# Understanding AI Attributes

> 

Source: https://help.kustomer.com/en_us/understanding-ai-attributes-rydWQsOtGg

Last updated: 2026-09-16T23:23:39.472Z

AI Attributes use AI to populate attributes in conversations, companies, or customer profiles, reducing the need for agents to tag them manually. This article introduces the three types of AI Attributes, explains how instructions shape their output, and covers common use cases.

**Who can access this feature?**

**User types**

### In this article

*   [What AI Attributes do](#what-ai-attributes-do)
*   [Types of AI Attributes](#types-of-ai-attributes)
*   [How AI Attribute instructions work](#how-ai-attribute-instructions-work)

### What AI Attributes do

AI Attributes read conversation content and populate attribute values based on instructions you provide. Instead of relying on agents to select values manually, the AI interprets the conversation and applies the appropriate value to the field.

### Types of AI Attributes

AI Attributes fall into three types, each drawing on a different source of data.

#### Conversation intelligence

Conversation Intelligence uses AI to evaluate information in Kustomer and automatically populate fields with the results with one of three types:

*   **Classifiers**
*   **Scores**
*   **Extractors**

See Understanding Conversation Intelligence for more details.

#### Company intelligence (Coming soon)

Company Intelligence uses AI to evaluate information in Kustomer and populate date fields.

#### Customer intelligence (Coming soon)

Customer Intelligence uses AI to evaluate information in Kustomer and populate date fields.

### How AI Attribute instructions work

Most AI Attributes require an instruction. The instruction tells the AI how to interpret the conversation and how to apply the value to the field. Instructions determine which behaviors the AI prioritizes, which it ignores, and what it does when a conversation doesn't clearly match any value.

Effective instructions share three qualities:

*   They are specific rather than general.
*   They describe behavior — what the AI should look for and what it should do.
*   They state how to handle edge cases explicitly.
