When 2 Pages are the Same Which Wins in Rovo? [Champions Slack Insider]

[#1]

Let’s kick it off with a great one from @Yong Yang - 填空题咨询:

“If we have two Confluence pages with the same knowledge from different years, how does Rovo decide what’s the latest?”

Short answer: It doesn’t “pick the newest” automatically.

Rovo isn’t scanning your space and saying, “Ah yes, 2024 beats 2022.”

Instead, it uses a retrieval system powered by the Teamwork Graph. That means it pulls content based on:

• Relevance to your query
• Your permissions
• Context (how content is connected, referenced, and used)

If two pages contain similar information… there’s a good chance Rovo retrieves both. Then the model does what it does best: synthesizes an answer across sources.

So where does “recency” come in?

Recency can influence results—but it’s not a guaranteed tie-breaker.

Signals that may help:

• Last updated date
• Page activity
• Links from other pages
• General usage patterns

But none of these override everything else. If an older page is still highly relevant or well-linked, it can still show up.

Why this matters (more than people think)

This is where things can get messy.

If:

• Your older page still has valid info
• Your newer page reflects updated features
• And both are active

Rovo may blend them together.

That can lead to:

• Slightly outdated answers
• Conflicting guidance
• Confusion for end users

What Champions recommended (and why it works)

There were a few solid takes in the thread:

1. Archive when you can
If a page is no longer the source of truth, archive it. Clean inputs = cleaner outputs.

2. Keep one source of truth
This is still the gold standard. Rovo performs best when your content does too.

3. Guide behavior with Agents
As @Matthias Gaiser _K15t_ pointed out, you can prompt an agent to

“Prefer newer pages and only supplement with older ones.”

This is a practical workaround when archiving isn’t an option.

What’s really happening under the hood

Rovo is doing exactly what modern AI systems are designed to do:

• Retrieve multiple relevant sources
• Weigh signals (including—but not limited to—recency)
• Generate a response that sounds cohesive

That last part is key. The answer may feel unified… even if the sources behind it aren’t.

[famous Videos]

Thanks for the explanation! 

Makes sense that Rovo doesn’t automatically pick the newest content but instead weighs relevance, context, and permissions. The tip about archiving old pages and keeping a single source of truth is especially useful to avoid outdated or conflicting answers.