@Krishna Modem
The most likely reason: your agent was never limited to your new source in the first place. Per Atlassian's docs, every agent has access to all organizational knowledge unless you configure it differently. Adding a page as a knowledge source doesn't replace what the agent reads, it adds to it. So if the old content exists anywhere else the agent can see (a copy in another space, an old export, a page with similar wording, even the trash, more on that below), deleting the original page and adding a new URL changes very little. Your new source is competing with everything else in the org's knowledge, and semantic search sometimes prefers the older, better-matching text.
Work through these in order:
1. Ask the agent to show its work. In the chat, look at the row of icons directly under the agent's answer: thumbs up/down, copy, and a { } icon. That curly-braces icon is the Debug response button (easy to miss, nothing labels it "debug"). Select it and you get the response details: the agent, agent ID and request ID behind that exact answer. The panel itself tells you these are what to include if you raise a support ticket about the response. Grab them before changing any config.
2. Test in a brand-new chat. Long-running conversations carry their own context. Before concluding the index is stale, start a fresh conversation and ask the same question. If the fresh chat answers correctly, your problem was conversational carry-over, not the knowledge source.
3. Check what the agent is actually allowed to read. Agent settings, then Knowledge. You'll see three options: All organizational knowledge (the default, and the radio button most agents are still sitting on), Custom knowledge, and No organizational knowledge. If yours is on the default, switch to Custom knowledge and select only the new page or space. This single change fixes most "it keeps using the old doc" cases, because it removes the competition outright. While you're there, note the Web search toggle below: if that's on, your agent can also pull from public websites, which is another way old information sneaks into answers.
4. Check the agent's Instructions field. Knowledge and instructions are separate things. If whoever built the agent pasted content or the old URL into the instructions, no amount of knowledge-source changes will stop it repeating that text. Instructions win.
5. Purge, don't just delete. A deleted Confluence page sits in the space trash. Whether Rovo's index retains trashed content isn't clearly documented, so eliminate the variable: purge the old page from the trash, and if copies of the old content exist in other spaces, deal with those too. Then give indexing some time. Atlassian documents indexing delay as a known cause of stale results, but doesn't publish exact refresh timing for Confluence content, so I won't invent a number.
6. Still wrong after all of the above? Raise a support ticket and include the request ID from step 1. At that point it's genuinely an indexing question only Atlassian can see into, and the request ID turns "my agent answers wrong" into a traceable case.
In my experience the fix is #3 far more often than people expect. The org-knowledge default is easy to miss, and it quietly explains the exact behaviour you're describing: "I gave it a new source, why is it not using it?" Because it never stopped being allowed to use everything else.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.