> ## Documentation Index
> Fetch the complete documentation index at: https://docs.aide.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Topics and how Aide tags conversations

> Topics describe what a customer message is about: "Where is my order", "Return request", "Discount code not working".

Topics describe what a customer message is about: "Where is my order", "Return request", "Discount code not working". Aide tags every incoming customer message with a topic, on chat and email, including chats your team handles. Scenarios, routing rules, analytics and the AI Agent all use them.

## How topics are organised

Topics live in a three-level taxonomy on the **Topics** page:

* a **category** is the top level, such as "Orders & shipping"
* a **sub-category** groups related topics, such as "Order status & tracking"
* a **topic** sits in a sub-category, such as "Late delivery"

Deleting a category or sub-category deletes what's inside it, and scenarios using those topics stop matching.

## Create a topic

1. Open **Topics**. If you have none yet, start with **Create a category**, then add a sub-category to it.
2. Use **+** on a sub-category to add a topic.
3. Give it a descriptive name and fill in **When to use this topic**: Aide reads it to decide which messages belong to the topic. The more specific, the better the match.
4. Add a couple of examples.

## Suggested topics

You don't have to write your topics up front. When none of your topics fits a customer message, the message shows a dashed **Suggested** chip with a topic Aide proposes. Open it to check the name, description and group, then choose **Add topic**. Adding it tags every message Aide suggested it for and keeps them as training examples. Messages that ask the same thing share one suggestion, and dismissed suggestions aren't proposed again.

Suggested topics don't trigger scenarios or routing until you add them.

## Teach Aide with examples

Each topic has **Examples**. Positive examples are messages that belong to the topic; negative examples look similar but should not be detected as it. A handful of real customer messages is the fastest way to fix a topic that is detected too often or too rarely.

* Add an example by hand from the topic, or use **Add topic** on a message in a conversation.
* Turn an example positive or negative, or move it to another topic.
* Aide flags examples to review: a message that may belong to the topic but has to be accepted, an outlier that sits apart from the others, or one it thinks belongs to another topic.

## Correcting a topic

Each customer message in a conversation shows its topic as a chip. Open it to see how sure Aide was, and mark it **Right topic** or **Wrong topic**: your answer trains Aide. Your team can also mark topics right or wrong from the Aide panel in Zendesk and Front (see [The Aide panel in Zendesk and Front](/integrations/aide-panel-zendesk-front)).

## Confidence

Aide records how confident it is about each topic. Scenarios can use that with a **topic confidence** condition.

## Topics in your helpdesk

Aide can write the topic back to the ticket in Zendesk, Gorgias or Front, so your views, reports and rules can use it. In the integration's inbox settings, under **Topics**:

* **Write topics as tags** adds the topic as a tag, with the **Tag prefix** you choose (for example `aide-refunds`).
* **Topic field** (Zendesk and Gorgias) fills one of your ticket fields, such as a "Reason" dropdown. A dropdown gets the option you map each topic to, otherwise the option named like the topic. A value set by a person is never overwritten.

An inbox in shadow writes nothing to your helpdesk. Front tickets get topics as tags only.

## What each topic shows

A topic's page shows its **AI replies** level, how many conversations it has (select it to open them), when it was **Last seen**, and whether it's **Automatable**. See [Control what Aide is allowed to do for each topic](/aide-ai-agent/topic-ai-levels) for what each level does.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.