How to Use Confluence AI to Summarise Long Pages Fast

Short answer: Confluence AI adds three things to Atlassian’s wiki: page summaries, question answering that cites sources across spaces, and inline draft/rewrite in the editor. It is part of Atlassian Intelligence bundled into Premium ($10.35/user/month) and Enterprise, not into Free or Standard.

For runbooks and 40-page specs, summaries work well. For decisions buried in comments, they do not — comments are often excluded from the context.

What exactly does Confluence AI include?

Confluence organises content as a page tree per space: nested pages, labels, comments and attachments. Atlassian Intelligence sits on top of that as a panel and an editor button.

The three features, precisely:

Feature	Where	What it reads
Page summary	Top of page	The page body, including child page excerpts on some templates
Ask a question	Search bar	Pages you can already view, across spaces
Rewrite / draft	Editor toolbar	Selected text or a prompt

What it does not read: Slack history, Jira issue descriptions in detail, or most inline comments. If a decision was made in a comment thread, assume the AI does not know.

How do you get a usable summary of a long page?

Four moves, in this order.

1. Add a TL;DR macro at the top with two sentences. Not for the AI — for the humans who will not read 6,000 words. The AI summary then has an anchor and tends to follow it.
2. Break the page before summarising. Pages over about 8,000 words get compressed badly. Split into child pages by topic, then summarise the parent, which pulls child excerpts.
3. Put the decision in an info panel, not inline prose. Highlighted panels survive summarisation more often than body paragraphs.
4. Use labels, then ask questions by label rather than by space. “What is our current deploy rollback process” filtered to runbook returns far better sources than an unfiltered search.

If you only do one: split the long pages. A 12,000-word page from 2023 is the single biggest cause of bad summaries in every Confluence instance I have looked at.

How good is question answering across spaces?

The citation behaviour is the good part. Answers come with links to the source pages, so you can check before you act. In testing on a 400-page space, answers were accurate when a single page clearly contained the answer, and muddled when the answer was spread across three pages written at different times by different teams.

That last case is the common one, and it is a content problem rather than a model problem. Conflicting docs produce confident conflicting answers. There is no “this page is outdated” signal Confluence AI can apply on its own — the page owner has to archive.

What does it cost?

Atlassian bundles Intelligence into Premium at $10.35 per user per month (annual, Cloud) and into Enterprise. Standard at $5.75 and Free do not include it.

That is the whole pricing story, and it is the reason many small teams skip it. Moving a 15-person team from Standard to Premium costs roughly $830 a year net of the Standard fee. Whether that is worth it depends entirely on how much of your institutional knowledge lives in Confluence.

Where it falls short

Three real limitations:

• Comments are largely invisible. A spec with a 20-comment debate about scope will be summarised as if the debate did not happen. In wiki-based teams, the decision is often in the comments.
• No cross-product reasoning. Asking “why did we delay the release” will not connect a Jira epic to a Confluence page unless someone wrote that link in prose.
• Editor drafts are generic. Inline rewriting makes text cleaner and also makes it sound like every other page. Keep the rewrite for grammar fixes.

On the plus side, permissions are respected strictly, which matters more than summary quality in regulated environments.

Who should use it

Use it if you are on Premium already, your space has 200+ pages, and new joiners ask the same five questions repeatedly. Onboarding is where these features pay for themselves — a new hire asking “how do we deploy” and getting a cited answer in 10 seconds beats three days of asking in chat.

Skip it if you are on Free or Standard and the wiki is small. Also skip it if your Confluence is mostly a meeting-notes graveyard; summarising stale notes gives you shorter stale notes. Archive first, then consider Premium.

Quick Summary

• Confluence AI = page summaries, cited Q&A across spaces, and inline rewrite. Bundled in Premium ($10.35/user/month) and Enterprise.
• Comments are mostly excluded from context — decisions made in comment threads are invisible to it.
• Pages over roughly 8,000 words get compressed poorly; split them into child pages first.
• Ask questions filtered by label (runbook, spec) for noticeably better sources.
• Biggest real payback is onboarding: cited answers replace repeated questions in chat.