Articles with a structure people can follow
Spaces, categories and tags rather than one endless list. Step-by-step procedures, policies, troubleshooting guides and FAQs each get a template, so a new article looks like the last one and a reader knows where the actual instruction starts.
Search that finds the answer, not the word
Full text across titles and body, weighted by how often an article is opened and marked helpful, with synonyms you define — your staff type refund, your policy says reimbursement, both should land in the same place. Filters by space and category narrow a large library fast.
Permissions: internal, team-only and public
Set who can read each space and each article. Commission structures and escalation scripts stay internal; delivery areas and warranty terms publish to your website as a help centre on your own domain. One library, one edit, two audiences — no maintaining a public copy by hand.
Owners, review dates and stale-content alerts
Every article carries an owner and a next review date. When the date passes, the owner gets a reminder and the article is flagged as due for review in the admin view. This single feature is the difference between a knowledge base and a graveyard of documents nobody trusts.
Drafts, approvals and version history
Edits can go live immediately or route to a reviewer, depending on the space. Every published version is kept, with who changed what and when, and any version can be restored. When a policy changed on a particular date, you can show exactly what it said before.
Feedback and the gaps it reveals
Readers mark an article helpful or not and can say why. Searches that returned nothing are logged as requested articles. Those two lists tell you what to write next, based on what people actually looked for rather than what someone assumed they needed.
The AI assistant answers from these articles
Your published knowledge base is what the free lifetime assistant reads, so it answers customers and staff from your policies and cites the article it used. Update the article and the assistant is updated the same moment — no retraining, no separate script to maintain.