Copilot won't fix a messy intranet.
AI is only as good as the content it leans on. An audit shows what in your intranet stops AI from working, and we fix it together: structure, metadata, lifecycle and, in the end, an intranet agent on top of your own content.
Why AI gives wrong answers from the intranet.
Outdated content no one cleans up
Old instructions, finished campaigns and forgotten drafts live in the intranet for years. AI pulls them into its answers just like up-to-date content.
The reliability of information suffers, and users learn that Copilot's answers cannot be trusted.
The same information in many places and many forms
The same instruction lives on a page, in a file and in a news post, each a different version. AI picks its answer from wherever it happens to land.
The same question produces a different answer each time, and no one knows which one is right.
Metadata and structure are missing
Without content types, audiences and owners, AI searches blind. Heading structures and links do not signal what matters.
Results are random, and even good content goes unfound.
Four steps toward an agent implementation.
In most situations the natural start is the audit. It tells you what to fix and in what order, and guides the decision of whether to fix the intranet or build a new one.
Intranet audit
We go through your intranet automatically and manually. You get a report on the current state and concrete, prioritized recommendations.
Fixes
Metadata, content types, and structure fixed. Training for content authors and a cleanup guided by the audit report.
Upkeep
A one-off cleanup is not enough: new content appears and old content ages. Tools that keep the foundation AI-ready next year too.
Agent implementation
An intranet agent whose knowledge source is your own, cleaned-up content. It guides the user to the right page and the right people.
Start with the audit.
The audit produces a script-based, site-by-site picture of what in your intranet stops AI from working. The report shows the current state before anything gets fixed.
Heading structures, ALT texts, and link texts fixed so AI understands what's on the page.
One topic per page, the core point at the top, clear paragraphs.
Splitting pages that cover several topics into findable units.
Internal, optimal paths that AI travels from one piece of content to another.
Duplicate pages, empty pages and forgotten drafts that still end up in AI's answers.
Poor content practices have pushed files into hiding in the wrong libraries.
Pages and documents where permission inheritance has been cut.
Default links, external sharing and link leaks made visible.
The content audit is done manually, site by site; the technical audit runs programmatically across the whole environment. The scope and price of the audit depend on the number of sites.
From audit findings to fixed, with guidance.
The audit produces ready reports and priorities. We fix the structure and train the authors; the actual cleanup is done by your team, guided by the report.
AI cannot tell content types or audiences apart
Metadata design and implementation
Content types, metadata fields, default values and library views. Confused and contradictory answers brought under control.
Content authors do not produce AI-ready content
Training for content authors
Training plus clinics for recaps and questions. Includes guidance for the post-audit fixes and for filling in metadata.
Messy structure, broken permissions, loose sharing settings
You fix, we guide
The prioritized report and scripts say exactly what to fix and in what order. Your team does the cleanup, with our guidance.
Content authors
- ▸ Removing unnecessary pages and files based on the report
- ▸ Moving misplaced files into the right libraries
- ▸ Accessibility fixes: subheadings, ALT texts, link texts
- ▸ Splitting pages that cover several topics
Administrators
- ▸ Removing individual permissions
- ▸ Fixing sharing settings and default links
Editor-in-chief
- ▸ Taking part in definitions, trainings and clinics
- ▸ Steering the content fixes
An example: cleaning up a site of about 50 content pages takes a content author roughly 1–2 working days (about 15 minutes per page). The audit's priority list keeps the work focused on what matters.
Without upkeep, the intranet decays all over again.
A one-off cleanup is not enough: new content appears and old content ages. Lifecycle management keeps the foundation AI-ready next year too.
Aging overview page
A search page showing outdated content by site and by owner. The situation at a glance.
Aging reminder
The owner of a page or file gets an automatic reminder when content has not been reviewed in a year.
News archiving
Outdated news is moved to a separate site and out of AI's reach.
An intranet agent, once the foundation is solid.
When the content and structure are in order, the agent can answer reliably. Done in the wrong order, an agent only speeds up finding the wrong answers.
- ▸ A Copilot Studio based intranet agent whose knowledge source is your own, cleaned-up content
- ▸ Guides the user to the right site no matter where they are
- ▸ In selected topics, directs the user to contact the right people
- ▸ A feedback feature for continuous improvement
Does this sound familiar.
Copilot gives wrong or contradictory answers from intranet content, and no one knows where to start.
The intranet's content has grown over the years, and no one has a picture of what is actually there.
You want an intranet agent, but the foundation needs fixing first.
Related reading.
The audit tells you where to start.
The audit's scope is sized by the number of sites, and the report guides every next step. In a first conversation we look at the size of your environment and your goals.