How does a support team get its AI to use what agents already solved in past tickets?

Short answer

Not by pointing the AI at old tickets. In Zendesk, AI agents answer from connected knowledge sources: your help center and external content. What an agent solved in a ticket must first be made available through a connected knowledge source, such as a help center article or external knowledge base. Zendesk's Knowledge copilot can draft articles from ticket data; someone still reviews and publishes them.

Where a support AI actually gets its answers

The answer to a customer is usually already in your ticket history: an agent solved the same problem last month, with a workaround nobody wrote down. The AI does not see that. In Zendesk, as of November 2026, an AI agent generates answers from knowledge sources, and there are two kinds:

Knowledge source What it is How current
Zendesk help center The help center of your account or of one brand Searched live at the moment of the question
External content Content brought in by a web crawler, an API call, or a knowledge connector to another knowledge base As of the last sync, usually every 24 hours

You connect them in the AI agents workspace under Build → Knowledge → Knowledge sources → Edit sources. Without at least one knowledge source, a generative reply produces a technical error. The same external content can also feed help center search, quick answers for agents in Agent Workspace, and, with the Copilot add-on, auto assist.

Tickets are not on that list. That is the gap for a support team.

From solved ticket to something the AI can use

  1. Find the gaps. Zendesk’s Knowledge copilot, in early access since May 2026 for Suite Professional and above, shows a knowledge base health page with coverage, freshness and “AI readability”, and recommends articles to create or update based on ticket trends. You enable it under Admin Center → AI → Knowledge → Knowledge copilot.
  2. Draft from tickets. Its Article builder generates article drafts from ticket data or a prompt. Its Procedure builder drafts auto assist procedures, which requires the Copilot add-on.
  3. Review like an editor. A ticket contains one customer’s situation. An article has to state when the fix applies and when it does not. That judgement is the support lead’s, not the model’s.
  4. Publish with the right visibility. Restricted articles are used only for authenticated customers; internal workarounds belong behind that restriction.
  5. Keep sources focused. Zendesk notes that too many knowledge sources can reduce accuracy and increase latency.

The common misunderstanding

Teams assume an AI on top of the helpdesk learns from every ticket agents close. It does not. It learns nothing on its own: the Zendesk AI agent generates knowledge answers from connected help centers and external knowledge sources. Every insight that stays in a ticket, a Slack thread or one agent’s head is invisible to the AI agent and to the next colleague. The work is to turn solved tickets into maintained knowledge, and to decide who owns that.

What this means for a support team

Support knowledge lives in three places at once: the help center (what you tell customers), the tickets (what actually happened), and the heads and chats of experienced agents (what they know works). The Zendesk AI agent generates knowledge answers from connected help centers and external knowledge sources, not from the other two. Agents who also use ChatGPT, Claude or Copilot to draft replies have a fourth place: their own chat history, which nobody else sees. The same holds for other knowledge work, as in how to make company documents searchable for AI and AI draft replies in a shared customer inbox.

Where Bombos fits

Bombos is a plugin that gives every AI tool a team uses one shared memory. For support, that means the fix one agent found, and the rule the team agreed on (“never promise a delivery date for backorders”), is available to the next agent’s ChatGPT, Claude or Copilot as well, with a trace to where it came from. It sits next to Zendesk rather than replacing it, in our cloud or on your own server, and one agent can try it first with their own agents.

Related questions

Can a Zendesk AI agent search old tickets directly?

Zendesk's documentation for generative replies lists help centers and external knowledge sources as the knowledge an AI agent answers from. Ticket data is used to suggest and draft articles, not as a direct answer source.

How fresh is the content the AI agent uses?

For a connected help center the agent searches the current content. For an external source it searches what was available at the last sync, which usually happens every 24 hours.

Do restricted articles leak to anonymous customers?

No. Zendesk states that AI agent answers respect article view permissions: authenticated customers can get answers from relevant restricted articles, unauthenticated ones only from public articles.

How many external sources can we connect?

Up to 50 external content sources per Zendesk account. Zendesk itself recommends keeping the number of sources per AI agent reasonable, because too many can reduce accuracy and add latency.

Sources

  1. Zendesk: Connecting knowledge sources to power generative replies in AI agents
  2. Zendesk: About connecting external content to Zendesk
  3. Zendesk: Announcing the Knowledge copilot Early Access Program
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.