Guide · Structuring your business

Why company wikis fail (and what works instead)

Almost every company has tried: a wiki, a shared space, a knowledge base. At first, everyone writes. A few months later, nobody knows what's current. It's not a tool problem, it's a design problem.

By Olivier Giry, right-hand man to business owners · 3 min read · Updated November 4, 2026

The symptoms of a dying wiki

  • The most-visited pages are more than a year old.
  • Nobody knows who owns which page.
  • People ask a colleague rather than search, because it's faster and safer.
  • Two pages say different things about the same topic.
  • New hires are the only ones reading it, and quickly learn to stop.

Five reasons wikis die

1. Writing happens beside the work, not during it

Documenting becomes an extra task that keeps getting postponed. Knowledge stays in heads, emails and chats.

2. Nobody owns anything

A page without an owner never gets updated. When everyone can write, nobody feels responsible.

3. Nothing signals that a page is ageing

A wrong page looks exactly like a right one. Without a date or reminder, the error stays invisible until the day it costs you.

4. No starting structure

An empty wiki lets everyone organise things their own way. The result: a tree only its author understands, and duplicates.

5. It isn't there when you need it

If information isn't where you work, you don't go looking for it. More and more, people work with AI: if the AI doesn't draw on the company's knowledge, the knowledge isn't used.

Without shared memory, a company drifts

Running a company without context, memory or shared knowledge was already hard. With AI, everyone produces fast, each on their own: it becomes very hard to know who does what, and on what basis.

Everyone should be able to speak up and contribute. But without a framework, everyone does things their own way: key messages, proposals, roles and processes stop being applied, and the company drifts. Shared memory is the framework that makes trust possible.

What works: a living memory

A knowledge base that lasts rests on five principles, each answering one of the five causes:

CauseWhat works
Writing happens beside the workWrite while working: a method gets written as it's practised, by answering questions rather than facing a blank page.
Nobody owns anythingEvery document has a named author and owner.
Nothing signals ageingEvery document is dated and comes back for review at regular intervals.
No starting structureStart from a known structure: strategy, organisation, processes, rituals, roles, with templates.
It isn't there when you need itThe knowledge is available where people work, including in everyone's AI, with the access their role allows.

It also helps to separate what sets the rule (principles everyone applies) from what informs. And to let anyone suggest an improvement, approved by leadership before it applies to all.

What if I already have a wiki?

No need to throw everything out. Start by:

  1. finding the ten pages that are actually used;
  2. giving each an owner and a review date;
  3. archiving what hasn't been read in a long time;
  4. filling the gaps on critical topics: strategy, who decides what, key processes.

For the full approach: how to structure your business. To choose a tool: the comparison. On passing knowledge on: knowledge transfer.

Frequently asked questions

Why is our wiki never up to date?

Because nothing signals that a page is ageing and nobody owns it. Dating every document, naming an owner and scheduling a recheck changes everything.

What's the difference between a wiki and a knowledge base?

The wiki is the tool; the knowledge base is what you put in it. A useful knowledge base starts from a structure (strategy, organisation, processes) and stays current, whatever the tool.

How do I get busy experts to document what they know?

Ask them questions rather than asking them to write. An interview, run by a person or an AI, then a review: often one session is enough.

How do I connect a knowledge base to AI?

Connect each person's AI assistant to the company's approved knowledge, with their role's access rights, rather than pasting extracts into every conversation.