Scheduled tasks in ChatGPT dots

Scheduled tasks are what make a dot feel always-on. Instead of opening ChatGPT to ask for something, you set it up once and your dot comes back with the result. This page covers how OpenAI documents creating and managing them, then offers a template for writing tasks your dot can finish without stopping to ask.

What a scheduled task is

OpenAI’s help center describes creating a task by asking for it in conversation, the same way you’d ask for anything else. It doesn’t mention a separate form.

Where to manage your tasks

OpenAI documents two places to see scheduled work.

Where	What you see	What you can do
Your dot’s profile → Scheduled	Tasks grouped as active, paused and completed	Open a task to change its repeat schedule, time, or completion notifications
Your dot’s profile on desktop (activity)	Work listed under In progress, Scheduled and Completed	See what your dot is doing now and what it has finished
Your conversation with your dot	Each run and any proactive updates	Reply, adjust, or ask for changes in plain language

OpenAI’s safety FAQ adds that Activity View on desktop shows ongoing and delegated tasks and the steps taken, which is the best place to check what a run actually did.

Pausing and deleting

There are three levels of stopping, and they are not the same.

• One task. The Scheduled list shows paused tasks alongside active ones. OpenAI doesn’t spell out the exact control for pausing a single task, so open the task and look there, or ask your dot in conversation.
• The whole dot. Open your dot’s profile, choose •••, then Pause. To resume, use “Paused • Tap to resume.”
• Everything, permanently. “Reset deletes your dot, including its conversations, saved memories, and scheduled tasks.” See memory and reset before you do this.

Background work follows the same limits

A task that runs while you’re away doesn’t get extra permissions. Two documented limits matter here:

• Proactive research is read-only. OpenAI says it “can read information from permitted connected sources and save private notes,” but its research tools cannot directly send messages to other people, change content through plugins, or control a browser or computer.
• Custom rules can’t lift built-in safeguards. Rules cannot change Auto-review or the proactive research restrictions.

How to write a scheduled task your dot can run without asking

OpenAI defines pre-approval narrowly: “‘Pre-approved’ means you explicitly requested the action in your prompt.” So the safest task is one that names every action up front: who receives anything, what it says, and when or under what conditions it goes out. A task written that way gives your dot everything it needs in one place.

When:   [schedule, time and time zone, and an end date if any]
What:   [the work, and which connected apps or sources to use]
Report: [where the result goes: this conversation, or a named channel]
Send:   [who receives anything, exactly what, and under what conditions]
Done:   [what counts as finished, and what to do if something is missing]

A few habits help:

• Name the send explicitly. “Post the summary to the team channel” is a request. “Keep the team updated” is not.
• Keep one task to one job. Separate tasks are easier to pause, edit and read in the Scheduled list.
• Say what “nothing to report” looks like. Otherwise you may get an update every run whether or not anything changed.
• Remember the approval doesn’t spread. “Approving one message does not give your dot ongoing permission to contact people on your behalf.”

Example task prompts

1. Morning calendar review. “Every weekday at 7:30 a.m. my time, review my calendar for today. Report here: the first meeting, any overlaps, and any meeting without an agenda. If there’s nothing unusual, just say ‘clear day.’”
2. One-off reminder. “Remind me tomorrow at 4 p.m. to send the signed contract back to Dana. Just the reminder, don’t send anything.”
3. Weekly research check. “Every Monday morning, look for new public information about our competitor’s pricing page and summarize what changed since last week. Report here only. Don’t contact anyone.”
4. Recurring team update. “Every Friday at 3 p.m., draft a five-bullet summary of this week’s project notes and post it to the project channel in Slack. Send it without asking me. If there are no notes, post nothing and tell me here.”
5. Inbox follow-up check. “Every afternoon at 5, check my email for messages from clients that I haven’t replied to in two days. List them here with a one-line summary. Don’t reply to anyone.”
6. Month-end nudge. “On the last weekday of each month, remind me to submit expenses and list any receipts you can find in my connected email from that month.”

Example 4 only runs hands-free if your custom rule for posting to that channel allows pre-approved actions. The others only read and report.

Usage and cost

OpenAI’s announcement says conversations with your dot don’t count toward your ChatGPT usage limits. When your dot starts or manages tasks in Codex or ChatGPT Work, “those tasks count toward your usage limits as usual.” The help center adds that for the next month, dots usage won’t count toward eligible Pro, Business, and Enterprise users’ plan allowances. See pricing for what comes after that.