Why Does Rovo Mess Up Confluence Formatting? [Champion's Slack Insider]

[#1]

This thread started by @Darryl Lee can be summarized in one word:

“IN. CON. SIS. TEN. CY!”

Let’s break down what’s actually happening.

The questions

• Is “Edit Confluence Page” in Rovo the same as Automation append?
• Why does formatting (like tables) look correct in chat but break on the page?
• Can we force a follow-up prompt like “I Agree” vs “Anonymous”?

1. Rovo edit ≠ Automation append

Short answer: They are not the same.

Automation for Confluence:

• Deterministic
• Uses native ADF handling
• Predictable results

Rovo “Edit page”:

• AI-generated output
• Then converted into ADF
• Subject to formatting drift

Same outcome, different engine.

2. Why your table looks correct… then breaks

This is the frustrating part:

• Rovo generates a clean table in chat
• You see a proper preview
• It writes to Confluence
• The structure collapses into a single line or loses formatting

What’s actually happening

There are two steps:

1. Generation (LLM) Produces formatted, markdown-like output
2. Conversion (to ADF) Translates that into Confluence’s internal format

That second step is where issues occur. This is especially noticeable with:

• Tables
• Mixed formatting
• Wiki markup

Why Automation feels more reliable

Automation avoids interpretation.

• No formatting guesswork
• No conversion ambiguity
• Direct ADF structure

That’s why: Automation is currently more consistent for structured updates like tables.

3. Why you can’t force a second prompt

Agents do not strictly control conversation flow. Even if you want: “I Agree” or “Anonymous”

Rovo may:

• Ask clarifying questions
• Continue the original context
• Ignore the intended flow

You can:

• Suggest options
• Add conversation starters

You cannot:

• Enforce a required next step

What to do today

If your use case involves:

• Structured updates (tables, rows, forms) > Use Automation
• Logic, drafting, summarization > Use Rovo
• Combination workflows > Use both
  • Rovo to generate content
  • Automation to write it reliably

What about fixing this long-term?

There are active requests around:

• Formatting reliability
• Confluence editing behavior

Example: ROVO-682 (worth following if this impacts your workflows)

Champion takeaway

Rovo is handling two complex tasks:

• Generating content
• Translating it into a structured system

Automation handles only one. That difference is what you’re experiencing.

If you’ve ever thought about testing this repeatedly to see different results, you’re not alone. These kinds of edge cases are exactly where patterns emerge—and where useful feedback for product teams comes from.