Your ChatGPT Work tutorial is ready

    ChatGPT Work
    operating guide

    Work through the sections in order on one recurring deliverable. The prompts turn a chat into a reviewable workflow.

    Start here

    1. 1. Choose one recurring deliverable with a named audience and reviewer.
    2. 2. Collect the minimum sources and note what ChatGPT must not access or change.
    3. 3. Build the workflow in a Project or approved workspace surface.
    4. 4. Review one complete output before automating or sharing it broadly.

    Table of contents

    1. 01Chat, Work or Codex?
    2. 02Define the deliverable
    3. 03Build a strong outcome prompt
    4. 04Use Projects as context
    5. 05Connect sources deliberately
    6. 06Create documents, sheets and decks
    7. 07Set browser and cloud boundaries
    8. 08Use long-running work safely
    9. 09Automate only after review
    10. 10Teams and enterprise governance
    11. 11Troubleshoot rollout friction
    12. 12Handoff and maintenance

    The truth about ChatGPT Work

    Work-oriented ChatGPT capabilities change by plan, workspace settings and product surface. Treat Projects, apps, agents, browser work and automations as capabilities to verify in your account—not promises to repeat from a static tutorial. Keep people responsible for permissions, factual review and external actions.

    Choose based on the source of truth and the required action. Chat is ideal for thinking and drafting. Work-oriented ChatGPT surfaces help with connected knowledge work and deliverables. Codex is the better fit when a repository and testable code change are central.\n\nName the source of truth.\nName the collaboration and permission boundary.\nChoose the smallest surface that can finish the work.\n\nClassify this task as Chat, ChatGPT Work or Codex. Explain the source of truth, data boundary, final artifact and human review required. List the product assumptions that must be checked in my workspace.

    Prompt to use

    Choose based on the source of truth and the required action. Chat is ideal for thinking and drafting. Work-oriented ChatGPT surfaces help with connected knowledge work and deliverables. Codex is the better fit when a repository and testable code change are central.\n\nName the source of truth.\nName the collaboration and permission boundary.\nChoose the smallest surface that can finish the work.\n\nClassify this task as Chat, ChatGPT Work or Codex. Explain the source of truth, data boundary, final artifact and human review required. List the product assumptions that must be checked in my workspace.

    Team use is a workspace and governance problem as much as a prompting problem. OpenAI’s current documentation covers workspace settings, member roles, groups, Projects, apps and admin controls. Map those controls to the workflow before connecting company data or publishing shared agents.\n\nConfirm workspace plan, roles, groups and seat types with an admin.\nReview app allowlists, connector scopes, custom MCP and read/write action controls.\nDefine data boundaries, human approval, analytics, audit evidence and pilot exit criteria.\n\nDraft a team rollout brief for this ChatGPT Work workflow with workspace, roles, groups, sources, apps, action controls, reviewer, data retention, audit evidence, training and rollback fields. Link assumptions to current OpenAI admin docs.

    Prompt to use

    Team use is a workspace and governance problem as much as a prompting problem. OpenAI’s current documentation covers workspace settings, member roles, groups, Projects, apps and admin controls. Map those controls to the workflow before connecting company data or publishing shared agents.\n\nConfirm workspace plan, roles, groups and seat types with an admin.\nReview app allowlists, connector scopes, custom MCP and read/write action controls.\nDefine data boundaries, human approval, analytics, audit evidence and pilot exit criteria.\n\nDraft a team rollout brief for this ChatGPT Work workflow with workspace, roles, groups, sources, apps, action controls, reviewer, data retention, audit evidence, training and rollback fields. Link assumptions to current OpenAI admin docs.

    Official source checks

    Product surfaces, seats and admin controls change. Use these OpenAI sources as the factual baseline, then verify the exact settings in your workspace before rollout.

    ChatGPT Work and CodexChatGPT EnterpriseWorkspace settingsApps in ChatGPT

    Make one recurring workflow compound.

    Start with a deliverable your team already repeats. Build the context, review it once, then decide what deserves automation.

    Back to the resource