CHATGPT DOT PLAYBOOK / PLANNING

Create a daily work brief with ChatGPT dot.

Ask your dot to review only selected work sources and return a short list of meetings, deadlines, replies and decisions. Keep every suggested message in draft until you approve it.

By Agent Taskbook◷ 10–20 min setupRisk: ModeratePublished Oct 8, 2026Reviewed Oct 8, 2026
01 / START HERE

When this responsibility makes sense

Use it when your day is spread across a calendar, a defined slice of email and a project list, and you repeatedly spend time deciding what needs attention first.

Have ready

  • The exact calendar and email account to connect.
  • A narrow email folder, label, sender set or time window.
  • The project list that controls deadlines and ownership.
  • Your time zone, delivery time and end date.

Stop & check

  • Do not connect a broader account than the responsibility needs.
  • Keep sending, invitation changes and shared-record edits behind approval.
  • Do not let a summary replace the original calendar, email or project record.
02 / GIVE THE DOT A RESPONSIBILITY

Copy the bounded brief

Replace bracketed details inside ChatGPT. Start with two weeks so you can review performance before making the responsibility indefinite.

RESPONSIBILITY BRIEF
Responsibility: Prepare one daily work brief on weekdays for the next two weeks.

Use only these approved sources: [calendar], [email label or folder], and [project list]. Do not search other accounts or channels.

At [time and time zone], give me:
1. Today's meetings and the preparation each one needs.
2. Deadlines or commitments due today or within two working days.
3. Replies I owe, with a link or reference to the original message.
4. Conflicts, missing owners, or information you could not verify.
5. The three decisions that most need my attention.

Draft suggested replies when useful, but do not send messages, accept invitations, edit shared records, move files, make purchases, or change accounts. Ask before sharing information with another person or channel.

Deliver the brief in this ChatGPT conversation. Notify me outside the scheduled brief only if a deadline due within one working day is newly at risk. Confirm the sources, schedule, duration, delivery location, and approval boundaries before starting.

03 / WORKFLOW

Set up the first run

  1. Connect only the calendar, email account and project source needed. Review each plugin's allowed actions.
  2. Paste the brief, fill in the source boundaries, time zone, delivery time and two-week end date.
  3. Ask the dot to confirm the saved schedule and restate which actions remain draft-only.
  4. For the first run, compare every meeting, deadline and owed reply with the original source.
  5. Correct omissions and false priorities before allowing the next scheduled run.
04 / YOU DECIDE

Keep these actions behind approval

  • Sending or posting any message, including a reminder drafted from the brief.
  • Accepting, declining, moving or creating a calendar event.
  • Editing shared project status, deadlines, owners or files.
  • Sharing information from a private email or conversation with a different audience.
05 / CLOSE THE LOOP

How to verify the brief

Use the original connected systems, not the dot's recap, as the source of truth.

If it misses something

  • Narrow the job to one source until the brief is consistently accurate.
  • Add an explicit sender, label, project or date-window rule rather than connecting more data.
  • Check whether the plugin connection expired or lacks permission to read the required source.
  • Pause the main work and separately inspect delegated tasks and Scheduled if you need all activity to stop.
06 / EVIDENCE & LIMITS

What we tested—and what we did not

Walkthrough record, October 8, 2026: In an authenticated Chrome session, we opened ChatGPT's Work mode and submitted a one-time brief made entirely from fictional calendar, email and project entries. We explicitly prohibited connected apps, account or computer data, external actions and recurring schedules.

The result correctly separated meetings, deadlines and replies; flagged an approval due before its related meeting; identified a missing owner; treated a possible shared evidence gap as uncertain; omitted a resolved item; and restated that it had not connected apps, taken actions or created a schedule.

Connected-source check: We then installed only the Google Calendar plugin, leaving Gmail, Contacts and Drive disabled. The install flow disclosed search/read access plus permission to create, edit and delete calendar events; it did not offer a read-only installation. We authorized that scope, but instructed Work to return only aggregate event counts and overlap counts for a fixed seven-day window, with no titles, people, locations, links or descriptions and no changes.

The response returned the requested privacy-minimized aggregate and stated that it had only read data. We did not audit the underlying API calls, so this is evidence of a successful read-path walkthrough—not proof that the granted write scope cannot be used or that every connected calendar will behave the same way.

Schedule check: We saved a five-run weekday schedule with an explicit time zone, end date, ChatGPT-only destination, Calendar-only source and no-change rule. Scheduled showed the task and next run. “Run now” completed and delivered an aggregate-only table. However, the Calendar source card exposed an event title while the task was running even though the final table omitted titles. Treat instructions as output guidance, not a guarantee that intermediate tool or source cards will hide metadata.

Limit: We have not yet observed a run triggered at its scheduled time or confirmed automatic cessation after the last date. Gmail retrieval, multi-source synthesis and external notification behavior also remain Guidance only. Reviewer: Agent Taskbook.

OpenAI documents recurring schedules, connected apps, background work, Activity review and explicit delivery instructions. That supports the untested parts of this workflow design, but does not establish that a particular account, plugin or scheduled brief will be complete or accurate.

Reviewed October 8, 2026. Product access, plugin behavior and interface labels can change. Read our evidence method and use the platform-neutral verification checklist.