How does AI generative search choose which article to pull information from?

Which data sources does generative search use?

Generative search respects user segments that you set in your Knowledge admin settings.

Quick answers that users see on your search results page will only be from articles that they have permission to access and view in your help center.
Generative AI evaluates your top matching articles and community posts, which then generates an answer.
With federated search, quick answers can also pull from your indexed external content.

When connected to Knowledge, generative search can also use external sources such as Confluence sites and spaces, web‑crawled content from public or authenticated websites, and other external records ingested through the Federated Search API. These sources are managed through Knowledge and are subject to the same access and visibility settings applied in your account.

This means generative answers can be based on information stored in your Zendesk help center, community posts, or approved external content indexed in Knowledge.

Quick answers use generative AI from OpenAI's large language model (LLM). Zendesk leverages OpenAI endpoints for zero data retention. Request and response bodies do not persist in any logging mechanism and exist only in memory in order to serve the request. OpenAI doesn't store any data after the delivery of the output.

OpenAI applies to quick answers in both help center generative search and the Knowledge context panel in the Agent Workspace.
Quick answers in the Agent Workspace are available to customers with the Agent Workspace and Knowledge and do not require the Copilot add-on.

You can also use PDF files as a knowledge source with Knowledge connectors through the PDF support for Knowledge connectors early access program (EAP). You can sign up for the EAP here.

Note: Connect external sources of knowledge to your Zendesk account, such as Confluence.

For more information, see