Guides

Training Your Agent

Best practices for adding content and improving your AI agent's response quality

Your AI agent is only as good as the knowledge you give it. This guide covers how to build a high-quality knowledge base that enables accurate, helpful responses.

How Your Agent Learns

FutureBase agents don't "learn" in real time — they reference the knowledge base you build. When a customer asks a question, the agent searches your content for the most relevant information and uses it to generate an answer.

This means the quality of your content directly determines the quality of your agent's responses.

Adding Content

There are three main ways to add knowledge to your agent:

1. URLs (Website Crawling)

The fastest way to get started. Point your agent at your website and it will automatically crawl and index the content.

Go to your agent's Knowledge Base section and select the URLs tab.

Add your website URL

Enter your website's URL. FutureBase will automatically discover and crawl linked pages from your sitemap or by following internal links.

Review indexed pages

After crawling completes, review the list of indexed URLs. You can remove pages that aren't relevant to customer support.

FutureBase automatically re-crawls your URLs periodically so your agent stays up to date with content changes.

2. FAQs

FAQs are the most effective content type for customer support. They give your agent direct question-answer pairs, which often produce the most precise responses.

Tips for writing great FAQs:

  • Write questions the way customers would actually ask them
  • Include variations — "How do I cancel?" and "How to cancel my subscription" can both be useful
  • Keep answers concise but complete
  • Group related questions together

3. Sitemaps

If your website has a sitemap (/sitemap.xml), you can submit it directly to discover and index all your pages at once instead of entering URLs manually.

Enter your sitemap URL

Enter the full URL to your sitemap (e.g., https://yoursite.com/sitemap.xml). FutureBase will parse the sitemap and enqueue all discovered URLs for crawling.

Review discovered pages

After crawling completes, review the indexed URL list and remove any pages that aren't relevant.

XML sitemaps and sitemap index files (a sitemap of sitemaps) are both supported.

4. File Uploads

Upload documents that contain relevant information. This is useful for:

  • Product documentation
  • Policy documents
  • Technical specifications
  • Training manuals

Supported file types: PDF, TXT, MD, DOCX

File size limit: Up to 5 MB per file on Hobby, up to 25 MB per file on Pro.

Improving Response Quality

Once you have content in your knowledge base, here are some strategies to improve responses.

Fill Knowledge Gaps

Review your Conversations tab to see what customers are asking. If your agent is struggling with certain topics, add content that covers those areas.

Be Specific

Vague content leads to vague answers. Instead of:

"We offer flexible pricing plans."

Write:

"We offer three pricing plans: Starter at $9/month, Pro at $29/month, and Enterprise with custom pricing. All plans include the AI chat widget and basic analytics."

Keep Content Current

Outdated information is worse than no information. Regularly review and update your knowledge base, especially after:

  • Product or pricing changes
  • New feature launches
  • Policy updates
  • Seasonal promotions

Use the Playground

Before going live with changes, test your agent in the Playground. Try asking questions that relate to the content you just added to verify the responses are accurate.

Avoid adding duplicate or contradictory content. If two pieces of content give conflicting answers, the agent may produce inconsistent responses.

Content Priority

When adding content, prioritize in this order:

  1. FAQs — highest signal-to-noise ratio
  2. Core product pages — features, pricing, how-to guides
  3. Support documentation — troubleshooting, setup guides
  4. Policy pages — refunds, shipping, privacy
  5. Blog posts — useful for product context, but lower priority

On this page