Skip to main content
AI is a powerful tool, but it has limitations. Understanding these limitations helps you set appropriate expectations, configure better safeguards, and maintain human oversight where needed.

​
What AI can do well

  • Answer common customer questions using your knowledge base
  • Qualify leads based on conversation data
  • Book appointments by checking calendar availability
  • Route conversations to the right team member
  • Generate content drafts for emails, social posts, and web pages
  • Handle high volumes of simultaneous conversations

​
What AI cannot do

  • Make complex judgment calls that require human empathy or nuanced understanding
  • Guarantee 100% accuracy in every response
  • Access real-time external data beyond what is in your knowledge base and system prompt
  • Handle sensitive situations (legal advice, medical diagnoses, crisis intervention) without human oversight
  • Learn from individual conversations automatically (training requires manual updates)

​
When to involve humans

Configure your AI agents to escalate to a human team member when:
  • The customer explicitly asks to speak with a person
  • The conversation involves a complaint or sensitive issue
  • The AI is not confident in its response
  • A transaction or decision requires human authorization
  • The topic is outside the agent configured scope

​
Transparency guidelines

  • Always disclose to customers that they are interacting with an AI
  • Set realistic expectations about what the AI can help with
  • Provide an easy path to reach a human when needed
  • Be honest about AI-generated content (do not present it as human-written)

​
Next steps

Guardrails

Set behavioral boundaries for your AI

AI safety overview

Our principles for responsible AI
Last modified on March 5, 2026