Use Confluence's AI to Draft Runbook and Data Dictionary Pages

What This Does

Runbooks and data dictionaries rot fast because writing them competes with pipeline work and always loses. Confluence's built in AI assistant, Atlassian Intelligence, expands rough bullet notes into a formatted page in the same space your team already checks during an incident, so the doc gets written where people will actually find it.

Before You Start

• You have a Confluence space where runbooks or data dictionaries already live
• Your organization is on a Confluence Cloud Premium or Enterprise plan. Atlassian Intelligence features sit behind those tiers, so ask your Atlassian admin if you are not sure which plan your instance runs
• You have five to ten bullet points about the topic (a DAG's retry behavior, a table's grain and refresh schedule, an escalation path)

Steps

1. Find the AI feature

Open the page you want to expand, or create a new page from the runbook or data dictionary template your team uses. Look near the top right of the editor, next to the lock icon, for a "Summarise" style AI control, or open the editor's insert menu for a prompt-based drafting option that generates a structured draft from a short description. The exact label has shifted between Atlassian releases, so if you don't see it where described, check the toolbar's AI or magic wand icon.

2. Tell it what you need

Type your bullet notes directly, or write a short prompt describing the page: "Draft an on call runbook for the fct_orders DAG covering retry behavior, who to page, and known false alarms." For a data dictionary entry, list the table name, each column, and what it means in plain terms, then ask the assistant to format it to match your space's existing pages.

3. Review and use the result

Read the draft against what you know is actually true. AI drafting tools reliably improve formatting and phrasing, but they can smooth over gaps in your notes with plausible sounding filler. Before publishing, have the person who last touched the system (or your own memory of the last incident) confirm the retry counts, escalation contacts, and any command examples are correct, then save the page in its normal location.

Real Example

Scenario: You just finished stabilizing a flaky Airflow DAG and want the runbook written before you forget the details.

What you type/do: Five bullet points: "DAG retries 3x on failure, alerts #data-oncall after 2nd retry, common cause is a slow upstream API timeout, manual fix is clearing the task instance in the Airflow UI, escalate to platform team if it fails 3 days running."

What you get: A formatted runbook page with a summary line, a "When This Fires" section, step by step manual recovery instructions, and an escalation section, ready for a quick edit pass rather than a blank page.

Tips

• Point the assistant at an existing well-written runbook first and ask it to match that page's structure. Consistent formatting across runbooks matters more during a 2am incident than clever prose.
• Don't paste connection strings, credentials, or internal hostnames into the prompt even for an internal runbook. Reference them as "the usual staging connection" and let the reader fill in specifics from a secrets manager.
• Rerun the same prompt after a real incident to fold in what you learned. A runbook that never changes after being written is the one that goes stale first.

Tool interfaces change. If a button has moved, look for similar AI/magic/smart options in the same menu area.