Plan requirement
| Subscription | Suite Team or higher, Support Team or higher |
| Access | Admin |
Past conversations show how your customers ask. The thing to know before you plan the work: this is a TSV upload with five named columns, not a list of conversations you tick. Building the file is most of the job.
The file
| conversationId | Groups messages that belong to the same conversation. |
| sendTime | ISO 8601, for example 2025-03-22T04:55:51+00:00. |
| senderType | Either agent or customer. |
| text | The message itself. |
| agentId | Optional, and only needed if you want per-agent statistics. |
Import it
- Select your AI agent in the workspace.
- Click the Conversation logs icon in the sidebar.
- Open the options menu at the bottom right and select Import TSV.
- Choose your file.
- Pick the benchmark model: Clustering for an established agent, or Import only for a new one.
- Click Import. The data appears in your conversation logs.
What clustering gives you
AI-based clustering identifies the most frequent topics in what you uploaded. That is a report on your support operation as much as a setup step: the phrasing customers use, and how often each subject comes up.
Zendesk's own guidance frames those topics as a starting point for intents. The intents meant there belong to the legacy AI agent, which Zendesk removes on 10 December 2026, so in the current experience the sensible destination for that list is your use cases, and the articles you have not written yet. Note that this is not the Intent field that intelligent triage fills in on a ticket: that is a different feature and it stays.
Choose the period deliberately
Recent conversations reflect the current product; old ones describe how things used to work. A year is usually the right window, and less if you have changed something significant.
Old answers were not all correct
Historical exchanges include agents who were mistaken, policies since changed, and workarounds no longer needed. Treat the import as a source of questions rather than a source of answers.
What happens to personal data
Zendesk states that all imported data is sanitised and GDPR compliant, and that no personal data is imported.
That covers what happens after the upload. It does not cover the export you build to get there: that file leaves your systems with real conversations in it, so handle it the way you handle any ticket export, and involve whoever owns that decision if the volume is large.
On email you may not need the file at all
Before you build a TSV, check whether your AI agent is on email. If it is, set the Report preset to Custom and the Source to Zendesk Support, and it reads your historical email tickets directly. Public comments only, no private notes, and messaging tickets excluded. Raise the Conversations sample to a thousand or two thousand to take in more of them. That is the whole import, without the export and the column mapping.
Read what came out
The topics it identified are worth more than the import itself. Anything frequent and undocumented is an article to write.
A topic needs at least five conversations before it appears, so a small sample can come back with nothing while the import itself worked perfectly. If you get the message ai agent does not have any conversations, read it as being about the source rather than the agent: the Quick preset looks only at messages classified as not understood, and a busy agent that understands most of what it hears has very few of those.
Comments
0 comments
Article is closed for comments.