Three Articles, One Process, Three Answers

Search for a process you know well and count what comes back. If you get three articles with almost the same title, you do not have good coverage. You have a fight waiting to happen.

One was written in 2023 and never archived. One is the “updated” copy someone made rather than editing the original. One is a team’s private version that escaped into the main library. They all look official. They all rank. An agent on a live call will open the first one and trust it, because they do not have time to decide which page is real.

Customers get three different answers depending on who searched, and in what order the results loaded.

How you get to three

Nobody sets out to do this. It happens because editing a live page feels riskier than writing a new one. The old page has comments, or a review date, or someone else’s name on it. A fresh page feels cleaner. Now you have two.

Or a team couldn’t find the original, so they wrote their own. Search was bad that week. Their version is slightly more accurate for their product, and slightly wrong for everyone else.

Or a change landed and someone published the new steps without taking the old ones off the board. The update exists. The leftover still answers.

Wikis are especially good at this. They make creating a page cheap and retiring one optional. That’s a big part of why teams outgrow them.

The expert on the floor already knows which of the three is “the real one”. Everyone else is guessing. When that expert is on leave, the guess gets worse.

What one version actually means

One URL. One owner. One review date. The others are history, not search results.

Archive is a decision, not a vibe. If a page can still be opened from search, it is live. Redirects, unpublish, whatever your tool calls it. “We’ll leave it up in case someone needs the old steps” is how you keep paying for the third answer.

If two products genuinely differ, say so in one place. “For product A, do this. For product B, do that.” Two slightly rewritten articles will drift the first time only one of them gets updated. I have watched that happen. The teams each thought they were looking after their own page. The customer who holds both products got both answers in one week.

Then check what else reads the library. The widget. The AI. Internal search. If you killed a leftover in the main view and an agent still answers from it, you didn’t finish. You hid it from yourself.

The honest bit

Cleaning this up is not fun, and a tool will not do the judgement for you. Someone has to read the three pages, decide which one is true, and disappoint the people who liked their copy. I will not pretend KnowledgeScout has a magic “merge my mess” button. We don’t. We give the winner a home, make the extras stop ranking, and put a review date so you notice when a second copy creeps back in.

What we can’t do is make your organisation have the argument. If two teams both believe their version is the real one, that is a meeting, not a feature.

Do the meeting once. Pick a page. Kill the other two. The next agent who searches should get an answer, not a shortlist.

The KnowledgeScout Team