TL;DR
ChatGPT Tasks runs a saved prompt on a schedule — a 7am market digest, a Sunday weekly summary — and notifies you by push or email when it finishes. As of June 2026 it requires a paid plan (Plus, Pro, Team, Enterprise, or Edu — not Free), caps you at 10 active tasks, and runs whether or not you’re online. It’s a personal scheduler, not infrastructure: runs are best-effort and web search occasionally doesn’t fire. This guide covers the task shapes that hold up, the silent failures that ruin them, and when to drop down to the API on cron.
What ChatGPT Tasks actually is
Tasks lets you schedule a recurring prompt without leaving ChatGPT. You describe what to run and when (“every weekday at 7am, search for funding news in fintech”), and ChatGPT executes it on that cadence, then pushes you a notification or email with the result. It runs server-side, so you don’t need the app open.
It is not cron, and it is not Zapier. There are no retries, no run logs beyond the chat thread, and no guaranteed delivery. Think of it as a calendar for prompts. For anything that feeds a production system, you want the OpenAI API on a real scheduler instead (covered at the end).
The hard limits, as of June 2026:
| Property | Detail (June 2026) |
|---|---|
| Plans with access | Plus ($20), Pro ($100/$200), Team, Enterprise, Edu — not Free |
| Active task cap | 10 per user; can’t be raised on higher tiers |
| Paused tasks | Still count toward the 10 cap |
| Cadence | One-time, daily, weekly, monthly, yearly |
| Time zone | Uses your account time zone unless you state one in the prompt |
| Notification | Push or email on completion; runs while you’re offline |
| Platforms | Web, iOS, Android, macOS (Windows app on the roadmap) |
| Default model | GPT-5.5 (Instant) for Plus/Pro |
Free users can’t create, view, or run Tasks at all. If you’re on Free and want recurring prompts, your only path is the API plus your own scheduler.
Who this is for
People who already have a recurring “I should check X every morning” habit and want it to run on its own — founders, analysts, researchers, ops. If you’ve never run the prompt manually a few times, don’t schedule it yet. You don’t know what good output looks like, and you can’t audit drift you’ve never seen.
Good fits for Tasks
- A daily news or market digest filtered to a specific industry or set of companies.
- A weekly KPI check-in where the question is fixed and the source is a public URL.
- A monthly “what’s new in my field” sweep across a set of blogs or arXiv categories.
- Birthday and anniversary reminders with a context-aware draft message.
- Plain reminders (“remind me to review the Q3 draft Friday at 3pm”) — Tasks works as a reminder, not just a content generator.
Before you start
- Run the prompt manually 5 to 10 times first. You want to know the output before you stop reading it.
- Define “good enough.” A digest nobody reads is just an email-volume problem.
- Pick the time window deliberately. A markets-open scan has to run after the open; a news digest has to run after morning publication, not before.
- Decide your failure tolerance. If a missed run is fine, Tasks is great. If a missed run breaks something downstream, build it on the API with monitoring instead.
Step by step
-
Draft and dry-run. Write the prompt in a fresh chat and run it 3 to 5 times across different days. Look at the variance: does it return useful output every time, or only on certain days?
-
Convert it to a Task. Once it’s stable, ask ChatGPT to schedule it. Be explicit about time and zone:
Run this prompt every weekday at 7:00am US/Pacific: Search the web for major news on [industries] from the last 24 hours. Keep only items with at least one of: funding, regulatory, product launch. Output 5 bullets max, each with a source URL and a one-line "why it matters." If nothing meaningful happened, say "no notable news" and stop. Don't pad. -
Verify the schedule. Confirm the Task appears in your Tasks list with the right cadence and time zone. Tasks default to your account time zone, so DST changes and travel both quietly break “7am” assumptions unless you pinned the zone in the prompt.
-
Read every run for a week. Watch for: empty results (search didn’t fire), repeated content (the model is recycling), wrong tone (fix the prompt, not the schedule).
-
Tighten with explicit constraints. Based on the week, add hard rules:
Constraints: - Every bullet must cite a URL published within the last 24 hours. - No opinion pieces or newsletters. Primary sources only. - If web search returns nothing fresh, say so explicitly and stop. -
Give it a downstream landing place. Have the Task email itself to your inbox or pipe its output into a notes file you review weekly. Tasks without a landing place get ignored within two weeks.
A prompt template for scheduled digests
You are running a recurring scheduled task.
Schedule: [frequency, time, time zone].
Goal: [one-sentence deliverable].
Inputs: [URLs, topics, names to track].
Constraints:
- Every claim must cite a source URL.
- If sources return nothing fresh, say so. Don't fabricate filler.
- Output format: [bullets / table / short paragraph].
- Stop at the format limit. No extra commentary.
Failure mode: if web search is unavailable, output "search unavailable — manual run needed."That last line matters most. Silent failures are the single biggest thing that ruins a Tasks workflow, and the only signal you’ll get is an output that looks plausible but skipped the search.
Quality check
- Read the first 5 to 7 runs in full. After that, scan weekly and audit a random run monthly.
- Compare a fresh manual run against the scheduled output on the same day. They should match closely. Drift means the prompt is brittle.
- If web search is part of the Task, occasionally confirm it actually fired. The model sometimes answers from training data without searching, especially on lower-priority schedules — the “search unavailable” line above forces it to admit when it can’t.
- Track your “nothing notable” outputs. Too many means the filter is too tight; too few means it’s too loose.
Managing the 10-task cap
You get 10 active tasks, and that ceiling doesn’t move even on the $200 Pro tier. Two practical consequences:
- Paused tasks still count. Pausing a Task for vacation preserves its prompt and schedule but does not free a slot. If you’re at the cap and need room, delete rather than pause.
- Audit your slots quarterly. Keep your scheduled Tasks in a dedicated Project so the system instructions and schedule stay coupled, and maintain a
scheduled-tasks.mdnote listing each Task’s purpose, cadence, and expected output shape. The Tasks UI is thin; your notes are the audit trail. Retire any Task that nobody reads — background-noise email is worse than no Task at all.
When to use cron-on-API instead
Tasks is best-effort by design. Move to the OpenAI API on a real scheduler when you need any of:
- Guaranteed runs with retries — a missed run can’t break the downstream system.
- Logging beyond the chat thread — you need run history, error capture, alerting.
- More than 10 schedules, or schedules that hand off to other services.
- Output that feeds production, not just your inbox.
A typical setup: a cron job (or GitHub Actions, Cloud Scheduler, a Vercel cron) that calls the API, parses the response, writes it somewhere durable, and pages you on failure. More wiring, but you get the reliability Tasks deliberately doesn’t promise.
Common mistakes
- Scheduling a prompt you’ve never run manually. You don’t yet know if it works.
- Ignoring time zones. A Task that says “7am” with no zone fires at your account zone’s 7am, which breaks on DST and travel. State the zone in the prompt.
- Trusting that web search fires every time. It doesn’t — model load and silent fallbacks happen. Add the “if search unavailable, say so” line.
- Letting outputs pile up unread. Tasks become noise within two weeks without a review habit.
- Wrong cadence. Daily news digests are usually fine; daily competitive-research sweeps are usually too noisy and decay fast.
- Pausing to free a slot. Paused tasks still count toward your 10. Delete if you genuinely need the room.
- Using Tasks where a missed run breaks something. It’s best-effort. For guaranteed runs, use the API on cron with retries and monitoring.
FAQ
- Do I need a paid plan?: Yes. As of June 2026 Tasks runs on Plus, Pro, Team, Enterprise, and Edu. Free accounts can’t create, view, or run them. The only Free workaround is the OpenAI API with your own scheduler.
- How many tasks can I run at once?: 10 active tasks per user, and that cap doesn’t rise on higher tiers. Paused tasks still count, so delete (not pause) if you need a slot back.
- What if a Task fails silently?: Check the chat thread — failed runs sometimes leave a trace there. Build a “did it run?” check into your weekly review, and use the “search unavailable” failure line so a skipped search announces itself.
- Can a Task use web search or other tools?: Yes, but name the tool you want in the prompt. The scheduler doesn’t always pick the right one on its own.
- How do I pause a Task for vacation?: The Tasks UI lets you pause or disable per Task. Pause preserves the prompt and schedule, but remember it still occupies one of your 10 slots.
- Tasks vs ChatGPT agent — what’s the difference?: Tasks is a scheduler for a single prompt. ChatGPT agent (Plus/Pro/Business/Enterprise/Edu) is for multi-step browsing and actions on demand. Schedule a digest with Tasks; hand a multi-site research-and-action job to agent.