An intranet is not a SharePoint project

When an organization rolls out Copilot and the first answers turn out to be outdated, the culprit is usually not Copilot. It is the intranet. Copilot pulls its answers from the same SharePoint that has accumulated ten years of sediment: three versions of the travel policy, two outdated org charts and thousands of pages whose owner left the company years ago. A human knows to skip these, because they recognize from context what is current. A language model does not. It treats all indexed content as true by default.

This is why we talk about an AI-ready intranet as a structural decision. The question is not what the front page looks like. The question is what content the environment holds, who owns it, and how a machine tells current from outdated.

Why the old intranet project does not solve this

A traditional intranet renewal is a communications project: a new look, new navigation, a news feed on the front page. Content is often moved from the old environment to the new as-is, because cleaning it up is laborious and no one owns it. The result looks better, but the structure of the information is the same as before.

From Copilot’s perspective nothing changed. It does not look at the front page or the navigation. It looks at content, metadata and permissions. If the same topic has three mutually contradictory pages, Copilot picks one of them, and the user gets an answer that may be wrong without anyone noticing right away.

This also creates a less obvious risk: permissions. Copilot shows the user everything they have technical access to, including things the user did not know existed. Sites and folders loosely shared over the years, which used to stay hidden because no one knew to look for them, now surface in search results. AI readiness therefore also means auditing access, not just cleaning content.

What the work actually consists of

The working order is the same one we have found in agent governance: first visibility, then structure, only then AI.

Visibility means a content audit: what sites and pages the environment holds, when they were last updated, who owns them and where access has been shared. This is mechanical work that moves fast when done systematically. The audit’s priority list tells which part of the content to clean first and what can be archived right away.

Structure means three decisions. First: every piece of content that remains has a named owner and a review cycle, so the decay does not start again immediately. Second: metadata describes the content so a machine understands what it is about and who it is for. Third: permissions match what is actually meant to be shared, not what has accumulated over the years.

AI comes last, and then it works. When Copilot answers from a cleaned, owned knowledge base, the quality of the answers becomes usable without touching the tool itself. The same holds for every agent later built on the same information: no agent is better than the knowledge base it draws from.

”We have too much content to clean”

This is the most common objection, and it rests on the false assumption that everything must be cleaned at once. It doesn’t have to be. The audit almost always reveals that the vast majority of searches and Copilot answers hit a small part of the content. When that part is put in shape first and the rest is archived out of the index’s reach, the effect shows immediately, even with old content still uncleaned.

Archiving is the key word here. Old content does not need to be destroyed; it just needs to stop appearing to AI as truth. That is a much smaller decision than deletion, which is why it can be made fast.

Where to start

If your organization has Copilot in use or on the way, the intranet’s condition is worth establishing before users learn the answers cannot be trusted. Trust is lost once and earned back slowly. A content audit is the lightest way to get a picture: it commits you to nothing bigger, but afterwards you know as facts how much of the content is current, who owns what, and which permissions need fixing.

An intranet renewal should therefore start with a question that has nothing to do with the design: if Copilot answered tomorrow based solely on our intranet, would you trust the answers? If the answer hesitates, the work starts with the structure of the information.