Your knowledge base mixes content of very different quality — articles you wrote carefully, pages a scan picked up, and drafts the AI wrote itself. Trust tiers make that difference count in search ranking.
The three tiers
| Tier | Multiplier | Assigned to |
|---|---|---|
| Curated | 1.3× | Articles you created by hand, and articles imported from Freshdesk, Zendesk, or Intercom |
| Standard | 1.0× | Pages found by the website scan |
| Generated | 0.8× | Articles drafted by AI from a knowledge gap |
The multiplier applies to the combined search score. Two articles that match a query equally well will rank curated first, then standard, then generated.
Why generated content is penalised
AI-drafted articles are useful — they turn a question your agent couldn't answer into a starting point. But they're written from your existing content, so they can amplify an error rather than correct it. Ranking them below human-reviewed content means a real article always wins when one exists.
Generated drafts are also created unpublished. Nothing AI-written reaches your customers until you've read and published it.
Where tiers matter beyond search
- Auto-replies. A drafted reply sourced only from generated content is never sent automatically.
- The AI surface.
llms-full.txtand the MCP server serve curated and standard content only — generated drafts are excluded, so external assistants never quote unreviewed AI text back at your customers.
Changing a tier
The tier records where content came from, and it doesn't change when you edit. Reviewing and rewriting an AI draft improves the article, but it stays in the generated tier and keeps the 0.8× weighting.
If you want a piece of knowledge to carry full curated weight, create it as a new article and write it in your own words. Editing an AI draft is still worth doing — a good draft that ranks slightly lower is better than a bad one — but for anything important, a hand-written article is what you want.