence and Rovo respect existing permissions?

Atlassian Intelligence and Rovo honor all existing permissions within each feature. Users will not be able to create or generate content based on resources they do not have access to.
Ex. #1. You would not see issues/projects that you do not have access to if you do a natural language search to JQL or you would not get Confluence pages sourced for an answer to a question if you did not have access to those pages.
Ex. #2. if a Confluence user executes a smart search, the results shown will take into account the pages and spaces the user has permission to view, and ignores restricted pages and spaces.