Settingsbeginner

Enable AI auto-tagging on a tag

Turn on the AI Auto-Tag switch so Sidekick applies a tag to conversations whose content matches the tag's description. The description is the only signal — write it carefully.

4 min read

Enable AI auto-tagging on a tag

When AI auto-tagging is on for a tag, Sidekick is allowed to apply that tag automatically as a conversation evolves. The tag’s description is the only thing the AI matches against — there’s no separate prompt or rule.

This is the difference between a quiet inbox where every conversation is correctly categorized and one where agents have to remember to tag every case.

Before you start

Steps

  1. Open Settings → Tag Management.
  2. Find the tag and open its edit panel.
  3. Make sure the Description is filled in — see Write a good description below. Tags without a description are skipped by the auto-tagger entirely, regardless of the switch.
  4. Toggle AI Auto-Tag on.
  5. Save.

That’s it. Within minutes, new messages on conversations begin running through the auto-tag worker; when a conversation’s content matches the description well enough, the tag is applied automatically.

How it actually works

  • When the switch is on for a tag, Atender embeds the tag’s description into a vector and stores it.
  • When new messages arrive on a conversation, an auto-tag job is queued with a short debounce — rapid back-and-forth turns into a single evaluation rather than many.
  • The worker fetches the new messages plus a rolling per-conversation summary it maintains, runs a vector search against the auto-tag-enabled tags, and asks the model in a single call to (a) update the summary and (b) pick which of the candidate tags apply.
  • Selected tags are written to the conversation. Nothing is removed automatically — auto-tagging adds, never strips.

Write a good description

The description is the only thing the AI sees. Write it for the AI, not for agents.

Better:

Customer wants to return a product. Mentions return, refund, exchange, sending back, or doesn’t want it anymore. Includes order numbers being returned, defects, wrong items, size issues.

Worse:

Returns.

Worse still:

(empty)

Two practical rules:

  • Lead with the customer’s situation in plain language. What is the customer expressing or asking for?
  • Include common phrasings. Synonyms, slang, multilingual variants if your inbox is multilingual.
  • Don’t include workflow notes (“send to finance team”). Those belong in automation rules. The description is purely about when this tag applies.

Verify it worked

Send a test message to one of your inboxes that clearly matches the description (e.g., “I’d like to return this — it arrived defective”). Within a minute or two the tag should appear on the conversation.

If it doesn’t:

  • Check that the description is non-empty and specific. An empty or one-word description disables auto-tagging on that tag.
  • Confirm the switch is actually on (it might have failed to save).
  • Try with a more obvious test message — the auto-tagger is conservative and skews toward not over-tagging.

Pairing with automations

The combination is the real power:

AI tags the conversation urgent-billing → an Automation listening for that tag assigns it to the billing team lead and applies a strict SLA.

The AI handles the classification. Automations react. Neither has to know about the other.

See also

Tags

Ai FeaturesHow To
Back to Settings