Skip to main content
Conversation tags categorize conversations so your team can filter and prioritize them. They do two separate jobs, and the second one is the important one: a tag can also decide whether the AI agent is allowed to answer at all. Tags are configured at the workspace level under Settings > Conversation Tags, not per agent. Every agent in the workspace shares the same set.

​
Default and custom tags

Your workspace ships with a set of default tags such as Payment Issue, Product Inquiry, and Technical Issue. Beyond those, you can create your own. Custom tags are how you capture the situations specific to your operation that no default will cover, for example a guest reporting that the property has run out of supplies, a claim being opened, or a lock code failing.

​
How a tag gets applied

The tag description is what tells the agent when to apply the tag. A vague description produces noise, so it is worth writing properly.

​
Writing tag descriptions

A vague description leads to over-tagging, where the tag fires on conversations it should not, or under-tagging, where it misses conversations it should catch. Write it so the agent knows exactly when to apply the tag and when not to.
  • Be specific about what qualifies. Describe the exact situation, not a general category.
  • Include what should not be tagged. Call out the most common false positives so the agent does not over-apply the tag to similar but unrelated conversations.
  • Use the guest’s own words. Give a few example messages so the agent recognises the language guests actually use, not just the abstract category.
  • Keep one tag to one purpose. If a description covers two distinct situations, split it into two tags.

​
Escalation level

Every tag carries an escalation level, and this is the control that matters most.
An urgent tag holds the agent back even on autopilot. This is the mechanism to reach for when a situation must always be handled by a person, rather than trying to instruct the agent not to answer.

​
A worked example

A guest messages to say the property has run out of toilet paper. This is not a question you want answered from a knowledge base, because the useful information behind it is internal, such as which supply closet to open and its access code. Nor do you want the agent inventing an answer. The fix is a tag:
  1. Create a custom tag, for example Supplies Missing
  2. Write a description covering what qualifies, such as a guest reporting that consumables at the property have run out, and what does not, such as a guest asking whether an amenity exists
  3. Set the escalation level to urgent
The agent now recognizes the situation, stops, and hands it to your team.
Tagging controls whether the agent answers. It does not control what the agent can read. If you have internal-only material such as supply closet codes, owner commissions, or management fees, keep it out of the guest agent’s reach by scoping where it lives rather than relying on a tag. See Knowledge Base.

​
Creating a tag

  1. Go to Settings > Conversation Tags
  2. Select Add tag
  3. Enter a short, descriptive Tag Name, for example Maintenance Issue
  4. Write a Tag Description covering when the tag applies and when it does not
  5. Set the escalation level
  6. Test it before relying on it
Use the Test Tags feature to send sample messages that should trigger the tag and sample messages that should not. Refine the description until both come out right. A tag that fires on everything is worse than no tag, because your team stops trusting it.

​
Where tagged conversations show up

The tag is visible on the conversation in the list and in the sidebar, and you can filter by it. Conversations carrying an urgent tag also gather under Escalations in the inbox. See Escalations and AI tags.

​
Conversation tags and AI flags are not the same thing

Both appear on conversations, so they are easy to confuse.

​
Frequently asked questions

No. Conversation tags live at the workspace level and are shared by every agent in the workspace.
A tag is workspace-wide, applied by the agent based on what the message says, and can stop the agent from replying. A label is inbox-scoped, sits on the contact rather than the message, is applied by hand or by a reservation rule, and never affects what the agent does.
On copilot nothing sends without a person anyway, so the visible effect is that the conversation surfaces under Escalations. The escalation level matters most once you move to autopilot, where it is what keeps the agent from answering.
Yes, from the conversation sidebar. Manual tagging is useful for situations the agent has no reliable way to detect.
The description, not the tag. Add explicit exclusions for the cases it is wrongly catching, then retest. See Writing tag descriptions.