People look for answers in Elium every day. When they do not find one, they leave a signal, such as a request or a search without results. This guide shows where to read these signals and how to turn each one into an article.

Where to see the gaps

Signal	What it tells you	Where to find it	Who sees it
Article requests	Someone needs an article and says so.	"Article Requests" in the menu on the left	The requester, the assignee, the administrators and coordinators of the space
Searches without results	People search for words that no article contains.	"Settings" > "Analytics" > "Search"	Platform administrators
Feedback on AI answers	Answer or a Smart Assistant gave a wrong answer, or no answer.	"Settings" > "Content management" > "Feedbacks"	Platform administrators
Annotations	A contributor thinks that part of an article is wrong or incomplete.	"Tasks" > "Annotations"	Contributors

Article requests

A request is the clearest signal, because the person tells you what they need. Requests are off by default: a platform administrator enables them in each space.

Then tell the members of the space how to ask. When a search does not give them what they need, they click "Request Article" in the search bar.

Searches without results

Click your profile picture in the top right corner, then click "Settings".
In the menu on the left, open "Analytics" and click "Search".
Choose the period, then click "Export searches".
In the file, sort the searches by number of results and keep those with no result.

The same word often appears several times, with different spellings. Group these rows and count them: a word that 30 people searched for has priority over a word that one person searched for.

Also read the "Top 10 searches" on the same page. For each frequent search, check that the first result is the article you expect.

Feedback on AI answers

Click your profile picture in the top right corner, then click "Settings".
In the menu on the left, open "Content management" and click "Feedbacks".

Each feedback shows the question, the answer and the articles that the AI used as sources. Read the sources first. In most cases:

No source, or sources on another subject: the knowledge is missing. This is a gap.
The right source, but a wrong answer: the article exists, but it is not clear enough. Rewrite the part that answers the question. See Answer: best practices.

Most people do not leave feedback. On the same page, you can export all the questions asked to Answer and read them in the same way as the searches. See Manage AI feedback.

Annotations

Annotations show the gaps inside existing articles: a step that is missing, a figure that is no longer correct. Owners find them in "Tasks". Platform administrators see all the open annotations in "Settings" > "Content management" > "Annotations".

Sort the gaps

Two groups share this task:

Space coordinators sort the requests and the annotations of their space every week.
Platform administrators read the searches and the AI feedback once a month. They send each gap to the right space.

The steps are the same for both:

Collect the new signals.
Group the signals on the same subject, and count them.
Search Elium for an article that already answers the need. Then decide:

What you find	What to do
An article answers the need, but people do not find it.	Use the readers' words in its title. Add the missing tags, and link the tags that mean the same thing. Add a summary. Tags help the search, but Answer does not use them: for AI answers, change the title and the text.
An article exists, but it is incomplete or wrong.	Create a request and assign it to the owner of the article.
No article exists.	Create a request in the right space and assign it to the person who knows the subject.
The subject does not belong in Elium.	Archive the request, and tell the requester where to find the answer.

Create a request for every gap that you keep, including those that come from the searches or from the AI feedback. All the gaps are then in one list, with an assignee and a status.

To learn how titles and tags affect the results, read Search algorithm.

Get the article written

The assignee is often an expert with little time. Make the work small:

Start from a template. A template tells the expert what to write, section by section.
Start from what exists. Import the file that already holds the answer, then correct the draft.
Let someone else write. A colleague interviews the expert and writes the draft. An AI assistant connected through MCP (Model Context Protocol) can also create the draft from notes. The expert then only reviews it, with an approval on request.

When the assignee starts, they set the status of the request to "In Progress".

Close each gap

A gap is closed when the person who asked has the answer.

In the request: link the new article under "Article for the request", then set the status to "Done". The requester gets a notification.
In the feedback on Answer: open the feedback and click "Provide the good answer" to link the new article, or click "Mark as resolved".
Test it. Type the same question in "Ask AI" and search for the same words. Check that the new article appears.

Follow three numbers

Look at these numbers once a month. They show whether the gaps close faster than they open.

Open requests: "Settings" > "Content management" > "Requests" shows, for each space, the number of open, in progress and done requests. A space where the number of open requests increases needs help.
Searches without results: count them in each export and compare with the month before.
Feedback on AI answers: the chart at the top of the "Feedbacks" page shows the positive and negative feedback over the period.

Next step

To review all the knowledge of your organisation, use the Knowledge audit checklist.
Once the articles exist, keep them correct: Set up a review routine.