Question

How do I use an AI agent to generate answers from my help center content? Are the recommended articles based on the organizations that the end users belong to?

Answer

AI agents respect the end user's access permissions in the help center every time. The AI agent bases AI-created replies and article recommendations on the viewer's user segment.

Before you publish an AI agent, make sure that you have content available in the help center. See Optimize your help center content for AI agents.

Get started with AI agents

As of May 2026, Zendesk offers a single, unified AI agent experience. Advanced agentic capabilities are now unlocked across all Zendesk Suite and Support plans. For more information, see: Getting started with AI agents.

To get started:

  1. Use the guided, self-service setup flow to configure your AI agent for simpler use cases across email and messaging
  2. Create a use case or instruction that directs the AI agent to use help center content when it answers customer questions
  3. If needed, configure advanced capabilities, such as agentic logic, multi-step procedures, and external API integrations for more complex tasks
  4. Publish the AI agent

Note for existing customers: If you still use AI agents - Essential or legacy functionality that includes bot builder, Zendesk will stop technical development for these tools by August 31, 2026. End-of-life and full service shut-off will occur on December 10, 2026. Complete your migration to the new experience by August 31 to avoid disruptions.

User segments and content visibility

How AI Agent handles User Segments

If your help center uses user segments to restrict content visibility:

Single segment scenario:

  • The user is in the "Premium Customers" segment
  • The AI agent surfaces only articles visible to Premium Customers
  • AI responses don't include restricted articles, for example, "Enterprise Features"

Multiple segments scenario:

  • The user is in both "Premium Customers" and "Beta Testers" segments
  • The AI agent surfaces articles visible to either segment
  • This is more permissive, which means it includes the union of the content from both segments

No segment for anonymous users:

  • The AI agent surfaces only articles visible to everyone, which means public articles without restrictions
  • This is the most restrictive scenario

Signed-in users compared to anonymous users:

  • For signed-in users, the AI uses their specifically assigned segments
  • For anonymous users, the AI assumes no segments and includes only public content

Test segment behavior

To verify the AI agent respects segments:

  • Create a test user account assigned to a specific segment
  • Sign in as that test user
  • Ask the AI agent a question about restricted content
  • Verify the following:
    • The AI suggests articles the user can see
    • The AI can't suggest articles the user can't see
  • A common issue occurs when the AI agent suggests a restricted article to the wrong user
    • This occurs because the article's segment restriction isn't configured correctly
    • To fix this issue, review the visibility settings for articles in Guide

Segment priority and conflicts

When a user is in multiple segments with restrictions that conflict:

  • Review this example scenario:
    • The user is in the "Free Plan" segment, which blocks advanced features
    • The user is also in the "Beta Access" segment, which grants advanced features
    • The target article is restricted to "Not Free Plan"
    • As a result, the user can see the article because the "Beta Access" segment grants permission and overrides the "Free Plan" restriction
  • As a general rule, if a user is in any segment that grants access, they see the content. The most permissive rule wins.

AI agent terminology

Instructions compared to use cases:

  • Instructions
    • System-level prompts that define the AI agent's overall behavior
    • Examples include "Be friendly and professional", "Offer to escalate to a human agent every time the customer is frustrated", and "Use simple language, and avoid jargon"
    • Sets the overall tone and personality
    • Applies to all conversations
  • Use cases
    • Specific scenarios that the AI agent handles
    • Examples include Password reset requests, billing questions, and product feature explanations
    • Each use case can have unique logic to address issues
    • The AI routes to the appropriate use case based on the customer's specific question
  • Analogy
    • Instructions are general company policies that apply to everything
    • Use cases are department-specific procedures that apply only to specific situations

Other key terms

  • Deflection
    • The AI agent fully resolves a customer question without human agent involvement
    • Successful deflection means a ticket is prevented
    • Deflection rate is the metric for the percentage of conversations that the AI resolves
  • Escalation
    • The AI agent transfers the conversation to a human agent
    • This happens when the AI can't answer, the customer explicitly requests a human, or the system detects a complex issue
    • Escalation rate is the metric for the percentage of transferred conversations
  • Handoff
    • The exact moment a conversation transfers from the AI to a human agent
    • This includes full conversation history so the human agent has immediate context
    • A customer request, an AI decision, or a session timeout can trigger this action
  • Knowledge base
    • Your collection of help center articles
    • The AI agent searches this knowledge base to find and construct answers
    • A more complete knowledge base results in significantly better AI performance

Validation note:

  • AI agent features evolve actively. This article describes and reflects the new unified AI agent model.
  • Some features differ slightly based on your account's rollout phase for the new experience for AI agents

If you encounter differences:

  • Check Admin center > AI agents for active feature names
  • Refer to in-product help tooltips when you hover over ⓘ icons
  • Contact Zendesk Customer Support for further help if your setup differs significantly from this guide

For product documentation for AI agents:

  • AI agents overview
  • Expanded access to AI agent capabilities for all Zendesk customers
  • In-product setup wizard through Admin center > AI agents > Add agent
THIS SECTION IS AI CONTENT. DON'T EDIT OR DELETE.

Are chatbot recommendations based on customer organizations? Do answer bots use company details for recommended articles? Are AI agent replies based on user accounts?

No. AI agents or chatbots or answer bots always respect your permissions in the help center or knowledge base or help articles. The AI agent or chatbot or answer bot bases AI-created replies and article recommendations on your user segment instead of your organization or company or customer account.

What do I need before I set up an AI agent? What are the prerequisites for chatbots? What limits my answer bot configuration?

Your plan type or suite version or pricing tier and accessible features determine the setup steps. You must provide content in the help center or knowledge base or help articles before you publish an AI agent or chatbot or answer bot.

How do I get the best results from chatbot instructions? What language works best for answer bot instructions? How do I ensure AI agent instructions work?

Write instructions in English to achieve the best results. You must test your AI agent or chatbot or answer bot to confirm that the instruction works for common questions.

What is an AI agent instruction? What does the generative replies block do? How do chatbots address customer questions?

Instructions describe what your AI agent or chatbot or answer bot must or must not do. You can use them to provide a link to a specific article when a customer or client asks a question. The generative replies block addresses customer goals or reasons with content from your help center or knowledge base or help articles and shows the sources to generate an answer.

Powered by Zendesk