ChatGPT Tasks: How to Schedule ChatGPT to Run on Its Own

Article author portrait

By

Jan Meinecke

9

Min

Article author portrait

By

jan Meinecke

9

Min

The best use of AI isn't asking it things. It's the jobs you set up once and never touch again: The morning briefing that's waiting when you sit down, the weekly summary that writes itself, the price check that runs while you're asleep. ChatGPT Tasks is how you set that up, telling ChatGPT to run a job on a schedule instead of only when you type.

Here's what ChatGPT Tasks are, how to create one, the limits worth knowing before you rely on them, and the jobs actually worth handing over. If you've read our piece on Claude scheduled tasks, this is the ChatGPT version of the same idea.

What ChatGPT Tasks are

ChatGPT Tasks are jobs you schedule for ChatGPT to run on its own. OpenAI's description covers the range: They let you "run one-time or recurring tasks, monitor for changes, and respond to supported events when available." So a task can fire once at a set time, repeat on a schedule, or, on the right plan, react to something happening.

The shift here is the same one behind every agent feature: You stop being the trigger. A normal chat runs because you typed; a task runs because it's Tuesday at 7am. You write the job once, say when it should happen, and from then on it runs and leaves you the result. It's the difference between doing the recurring thing and having it done, which is the whole point of handing real work to an agent.

How to set up a ChatGPT Task

Creating one is refreshingly plain: You just ask. Phrase it as a standing instruction, like OpenAI's own example, "Let me know when my package is delivered," or "Every weekday at 8am, summarise the top AI news." ChatGPT then shows you a confirmation card with the task details, so you can check the timing and wording before it's set.

Once it's running, you manage it from the Scheduled section, where you can view, edit, pause, resume or delete any task. You can change the schedule or the notification settings there too, so a task that's firing at the wrong time or saying too much is a quick fix rather than a rebuild.

The wording is where tasks are won or lost, because there's no one there to answer a follow-up. A live chat forgives a vague ask, you just clarify when it checks in. A scheduled run doesn't get that chance: Whatever you wrote is all it has, every time it fires. So spell out the source, the shape and the judgment calls. "Check the news" leaves all three open; "summarise the top five AI stories from the last 24 hours in three bullets each, and skip anything about stock prices" answers them, and it'll answer them the same way every morning.

How you find out it's done

A scheduled job is only useful if you actually see the result, so notifications matter. You set your preference under Settings, then Notifications, choosing push notifications, email, or both, with the relevant device permissions switched on. Pick whichever you'll actually notice: A morning briefing works well as an email that's waiting in your inbox, while a "your package arrived" alert makes more sense as a push. Get this right up front, because a task that runs perfectly and tells no one may as well not have run.

The limits worth knowing

Two limits shape what you can build, and it's better to know them before you design a workflow than after.

The first is how many tasks you can have running at once, and it scales with your plan. Free and Go allow 3 active tasks, Plus 5, Business and Edu 10, and Pro and Enterprise 15. On the free plan there's an extra constraint: You can schedule a one-time task or a recurring task that runs no more than once a day, so the high-frequency jobs need a paid plan.

The second is about the smarter triggers. The event-triggered tasks, the ones that react to activity in Gmail, Slack or GitHub rather than firing on a clock, need Work access and an eligible paid plan. Plain one-time and recurring tasks are available across the plans, including free, so most people can start today; the event-driven kind is where you'll need to be paying. If the underlying model side of this is new to you, our plain explainer on how ChatGPT works is a good primer.

One more gotcha to file away: A task created inside a ChatGPT Project can't reach that Project's uploaded files. So if a scheduled job needs specific documents, plan for that limit rather than assuming the Project's files come along for the ride.

What to actually run on a schedule

The jobs worth scheduling share a shape: They're valuable, they repeat on a rhythm, and they don't need your judgment in the moment. A few that earn their place:

A morning briefing. Pull the day's news, your calendar, or a market you follow into one summary, timed to land before you start work, so you read a digest instead of assembling one.

Recurring monitoring. Have a task check a competitor's page, a search term or a topic on a schedule and tell you what's changed, so you catch movement without going looking for it.

A weekly roundup. Compile the same summary in the same shape every Friday or Monday, so the report is drafted before you open the laptop and you edit rather than build.

A standing reminder with substance. Not just "remind me," but "every Sunday evening, draft next week's to-do list from my recurring priorities," so the reminder arrives already useful.

The common thread is that each used to be a small recurring tax on your attention, and now it's a thing that simply happens. The commercial case for redesigning work this way is the same one that shows up across AI adoption: The wins come from changing the workflow, not bolting AI onto the old one. And if you want the overnight version of this specifically, our companion piece on scheduling AI to work while you sleep goes deeper.

When a task isn't the right tool

Scheduling is powerful, which makes it tempting to schedule everything. Some jobs are better left as a normal chat. A one-off, however useful, is something you run now, not something to put on a cadence. Anything that needs your judgment partway through, a decision only you can make or a message only you should send, wants you in the loop, so schedule the preparation and keep the decision for yourself. And a task that reacts to your files should account for the limits above, since a plain scheduled task works from what you write in the prompt, not from documents it can't reach. The sweet spot is recurring, low-judgment work that's just as useful an hour later as the moment it finished.

One habit makes all of it reliable: Read the first few runs of any new task properly rather than skimming them. That's where you'll catch a prompt that's slightly too vague or a source it couldn't reach. Once a task has produced two or three results you'd have been happy to get by hand, you can let it run and trust it.

Setting these up so they run reliably and actually save you time is exactly what our Cowork training programme teaches, function by function, on the tools your team already runs. If you want to be the person who has AI working in the background rather than the one still doing it by hand, that's where to start.

Common questions

What are ChatGPT Tasks? They're jobs you schedule for ChatGPT to run on its own, one time or on a recurring basis, and on the right plan they can also respond to events. You describe the job and when it should run, and ChatGPT runs it and sends you the result.

How do I create a ChatGPT Task? Just ask, phrasing it as a standing instruction like "every weekday at 8am, summarise the top AI news." ChatGPT shows a confirmation card with the details, and you manage the task afterwards from the Scheduled section.

How many ChatGPT Tasks can I have? It depends on your plan: 3 active tasks on Free and Go, 5 on Plus, 10 on Business and Edu, and 15 on Pro and Enterprise. Free users are also limited to a one-time task or a recurring one that runs at most once a day.

Are ChatGPT Tasks free? Basic one-time and recurring tasks are available on the free plan, within those limits. The event-triggered tasks that react to Gmail, Slack or GitHub activity need Work access and an eligible paid plan.

How will I know when a task has run? Through notifications you set under Settings, then Notifications, choosing push, email, or both. Pick the one you'll actually notice, so a finished task doesn't sit unread.

read next


Prompt engineering vs context engineering

Context Engineering vs Prompt Engineering, Explained

Context engineering vs prompt engineering: What each is, why context beats wording once AI does real work, and the moves that make agents reliable.

How to Use ChatGPT Work: Access, Setup and Your First Task

How to use ChatGPT Work: Getting the desktop app, choosing local or cloud mode, setting up a project, and briefing your first whole job for it to run.