You’ve probably already got annoyed on the hype: “AI will transform everything!” But here’s the dirty secret: your AI is only as smart as the knowledge it’s fed. Oh of course, you already knew this too. But yet, we aren’t doing anything?

As an Atlassian admin, your mission isn’t just enabling cool GenAI features, it’s maintaining a healthy, evergreen knowledge base that powers real success. So with the fact GenAI requires great knowledge, what are the steps us admins need to carry out

AI Loves Fresh Context. Stale Info? Not So Much

Studies in knowledge management show that AI struggles with knowledge inertia. Agents often dig up outdated, stale content that are leading to major errors and misrepresentation. Let’s think about an example. Imagine a user asking, “What’s the current API?” and your AI points to a 2018 spec. Companies may be amazing at creating documentation, but not so great of purging or retiring data. All data has a lifecycle. Some are short, think events planning, while others may be a lot longer when properly maintained, API docs. Correctly maintaing the sources of truth, either retiring previous versions or just keeping a single page with all the right information ensures AI agents correctly deliver contextually accurate, relevant insights. Remember: freshness = trust.

Quality Knowledge = Real AI ROI

Without reliable data, AI becomes a fancy toy. I often here around me that “a fool with a tool is still a fool, just faster”. This is what unstructured AI use can often feel like. McKinsey and Bloomfire found that well-structured, high-quality knowledge can boost productivity by up to 25% and reduce search time by 35%, and this is without the introduction of AI in the mix. Adding AI on top. A tool that we ask to act and think as human as possible, needs the same stimulus and inputs as we do, so it is no surprise that a recent ScientificDirect survey also emphasises that good knowledge management (KM) is critical for AI integration across enterprises.

Atlassian admins are the companies best role for championing quality content, uniform document designs, mandatory labels and categorisations as well as regularly gardening of content.

Page templates provide great information digestion

Page templates are nothing new, nor are they unique to Confluence. Almost every application vendor recommends templating regularly used content types. This helps promote easy information digestion for users. Whether you are scanning for a particular piece of data or need guidance on how to fill in a page, templates have always helped. NOW, they are helping AI agents.

Agents depend entirely on clearly written, well-structured pages Atlassian, so as an admin, when teams are asking for advice or help on creating templates, make this a priority. How can you do this?

Enforce templates across your sites.
Help get a Power User team that can advise and promote healthy uses within their teams.
label pages with uniform template pages, even if they are not 100% the same template format.

All this helps search, summarization, and indexing work better—and keeps AI from trotting out outdated, orphaned content.

Rovo & Smart Indexing: Filter the Good Stuff

Rovo’s agents, similar to any major AI application, deliver optimal results when provided with a clear task and well-defined, focused knowledge. Defining scopes for agents carefully, ensuring that outdated or irrelevant data, such as legacy archives, is excluded from the processing gives your agents a better rate of success. The creation of blocklists and taxonomy rules helps in filtering out unnecessary or low-quality information, allowing the agents to concentrate solely on valuable content.

External sources should 100% be connected in order to create a more holistic picture, but not at the rate of skewing results due to garbage. Validate data sources to maintain high standards.

Advanced Automations Need Reliable Triggers

Native Atlassian automations have continued to grow in functionality exponentially over the last two years. This year, Atlassian announced the use of Rovo agents within native automations as well as agent stacking (use more than one agent in a rule with the output of another agent). But, as I have spoken about many times, Poor structure automations can create more havoc then benefit. This is still very much the case even when using agents. While the rule may get a little smarter, if you do not provide the right parameters or have to make too many assumptions, your rules could create duplicate pages (leading to more knowledge overload), random Jira work items or fire wrong web calls to external systems. Understand the aim of the rule. Understand whether the agents, you are including have the right knowledge and focus to provide a valuable outcome and always refer back to best practices for automation.

“Quality knowledge means automations trigger predictably, not randomly”.

Healthy KM Boosts Adoption and Culture

A recent psychology study found that AI adoption encourages knowledge sharing when employees recognise real learning benefits, creating self-reinforcing adoption loops. However, if knowledge is disorganized or unclear, people tend to disengage, causing AI to lose relevance.

Atlassian admins play a crucial role in empowering users and promoting genuinely learning-friendly environments. By providing spaces where users can form communities to ask open questions, help each other, and share advice on prompt engineering and knowledge structure best practices, while also celebrating successes, organisations can naturally foster AI adoption.

For example, if one user shared, “AI suggested this summary and reduced my meeting prep by 30%.”, community members would be able to review, relate and adopt new practices to do more with less.

Metrics That Matter: It’s Not Vanity, It’s Validation

Atlassian admins need to provide businesses the ability to accurately measure the overall quality of their Confluence knowledge to fully maximise the potential benefits offered by AI agents.

It is essential to carefully track several key performance metrics, focused soley on the development of knowledge quality, while others will be on how agents have used that knowledge to provide meaningful results.

These include;

Query success rates and search click-throughs.
- You can get some basic level success and click through rates from the new built in site and space mision control statistics.
- These serve as important indicators of how effectively users are able to find relevant and useful information within the system.
Time-to-answer and number of automation rules triggered.
- When combined, time to answer and number of rules triggered provides clear evidence of the efficiency gains driven by integrating AI technologies.
Analyse summary usage on a per-space basis, in combination with the ratio of stale versus fresh pages
- Understanding whether spaces and their content can be archived, or where content is within its lifecycle in general, offers valuable insight into the ongoing relevance and up-to-dateness of the content.

Presenting these carefully gathered metrics to leadership teams transparently demonstrates how robust knowledge management practices directly contribute to a measurable and meaningful AI impact. At the same time, it highlights that neglecting these areas can lead to wasted investments, missed opportunities for optimisation, and ultimately a failure to capitalise on AI’s full capabilities.

Final Thoughts 🎯

High‑quality, fresh knowledge isn’t optional, it’s the foundation of any AI rollout. Stale, disorganised content? That’s AI kryptonite.

Your role as Atlassian admin is equal parts, gatekeeper, enabler, architect and librarian. Maintaining knowledge health so your AI assistant delivers real value will become more and more valuable to businesses, helping avoid potential embarrassing, outdated answers.

Clean KB + smart workflows = confident AI adoption.

Knowledge Is the Key to AI