Train AI Agent on website content

• Open Agent Setup > Sources > Prompts to tell your AI Agent how to behave or what to keep in mind, without a file or a website.
• Write a prompt, or record a voice note: it is transcribed, and you read and correct the text before it is saved.
• Your AI Agent follows every prompt that is on in every reply, alongside its other sources. An AI Agent holds up to 30 prompts and 8,000 characters of them in replies.
• Edit, turn off or delete a prompt at any time. Use a Plain text source for long documents, so they are searched only when a question needs them.

• Use website crawl sources for public pages that answer visitor questions directly.
• Use manual Q&A for exact answers that are missing from the website or need controlled wording.
• Use file and document sources for policies, menus, service guides, product sheets, and support material that should not be copied into a web page yet.
• Review source references in conversations so weak pages can be rewritten or replaced.

• Add one source type at a time from the training page.
• Review source status, character count, knowledge sections, crawl pages, and errors.
• Retrain stale sources after source content changes.
• Delete old sources when answers should no longer use that knowledge.

Training uses global upload and crawler safety caps. Website crawlers reject local and private network targets.

Use manual Q&A for exact gaps found in analytics. Prefer public help pages and clean policy documents over noisy source material.