Phase 1 · ChatGPT · Level 3 · Power User
Tasks: scheduled and recurring runs
By the end, you'll be able to…
- Set up a one-off and a recurring scheduled Task in ChatGPT
- Write a Task instruction that produces something useful on a schedule, not noise
- Manage, review and pause Tasks so they don't quietly drift or pile up
Why it matters
Most ChatGPT use is reactive: you ask, it answers. Tasks flip that round: ChatGPT does a job on a schedule and brings you the result. A Monday-morning summary, a weekly nudge with everything you need to prep, a recurring draft that's waiting when you sit down. Done well, it turns a chore you keep forgetting into something that just arrives. Done badly, it's another notification you learn to ignore, so this lesson is as much about what's worth scheduling as how.
What a scheduled Task is
A scheduled Task is an instruction you give ChatGPT to run later, either once or on a repeating schedule, without you being there to prompt it. When it runs, ChatGPT carries out the instruction and notifies you with the result. You manage them all from a Scheduled area in the sidebar, which is the hub for creating, editing, pausing and reviewing your Tasks.
The mental shift is from pull to push. Normally you pull ChatGPT into a job when you remember to. A Task pushes the result to you on a rhythm you set: every weekday at 8am, every Monday, on the first of the month. If you've ever thought "I really should do X every week and I keep forgetting", that's a candidate for a Task.
A few practical limits worth knowing up front, because they shape what's sensible to schedule. Tasks run at most about once an hour; this is for daily and weekly rhythms, not second-by-second monitoring. And Tasks that run and run without you ever opening the results may be paused automatically, so a Task you actually read stays alive; one you ignore eventually stops. Scheduled Tasks are a paid-plan feature.
Our running example: Fernway's operations team has a weekly sync every Tuesday, and prep is always a scramble. We'll build a recurring Task that, every Monday afternoon, drafts a prep note for the sync, so it's waiting when the team sits down.
Setting up a Task
Go to the Scheduled area in the sidebar and create a new Task. You'll give it three things: an instruction (what to do), a schedule (when), and a name so you can find it later. You can also create a Task straight from a chat: if you've just worked out a good prompt, you can ask ChatGPT to turn it into a recurring Task there and then.
The instruction is where the quality lives. A vague Task produces vague output on a schedule, which is worse than nothing because it trains you to ignore it. Everything you learned about good prompts applies (role, task, context, format), plus one Task-specific rule: say what a useful result looks like, because you won't be there to iterate when it runs.
Every Monday at 3pm, draft a prep note for Fernway's Tuesday operations sync. Structure it as: (1) three or four things worth raising this week based on what I've told you in this Task's chat, (2) any decisions still open from last week, and (3) two short questions to push the meeting forward. Keep it under 200 words, UK English, plain and skimmable. If you don't have enough recent context to fill a section, say "nothing new to flag here" rather than padding it.
Why this works: It's specific about output (a fixed three-part shape), about audience (the ops team), and about honesty ('say so' rather than invent). Because a Task runs unattended, baking the format and the guardrail into the instruction is the only chance you get. There's no follow-up nudge at 3pm on Monday.
When you save this, ChatGPT confirms the schedule in plain language ("This will run every Monday at 3:00pm"). Check that confirmation, because a Task set to the wrong time or the wrong day is a common and quietly annoying mistake.
One-off Tasks and reminders
Not every Task repeats. A one-off Task is a scheduled reminder-with-work-attached: "at 9am on Thursday, before the Harlow renewal call, draft me three talking points and the two numbers I'll need." It runs once, delivers, and it's done. This is the gentlest way into Tasks: a reminder that arrives already useful rather than just a nudge saying "you have a call".
This Thursday at 9am, remind me about the Harlow account renewal call and draft three short talking points for keeping the account, plus a note of the two figures I said I'd bring (current spend and the proposed discount). One screen, no preamble.
Why this works: A reminder that also does the prep is worth far more than a bare alert. Naming the exact moment (the morning of the call) and the exact deliverable (talking points plus the figures) means the result lands when it's useful and needs no further work.
Manage them so they don't drift
The failure mode with Tasks isn't setup; it's neglect. Left alone, a recurring Task can outlive its usefulness: the project ends, but the weekly summary keeps arriving; your priorities shift, but the Task still reports on the old ones. So build in a light review habit.
From the Scheduled area you can see every Task, edit its instruction or timing, pause one you don't need right now, and delete the ones whose job is done. Two habits keep it healthy: pause rather than ignore: if a Task's output stops being useful, pause or edit it rather than mentally filtering it out (an ignored Task may get auto-paused anyway, but do it deliberately); and review monthly: glance down the list and ask of each one, "is this still earning its place?" A short, curated set of Tasks you actually read beats a long list you've learned to skim past.
Update this weekly Task: the product launch it's been tracking is finished, so drop that section. Instead, focus the prep note on the office-move decision and the August holiday rota. Keep the same Monday 3pm schedule and the same under-200-words format.
Why this works: Editing the instruction of an existing Task is better than deleting and rebuilding, since you keep its history and context. Telling it precisely what's changed ('drop the launch, add the office move') retargets it without starting over.
Try it now
Common mistakes
- Scheduling noise. A vague instruction produces vague output on a timer, and you quickly learn to ignore it. Be as specific about the format and the "what good looks like" as you would in a normal prompt, more so, because you can't iterate at run time.
- Wrong time or day, unnoticed. It's easy to set 3am instead of 3pm, or the wrong weekday. Read ChatGPT's plain-language confirmation of the schedule when you save.
- Set and forget. Recurring Tasks outlive their usefulness silently. Review the list monthly; pause or edit rather than mentally filtering out the ones that no longer help.
- Trusting a scheduled output because it arrived on time and looks finished (over-trust). A Task that runs reliably every Monday feels dependable, and that reliability can lull you into skimming its content. But the run is only as good as its inputs: a prep note can confidently flag a decision as "still open" that was actually closed last week, or summarise context it half-remembers. Read a scheduled output as critically as any other ChatGPT answer; punctuality isn't accuracy.
Keeping current
Scheduled Tasks are a fast-moving feature: the scheduling options, limits, the Scheduled hub and how notifications work are all still evolving (Tasks recently absorbed the earlier proactive-briefing feature). For the current details, see OpenAI's Tasks in ChatGPT help article and the ChatGPT release notes. Accurate as of 13 July 2026.