A conversation attribute is a value stored on a conversation record alongside the transcript: what the conversation was about, how likely the customer is to leave, which plan they chose, whether a payment link was sent. Attributes turn conversations into something you can filter, count and export on your own business terms.
Two kinds sit under Project Settings > Conversation Attributes, one on each tab. This page covers what they have in common: what switches them on, and the places their values appear. Each kind has its own article for how you define it.
The two kinds
| AI Insight | Custom Attribute | |
|---|---|---|
| Where the value comes from | A language model reading the finished transcript | The flow itself, through a Set Attribute node |
| When it is decided | After the conversation has ended | At the point in the flow where the answer is already known |
| Work needed in the flow | None | A node on the branch that knows the answer |
| What it can tell you | An interpretation of what happened | Only what the flow explicitly wrote |
The difference matters for both cost and trust. An insight is inferred, so it can describe things nobody designed the flow to record, but it costs a model call on every conversation and it is an interpretation, which means it can be wrong or come back empty. A custom attribute costs nothing extra and is exact, but it exists only for conversations where the branch carrying the node actually ran. If the flow already knows the answer, record it as a custom attribute, and keep insights for questions only a reader of the whole conversation could answer.
Once recorded, the two behave identically. Same conversation record, same columns, same filters, same export.
Switching it on
Conversation Attributes is available in the Premium edition only. On a Basic or Standard tenant the section does not appear in Project Settings. On a Premium tenant, each project has its own Conversation Attributes switch on the General tab of Project Settings, and that one switch governs both kinds. While it is off, nothing is recorded, and no values are shown or exported on the Conversations and Datasets pages.
📝 Note: The switch stays disabled until the tenant's Add-On AI Features LLM configuration is filled in. This holds even if you intend to use custom attributes alone: the AI Insights switch is the only thing that turns them on, so the LLM configuration has to exist first.
Turning the switch back off later hides the values and stops new ones being recorded, but it does not delete what was already recorded. Switch it on again and the earlier values are back on the records.
Definitions on both tabs belong to the project version and are edited on the Draft version, by someone with permission to edit projects. On a published version both tabs are read only, and a new or changed definition reaches live conversations only after the project is published.
Where the values appear
On the conversation record. Open Insights > Conversations and click a conversation's Date link. The detail carries an AI Insights block and a Custom Attributes block, each listing one entry per recorded value. Both blocks are built from what was stored on that conversation, in the order the definitions were in at the time.
As columns on the Conversations page. Every enabled attribute and insight is available as a column on the conversation list. Open Column options (the wrench icon immediately left of Export), add the ones you want, and save.
📝 Note: These columns are hidden by default and the choice is per user, so after defining attributes the Conversations page looks exactly as it did before until somebody switches the columns on. Switching them on changes only that person's view. The columns cannot be sorted: clicking the header does nothing.
As filters. The filter panel on the Conversations page carries an AI Insight Filters section and a Custom Attributes section. Add a filter, choose the definition, give it a value, and the list re-queries. The same filters are available when you build a dataset on the Datasets page of AI Testing.
📝 Note: With several projects selected in the Conversations filter, whether values load at all is decided by the first project in the list. If that project has the switch off, no values appear for any of the selected projects. Put a project that has it on first, or filter one project at a time.
In the export. The conversation list export carries one column per currently defined insight and attribute, in the definitions' order. A single conversation export carries the values stored on that conversation.
Values are stored with the conversation, so they are removed with it when a data retention policy expires the session. A report built on attribute values reaches back only as far as your retention window, however long the definitions have existed.
Next
- AI Insights: the built-in insights, defining your own, output types, and when they run.
- Custom Attributes: defining a key, the Set Attribute node, and writing the value from a flow.