Training Best Practices

Best practices for training your agent on company knowledge

The quality and relevance of your training data directly impacts how well your agent performs. This guide covers what to train on, what to avoid, and how to get the best results.

Where to connect your knowledge

Source	What it provides	Where to connect
Help center	Published support articles	Zendesk, Freshdesk, Intercom, Gorgias, Jira
Past tickets	Your team's resolved conversations	Zendesk, Freshdesk, Intercom, Gorgias, Jira
Macros/Canned responses	Pre-written response templates	Zendesk, Freshdesk, Gorgias
Google Docs, Sheets, PDFs	Documents from Drive	Google Drive
Notion pages	Wiki and knowledge base content	Notion
Confluence pages	Team documentation	Confluence
Website pages	Any public web content	Website Crawler
Files	PDFs, docs, CSVs, or custom content	Files

How syncing works

Once connected, eesel AI keeps your knowledge up to date:

• Automatic re-sync — Content is periodically refreshed to pick up changes

• Manual sync — You can trigger a re-sync at any time from the integration settings

• Status indicators — The sidebar shows sync status (syncing, complete, failed)

What makes good training data

Help center articles

Your help center is usually the best starting point — it's already written for customers and covers common questions.

• Make sure articles are up to date before syncing

• Published articles work better than drafts

• Well-structured articles (clear headings, short paragraphs) produce better answers

Past tickets

Training on resolved tickets helps your agent sound like your team and handle real-world scenarios.

• Focus on tickets with good resolutions — your agent learns from the quality of replies

• Filter by status (resolved/closed) to avoid training on incomplete conversations

• Recent tickets are more valuable than old ones

Internal documents

Confluence, Notion, and Google Drive docs give your agent internal context that help center articles might not cover.

• Start with your most frequently referenced documents

• Avoid training on out-of-date or deprecated content

• Organize documents clearly — the agent uses document structure to understand context

Website content

Crawling your website is a quick way to add product information, pricing, and FAQs.

• Use path filters to include only relevant sections

• Exclude marketing pages, careers pages, or other non-support content

• The crawler respects robots.txt

Best practices

More relevant knowledge > more knowledge. A focused set of high-quality articles works better than dumping everything. Irrelevant content can confuse the agent.

Keep sources up to date. If your help center articles are outdated, your agent gives outdated answers. Update the source and re-sync.

Combine source types. Help center + past tickets + internal docs gives the most complete coverage:

• Help center → what to answer

• Past tickets → how to sound

• Internal docs → extra context and edge cases

Test after adding sources. After adding new knowledge, test in the chat to make sure the agent finds and uses it correctly.

Monitor unanswered questions. Use Analytics to find questions your agent can't answer, then add knowledge to fill those gaps.