Skip to main content

​
Overview

A well-built knowledge base is critical for accurate, domain-specific AI responses. The knowledge base provides context on a per-interaction basis — the system compares the contact’s message against your knowledge base to generate specific, relevant answers.
The AI doesn’t “know” your knowledge base — it uses it as contextual input to generate responses. The quality of output is directly tied to the quality of input.

​
Key Concepts

  • Knowledge as context — Information is injected per interaction, not memorized
  • Best content types — FAQ pairs, objection handling, cause-effect relationships
  • Quality in = quality out — Poor or unclear data leads to poor responses

​
Best Practices

​
Preferred Input Types

  1. FAQ format — Precise Q&A pairs deliver the most reliable results
  2. Raw text — Sourced from your website, documentation, or transcripts
  3. Document uploads & website scrapes — Useful but may introduce conflicting information
Source raw text from your website, YouTube transcripts, or documentation. Use ChatGPT or Claude to generate FAQs from this material — tailored for sales, support, or in different languages.

​
Optimization Tips

  • Use FAQ format for common questions — best accuracy
  • Keep entries focused — One topic per entry
  • Update regularly — Outdated info causes wrong responses
  • Test thoroughly — Ask the AI questions requiring knowledge base data and verify
  • Use specific language — Be precise about services, pricing, policies
  • Organize by category — Group related entries
  • Set temperature low (0–0.2) for businesses needing precise, deterministic answers

​
Monitoring Your Knowledge Base

In conversation logs, click the { } bracket icon to open transparency logs. Look for:
  • “Embedding” — The knowledge base was queried
  • “Embed complete” — Open this to see exactly what content was returned for the contact’s message
If the returned content isn’t ideal, go back to your knowledge base and add a targeted FAQ for that question.

​
Architecture: Where Things Go

ComponentPurpose
PromptPersonality, response guidelines, style guardrails, instruction set
Knowledge BaseDomain-specific knowledge, FAQs, objection handling, pricing
ToolsContext injection, conditional logic, data retrieval, actions

​
What About Live Data or Complex Pricing?

Use custom tool calls to fetch live data or integrate with third-party services for complex quoting and pricing operations.

Chat Response Optimization

Optimize response times

Data Extraction

Extract user information

Google Sheets Integration

Use spreadsheets as a knowledge source

Custom Tools

Extend AI with custom tools