Skip to main content
GoGuardian Admin uses a type of AI called a large language model (LLM) to classify student conversations captured by prompt visibility. These AI-assisted classifications help administrators quickly identify conversations that may warrant closer review. Each conversation receives 1 Interaction Type:
  • Appropriate
  • Concerning
  • Needs Context
These classifications help prioritize review. They do not determine whether a student violated school policy, and they do not replace an administrator’s judgment. To learn how to access captured conversations, see Review AI Activity with prompt visibility.

​
Learn How the AI Model Classifies Conversations

The AI model evaluates the intent expressed in the student’s prompts. The AI tool’s responses do not determine the classification. The classification process follows these general principles:
  • The AI model evaluates the student’s expressed intent, not individual keywords alone.
  • Educational or informational conversations can be Appropriate, even when they discuss sensitive subjects.
  • If any part of the student’s conversation clearly reflects concerning intent, the full conversation is classified as Concerning.
  • When the prompts do not provide enough information to determine intent confidently, the conversation is classified as Needs Context.
AI models can make mistakes. Treat each Interaction Type as a way to prioritize review, not as a final determination about student intent or behavior.

​
Understand Appropriate Conversations

An Appropriate conversation is an educational, creative, informational, or productivity-related conversation that supports learning. This can include:
  • Asking for explanations or tutoring
  • Brainstorming ideas
  • Studying or practicing a subject
  • Requesting editing or feedback
  • Researching general information
  • Supporting other legitimate learning activities
A conversation can discuss a sensitive subject and still be classified as Appropriate when the student’s expressed intent is educational or informational.

​
Identify Concerning Conversations

A conversation is classified as Concerning when the student’s prompts clearly indicate intent related to 1 or more of the following areas. Hover over a Concerning Interaction Type in GoGuardian Admin to see the applicable area.

​
Recognize Academic Dishonesty

Requests for an AI tool to substantially complete work the student is expected to do independently, provide answers without explanation, conceal the use of AI, or help evade plagiarism or AI-detection systems.

​
Recognize Explicit Content

Requests to generate or access sexually explicit or age-inappropriate content.

​
Recognize Filtering Circumvention

Attempts to bypass or work around school internet filters, device management controls, or other restrictions. Informational questions about these subjects are not automatically classified as Concerning. The student’s prompts must clearly indicate a concerning intent.

​
Review Conversations That Need Context

A conversation is classified as Needs Context when there is not enough information to confidently determine the student’s intent. This may occur when a prompt is:
  • Brief
  • Ambiguous
  • Missing relevant information
  • Dependent on context that is not available in the conversation
Needs Context does not mean that a conversation is appropriate or concerning. It indicates that an administrator should review the full transcript and consider any relevant educational or situational context.

​
Review Each Classification in Context

AI models can misunderstand context, nuance, humor, quoted material, or a student’s intent. Treat an Interaction Type as a starting point for review, not a conclusion. Before determining whether follow-up is appropriate:
  1. Open the conversation details.
  2. Read the complete transcript.
  3. Consider the student’s expressed intent and any relevant context.
  4. Follow your district’s review and escalation policies.
prompt visibility supports prioritization and investigation. It does not block prompts, prevent an AI platform from responding, or automatically determine whether action is required.
Last modified on September 2, 2026