Knowledge Base Preparation

Structure and connect your knowledge sources so TR-AI can index them accurately.

TR-AI's resolution quality is directly proportional to the quality of your knowledge base. A well-structured KB with clear, specific articles will produce high-confidence resolutions. A sparse or poorly organised KB will produce more triage escalations.

Supported KB sources

Source	How it works	Best for
Website / help centre	TR-AI crawls your site starting from the root URL, following internal links up to 1,000 pages	Zendesk Guide, Freshdesk Solutions, custom help sites
Confluence	Connects via API token; indexes all pages in the specified space	Internal wikis, engineering runbooks
Notion	Connects via integration token; indexes all pages shared with the integration	Product docs, SOPs, onboarding guides

What makes a good KB article

One topic per article

Each article should answer one specific question. Avoid combining multiple unrelated topics in a single page — the vector search works best when each chunk is semantically focused.

Clear, descriptive titles

Titles are weighted heavily in the similarity search. Use titles that match how customers phrase their questions:

Good: "How to reset your password"
Good: "Why is my invoice showing the wrong amount?"
Avoid: "Password" or "Billing issue"

Concrete answers, not just navigation

Articles that say "contact support for help with X" are not useful to the AI. Include the actual steps, values, or explanations that resolve the issue.

Keep articles under 1,500 words

TR-AI chunks articles into ~400-token segments for indexing. Very long articles are chunked automatically, but shorter, focused articles tend to produce cleaner matches.

Content to exclude

Some content should be excluded from the KB to avoid confusing the AI:

Release notes and changelogs (version-specific, frequently outdated)
Marketing pages and pricing pages
Legal documents (terms, privacy policy)
Navigation-only pages with no substantive content

Tip:If your help centre mixes support articles with marketing content, use the KB URL to point at the support subdomain only (e.g. https://help.yourcompany.com rather than https://yourcompany.com).

Keeping the KB up to date

TR-AI re-crawls your KB automatically on a schedule. You can also trigger a manual re-index from the Dashboard → Knowledge Base card → Re-index button.

After adding new articles or making significant edits, trigger a manual re-index so the changes are reflected immediately.

Note:A re-index rebuilds all embeddings from scratch. Tickets that arrive while a re-index is running are routed directly to your triage group (routed_to_agent outcome) to avoid low-quality resolutions.

Minimum viable KB size

TR-AI can work with as few as 10–20 articles, but resolution rates improve significantly with 50+ focused articles covering your most common ticket types. Review your last 90 days of tickets and make sure the top 20 ticket subjects each have a corresponding KB article.

Confluence setup

Required Confluence credentials

Base URL:    https://yourcompany.atlassian.net
Space key:   SUPPORT  (or whichever space holds your KB)
Email:       your-atlassian-account@company.com
API token:   generated at id.atlassian.com → Security → API tokens

Notion setup

Required Notion credentials

Integration token:  secret_xxxxxxxxxxxxxxxxxxxx
  → notion.so/my-integrations → New integration → Internal
  → Copy the "Internal Integration Token"

Root page ID:  the page ID from the URL of your top-level docs page
  → notion.so/{workspace}/{page-id}

Share the root page (and all sub-pages) with your integration:
  → Open the page → Share → Invite → select your integration

PreviousQuick start guide NextGo-live playbook