Rovo answers from every Confluence page the asking user can access, so the dependable way to stop it quoting an outdated page is to change or remove that page, not to prompt around it.
Archiving is one outcome of that work, but the first goal is simpler: make it obvious which pages can be trusted, who stands behind them, and when they were last confirmed.
This article shows how to find the outdated pages, put an owner and a status on them, and act on the ones nobody confirms.
Why agent settings and prompts do not remove an outdated page
Narrowing an agent's knowledge sources and telling it to ignore old data both help, and community answers on agents that keep reading an old source list those checks. They do not touch the page itself. A copy of last year's table in another space, a duplicated runbook, or a replaced policy stays readable for every other search and every other agent. When two pages disagree, Rovo doesn't reliably pick the right one, but it still burns credits trying.
Volume matters too. The more outdated and duplicate pages sit in the pool, the more noise Rovo has to sift through, which makes answers slower and less precise for everyone, not only for the one question that quoted a stale page. Removing what nobody needs is a quality measure for AI answers as much as a housekeeping task.
Confluence has no built-in way to keep that from happening. Page status is a label someone sets by hand (maximum 5 can be managed centrally), and there is no owner role that an automation can use to reach the person who should decide.
How to find and fix outdated pages before Rovo reads them
- Measure the problem. List the pages that have not been updated or viewed for a period you choose. In Better Content Archiving and Analytics, the Content List report shows the result with owner, status and dates, so you can see the size of the backlog before you change anything.
- Give every page one or more owners. Set owners (yes, multiple owners are possible) at the top of each page tree and let them be inherited down the tree, then work through the pages that still have none. The page owner report shows which pages are still unowned.
- Let the status follow the evidence. A content status scheme assigns statuses "Up to Date", "Needs review" or "Outdated" from last update and last view, so nobody sets a label by hand. The first matching rule wins, and you can always check why a certain status has been applied.
- Ask the owner once, with everything in one email. The notification emails go to the owners automatically, one email per owner listing all of their pages due for review. The owner updates the page or confirms it as valid, which records who confirmed it and when, and restarts the clock.
- Act on what nobody confirmed. An automation archives, deletes, or purges the pages that stayed outdated after the confirmation window. Which one you pick matters, see the next section.
Archive, delete, or purge for pages Rovo must not read
Choose the action by what should happen to the content. Archive suits pages people may still need to find on purpose, because it takes them out of everyday browsing and keeps them recoverable.
Delete suits pages with no remaining value, and purge suits content that must never be read again, because purging cannot be undone. When a page must not reach an answer under any circumstances, delete or purge it rather than relying on archiving.
The automations in Better Content Archiving and Analytics support all three actions, and purge is off until an admin enables it in the global settings. Pick the action your retention policy allows.
Where a regulation requires deletion after a fixed period, the two combine in one flow: archive first, then delete or purge what has been archived for the required time, as described in what to add to Atlassian's six-month automation rule.
Two safeguards apply during archiving. By default the automations skip pages that have children, so the page tree does not reshuffle. They also keep both direct and inherited view restrictions, which matters because the built-in archive action can drop a restriction a page inherits from its parent, as the open request CONFCLOUD-83913 describes.
Not everything old should go. Lessons learned, post-incident reviews and decision records keep their value because they are history, and they need a different rule than a how-to page. The article on setting owners and review cycles per content type shows how to keep that history on purpose, with a status that tells readers and Rovo what the page is.
How to check that Rovo stopped quoting the page
Run the check in a new chat, because a long conversation carries its own context. If an answer still shows the old content, use the Debug response icon under the answer and note the agent ID and request ID, and allow time for Rovo to re-index after the change.
What a rule-based layer adds that Rovo does not
A rule runs the same way on every page on every run, and the job audit log keeps the counts of pages archived, deleted and purged for each execution. That is the control a retention or compliance review asks for, and it does not depend on how a model weighs two pages.
nCino replaced Confluence Automation with Better Content Archiving and Analytics when it stopped scaling and cut its page count by 30%.
Storck did the same across 4,000 users during its move to Confluence Cloud. For the argument behind this procedure, see Rovo is about to quote your most outdated Confluence page.
Start a free trial of Better Content Archiving and Analytics for Confluence, or read the documentation.
