Best practices

A chatbot is only as good as the content and configuration behind it. These practices help you ship an assistant that's accurate, on-brand, and genuinely useful.

Curate your sources

• Quality over quantity. Add the pages and documents that actually answer customer questions. Crawling everything dilutes results with stale or irrelevant content.
• Keep sources fresh. Re-crawl websites and re-upload documents after major changes so answers reflect your latest information.
• Remove duplicates and outdated versions. Two documents that disagree produce confident but wrong answers. Run KB Health to catch these.

Write a clear persona

Set a custom prompt that defines who the assistant is, what it should help with, and what it should decline. Be specific about tone (formal vs. friendly) and scope (e.g., "only answer questions about our product").

Guide the first message

• Use a welcome message that states what the bot can do.
• Add suggested questions that showcase your best use cases — they set expectations and reduce dead-end first messages.

Trust, but verify with citations

Keep citations on. They let users confirm where an answer came from and help you spot when the bot is reaching for the wrong source.

Test like a user

Before publishing, ask the questions your customers actually ask — including edge cases and rephrasings. Review chat logs after launch to find weak answers, then improve the underlying sources.

Match access to audience

• Customer-facing? Set it Public and embed it.
• Internal only? Keep it Private so it stays within your workspace.

Measure what matters

Watch statistics for conversation volume, satisfaction (thumbs up/down), and top topics. Falling satisfaction usually points to a content gap you can fix.