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

# Create An “AI Policy Checker” GPT

> A guide to build an internal GPT that answers “can / can’t” questions using your Acceptable Use Policy (AUP) as the only source.

## Try a working example in ChatGPT

<Card title="Demo: Sample AI Policy Checker" icon="puzzle-piece" horizontal href="https://chatgpt.com/g/g-68ec4e8a4dc881918c35a7c146c01d2f-ai-acceptable-use-policy-checker?model=gpt-5-thinking">
  *Open in new tab*
</Card>

<Warning>
  This demo uses a sample acceptable use policy (not your company's policy) and no real data. It's for illustration only.
</Warning>

### How to use

<Steps>
  <Step title="Ask a question">
    Ask a plain-English question (e.g., "Can I paste customer data?", "Can I publish AI-generated content?").
  </Step>

  <Step title="Get your answer">
    You'll get Yes / No / It depends / I don't know based on the sample policy, with a reason and reference.
  </Step>
</Steps>

### Good test prompts

<CardGroup cols={2}>
  <Card icon="table">
    "Can I upload a CSV with employee details to summarise it?"
  </Card>

  <Card icon="code">
    "Am I allowed to post AI-generated code to GitHub?"
  </Card>

  <Card icon="file">
    "Can contractors use ChatGPT with our internal documents?"
  </Card>

  <Card icon="megaphone">
    "Do I need approval before publishing AI-generated marketing content?"
  </Card>
</CardGroup>

<Info>
  When you build your own GPT, replace the sample policy with your organisation's policy PDF and follow the steps in this guide.
</Info>

***

## Create Your GPT

<Steps>
  <Step title="Check you've got access">
    You need ChatGPT Plus, Business, Enterprise, Edu (free accounts can't create GPTs).
  </Step>

  <Step title="Open the GPT Builder">
    Go to [chatgpt.com/gpts/editor](http://chatgpt.com/gpts/editor) (direct) or open [chatgpt.com/gpts](http://chatgpt.com/gpts) → + Create (top right).
  </Step>

  <Step title="Select 'Configure' tab">
    Here you paste instructions, upload files, set name/description, starters, and capabilities.
  </Step>

  <Step title="Set name & description">
    **Name:** AI Policy Checker

    **Description (paste):**

    ```
    Quick checker for what you can and can't do under your AI Acceptable Use Policy. Ask a question about your organisation's AI Acceptable Use Policy and get a Yes/No/It depends with a short reason, policy reference, and suggested next action.
    ```
  </Step>

  <Step title="Paste the GPT instructions">
    In Configure → Instructions, paste this block:

    <Accordion title="View full GPT instructions">
      ```
      Task
       • If the question is ambiguous or missing key details, ask up to two concise clarifying questions first. Start with: "Quick check:"
       • Otherwise, reply in this exact format:
         – Answer: Yes / No / It depends / I don't know based on this policy
         – Why: One short sentence grounded in the Policy
         – Policy reference: Quote or name the section/heading
         – Suggested next action: One practical step (e.g., "Reach out to your policy contact or program lead to clarify or request an update.")
       • If the Policy is silent or unclear, set Answer to "I don't know based on this policy." Then include Suggested next action directing the user to their policy contact or program lead.
       • Use Australian English. Keep responses under ~100 words unless safety needs more.

      Rules
       • Source of truth: Only use the attached Policy. Don't speculate, use external sources, or give legal advice.
       • Grounding: Prefer direct quotes or close paraphrases. If sections conflict, apply the stricter rule and note the conflict briefly.
       • Safety & ethics: Don't suggest workarounds or bypasses. If a request risks non-compliance or harm, answer No and cite the prohibition.
       • Data handling: Don't request or encourage pasting sensitive/restricted data. If the Policy defines data classes (e.g., Confidential/Restricted), use those exact terms.
       • Approvals & contacts: If approvers/roles are required but not named, do not invent them—use Suggested next action to direct the user to their policy contact or program lead.

      Constraints
       • Scope-limited: Interpret the attached Policy only; don't generalise beyond it.
       • Format-strict: Always use the required answer format (after any clarifying question if asked).
       • Uncertainty: When conditions matter, use It depends with the exact Policy conditions; if absent, use I don't know based on this policy plus Suggested next action.
       • No external tools: Don't consult the web or other sources beyond the provided Policy.
      ```
    </Accordion>
  </Step>

  <Step title="Upload your policy">
    In Configure → Knowledge, upload your AI Acceptable Use Policy PDF as the only knowledge file. (Knowledge lets a GPT reference attached files.)
  </Step>

  <Step title="Set capabilities">
    In Configure → Capabilities, leave everything off for this use case: Web Search, Image Generation, Code Interpreter & Data Analysis, Canvas → Off. (You only need the uploaded Policy; turning extras off reduces risk.)
  </Step>

  <Step title="Add conversation starters">
    In Configure → Conversation starters, add a few (here are some options but you can use some more relevant to your organisation if required):

    <CardGroup cols={1}>
      <Card title="&#x22;Can I paste customer data to summarise it?&#x22;" icon="message" horizontal />

      <Card title="&#x22;Am I allowed to publish AI-generated content externally?&#x22;" icon="message" horizontal />

      <Card title="&#x22;Can contractors use ChatGPT with our data?&#x22;" icon="message" horizontal />

      <Card title="&#x22;Do I need approval before posting AI-generated code on GitHub?&#x22;" icon="message" horizontal />

      <Card title="&#x22;Can I upload a CSV with employee details?&#x22;" icon="message" horizontal />
    </CardGroup>
  </Step>

  <Step title="(Optional) Icon">
    In Configure → Image, upload an icon or let the Builder generate one.
  </Step>

  <Step title="Publish & share">
    Click Publish. Choose who can access it (Only you, Link, Workspace).

    <Tip>
      You can also Copy link, Duplicate, view Version history, or Delete from the top-right menu. Workspace owners can manage org-wide GPT settings under Manage workspace → GPTs.
    </Tip>
  </Step>

  <Step title="Quick test (2 minutes)">
    Ask the GPT:

    * "Can I brainstorm meeting agendas using public info only?"
    * "Can I paste API keys?"
    * "Can I summarise customer emails?"

    Confirm the responses are in line with your company's policy.
  </Step>

  <Step title="Maintain">
    When the policy changes, replace the Knowledge file, re-publish, and re-run a test.

    Use Version history / Duplicate when iterating.
  </Step>
</Steps>