Executive Summary
When using Zendesk AI, you get a fully managed AI service where Zendesk handles model selection and orchestration under the hood. No, you cannot manually choose or swap models (like toggling between GPT-4o, Claude 3.5, or Gemini) within the native Zendesk interface.
Regarding data storage: your core data remains in your designated Zendesk regional data center, while any generative AI prompt data sent to external LLMs is processed with Zero Data Retention (ZDR)—meaning it exists transiently in memory to generate a response and is never stored or used to train public AI models.
1. Can You Select Which AI Model to Use?
Short Answer: No (For Native Zendesk AI)
Native Zendesk AI (including Zendesk AI Agents, Agent Copilot, and QA insights) functions as a managed solution. Zendesk routes customer queries through a multi-model architecture combining:
Proprietary Zendesk Models: Fine-tuned specifically on CX interactions for intent detection, sentiment, and automated routing.
Third-Party LLMs: Models from vendors like OpenAI, Google (Gemini), and LLMs hosted on infrastructure like AWS Bedrock or Microsoft Azure.
Zendesk automatically determines which model handles a specific task to optimise accuracy, speed, and safety. There is no admin setting or dropdown menu to force a specific foundation model.
The Workaround: Third-Party AI Platforms
If selecting specific models (or bringing your own LLM API keys) is a hard requirement for your compliance or technical team, you can integrate a third-party agentic platform (such as ClearFeed, Lorikeet, or custom API bots built via Sunshine Conversations). These platforms treat Zendesk as the ticketing system of record while allowing you to choose and manage the underlying models.
2. Data Location & Storage Architecture
Data handling for Zendesk AI is governed by two separate mechanisms: where ticket data lives and how AI models process that data.
Where is your data stored?
Primary Service Data: Your support tickets, customer details, end-user data, and knowledge base articles remain stored in the Zendesk Data Center region you selected when setting up your account (e.g., United States, EU, Australia, etc.) under Zendesk's standard Regional Data Hosting policy.
How is AI prompt data handled?
Zero Data Retention (ZDR): When Zendesk sends a prompt to direct third-party LLM providers (e.g., OpenAI, Google), it uses enterprise ZDR endpoints. Data exists only transiently in memory to serve the immediate request and is purged immediately afterward. It is not written to disk or stored in logs by the LLM vendor.
Hosted Cloud Infrastructure: Many generative features run on models hosted directly inside isolated cloud environments (such as AWS Bedrock or Microsoft Azure). Under this architecture, third-party model vendors never touch or access your Service Data.
3. Zendesk AI Agreements, Privacy & Compliance
Your use of Zendesk AI features is governed by your primary Zendesk contract, backed by specific AI Trust and Data Privacy commitments.
| Policy Area | Commitment Details |
| Model Retraining | Zendesk contractually guarantees that your data is never used to train public third-party LLM models. |
| Legal Framework | Covered under your Main Services Agreement (MSA), Data Processing Agreement (DPA), and Supplemental Terms for AI. |
| Security Standards | Certified under ISO/IEC standards, SOC 2 Type II, and aligned with the NIST AI Risk Management Framework. |
| HIPAA & GDPR | Supports HIPAA compliance (with an executed BAA) and offers native GDPR tools, including automated PII redaction and configurable data deletion schedules. |
Key Takeaway: Zendesk prioritises enterprise security and zero data retention over granular customisation. You receive strong data protection and compliance out of the box, but lose the ability to pick specific foundation models natively.
Comments
Please sign in to leave a comment.