ChatGPT scheduled tasks: the free plan gets them, with three limits

On January 14, 2025, announcing scheduled tasks, OpenAI wrote a sentence that went unfulfilled for nineteen months: "Eventually this will be available to anyone with a ChatGPT account." Eventually turned out to be August 25, 2026.

That day's release note has three parts. Two of them are for paying subscribers: tasks triggered by a Gmail, Slack or GitHub event, and task sharing by link. The third runs to three sentences and has barely been picked up anywhere: free accounts can create up to three active scheduled tasks.

This article covers what those three tasks actually let you do, how to create them, and why you can easily land on an OpenAI page telling you the exact opposite. Everything below comes from pages OpenAI has published, and every claim points to its source.

What changed on August 25, 2026

A scheduled task is an instruction ChatGPT carries out on its own, at a moment you choose, without you opening the app. "Summarise my industry news every morning." "Tell me when this page changes." "Remind me to follow up with this client on Friday." ChatGPT starts the conversation on your behalf and sends you the result as a notification.

The feature is not new, but its arrival on the free plan is. Here is the sequence, exactly as it appears in OpenAI's release notes — and nowhere else.

Here is the August 25 note in full, under the heading "Scheduled tasks for Free users":

Free users can create up to three active scheduled tasks. Tasks can run once or recur no more than once per day using flexible scheduling windows. Free and Go users cannot create webhook-triggered tasks.

Everything is there, restriction included. Three tasks, one daily run at most, and a window rather than a time. That last point is the most constraining one, and I come back to it below.

The June sunset of Pulse deserves a footnote: the proactive updates that product sent to Pro subscribers were absorbed into scheduled tasks. OpenAI's own advice to former Pulse users — ask ChatGPT to schedule a daily briefing — is exactly what the free plan has been able to do since August 25.

Why OpenAI's French pages say the opposite

Look this up in French and you will hit two OpenAI pages telling you the free plan does not get scheduled tasks. That is not a mistake on your side: it is the French localisation lagging behind. I checked both on August 27, 2026.

The French help page

The help page "Scheduled tasks in ChatGPT" exists in French and in English under the same article ID. Both display "Updated: yesterday". They do not say the same thing.

The French page is not wrong: it is the June revision. Every one of its claims matches the June 17, 2026 release note word for word, including the "Plus, Pro, Business and Enterprise" and the "no more than once per hour". It is simply two months behind, and its date banner does not say so.

The French pricing page

The second gap is even more visible. In the comparison table on the French pricing page, the "Scheduled tasks" row shows, in the Free column, a cross whose accessible text is unambiguous: "Plan: Free, Feature: Scheduled tasks, No". The Go, Plus and Pro columns read "Yes".

In other words: OpenAI's release note grants three tasks to the free plan, and OpenAI's pricing page, in French, shows "no" on that same row, on the same day.

I could not check what the English pricing page shows on that row: from France, chatgpt.com/pricing redirects to the French version. The gap I am describing is established between the two help pages, and between the release note and the French pricing page — not between the two versions of the pricing page.

Creating a scheduled task, step by step

There are two routes, and the second is faster.

Through the Scheduled page

The Scheduled page is reachable from the ChatGPT sidebar, on the web, on mobile and in the desktop app. You create a task there and choose when it should run. It is also where you will find every task later on: when it runs next, what it returned, and the controls to pause, resume, edit or delete it.

In the desktop app, if the page does not show up, OpenAI says to go through the menu at the top left. And if it stays missing, the English help page simply recommends using ChatGPT on the web: availability depends on the app and its version.

Just asking for it

The short route: write your request in an ordinary conversation, phrased as a deferred action. OpenAI's own example is "Let me know when my package is delivered." ChatGPT creates the task and shows a confirmation in the conversation.

This is the mode I would start with, because it forces you to write the instruction as a sentence rather than fill in a form — and a scheduled task is nothing more than a well-written instruction.

Turning on notifications, or none of it shows

The step people skip, and the one that makes the whole thing pointless: without permission, the task runs but you never hear about it. Go to Settings → Notifications and pick push, email, or both.

On a computer, the browser will ask for its own permission on top of that. If ChatGPT detects that it is missing, it shows a notice in the conversation with an Allow button. On mobile the path is different: create your first task from the mobile app and grant permission when prompted. Once granted, task notifications follow you across platforms.

Editing or pausing a task

From the conversation, select the task title: a side panel opens where you can edit, pause, manage or delete it. To see the full list without going through the sidebar, two shortcuts: Settings → Notifications → Manage tasks, or the three-dot menu of a conversation containing a task.

The real limits, plan by plan

Here is the documented state as of August 27, 2026, according to the English help page, which is the current revision.

Two honest notes about that table. The Go row carries two "not specified" entries: OpenAI groups Go with Free for the number of active tasks, but the sentence describing the daily frequency and the windows names Free only. The text then says hourly schedules and exact delivery times require an eligible paid plan, without stating whether Go counts as eligible. I am not filling that gap with a guess.

Second, tasks do not live in a world of their own: they consume your usual usage limits. The French help page puts it plainly — your ChatGPT plan's usage limits apply to tasks as well — and the English version rephrases it by saying model availability and usage limits depend on the task, the account plan and workspace settings. On a free plan, a daily task that triggers a web search therefore draws from the same pot as your conversations. What is unlimited in text stays unlimited; what was capped stays capped. I covered that split in what "unlimited" really covers on free ChatGPT.

What tasks cannot do, on any plan

Two exclusions, identical on both versions of the help page: voice chats and GPTs. A task cannot call you, and it cannot lean on a custom GPT.

And one confusion to clear up straight away: Codex automations are not scheduled tasks. They are targeted workflows that run inside Codex, with their own interface. The Scheduled page does not list them.

The constraint that actually matters: window, not time

On the free plan, you do not pick "8:30 am". You pick "morning". OpenAI calls them "flexible scheduling windows, such as morning, afternoon, or night".

That constraint rules out an entire use case: the timed reminder. A free task will not replace an alarm, will not ping you ten minutes before a meeting, and will not message you the moment a ticket sale opens. For anything that turns on a minute, stay with your phone's calendar.

It gets in the way of nothing that is about rhythm, though: a morning briefing, an evening round-up, a daily watch. Nobody insists on receiving their press summary at 7:12 sharp.

The monitoring task: the format that pays off on free

This is the one point I would want you to keep. With three tasks and one run a day, the temptation is to schedule three daily briefings and receive three notifications a day, two of which say nothing new. That is the fastest way to turn notifications off within a week.

Monitoring tasks answer exactly that problem. OpenAI describes them like this: ChatGPT checks periodically whether something has changed and notifies you only when there is something worth reporting. Two properties follow, and both are documented: previous runs are remembered — so the task knows what it has already told you — and a monitored task can stop by itself once its end condition is met.

In practice: a daily monitoring task only costs you a notification on the days when something happens. Out of a quota of three, it is the format with the best return. And a one-time task that finishes frees its slot — the help page says so explicitly for the case where you are at the cap: pause a task, delete one, or wait for one to finish.

Event-triggered tasks: what free will not get

This is the real August 25 news for paying subscribers, and it is worth understanding so you know what you are not missing.

A classic scheduled task answers to a clock. An event-triggered task answers to a fact: a new message arrives, a Slack channel moves, a pull request changes state. It runs inside ChatGPT Work rather than in an ordinary conversation.

The three supported triggers are documented:

• Gmail: a new message, with the option to filter by sender or subject.
• Slack: a new message in a channel. You have to add @ChatGPT to each monitored channel.
• GitHub: supported pull request activity in an authorized github.com repository.

Setting one up goes through Settings → Apps to connect the account, then Work, where you describe the event and the action you expect. ChatGPT then presents three items to review: the trigger, the condition and the prompt. The task then joins the Scheduled page like any other.

The throughput is capped, and the figure is precise: up to 30 runs per hour and 720 per day, across all your event-triggered tasks combined. Several events may be grouped into a single run. And any action requiring approval — sending a message, changing external data — pauses the task until you review it.

Who gets them: Plus, Pro, Business, Enterprise, Edu, plus eligible ChatGPT for Healthcare workspaces. Not Free or Go, not FedRAMP workspaces. On Enterprise, Edu and Healthcare workspaces an admin must enable the option first; in Healthcare it is off by default, is not covered by the Business Associate Agreement, and OpenAI explicitly asks that no protected health information pass through it.

Should you regret this on the free plan? Only mildly. Event triggering is a professional automation building block: it assumes a connected Gmail inbox, a team Slack or a code repository. For personal use, a well-phrased daily watch covers most of the need — and it is a decent preview of what AI becomes when it acts without being asked, a subject I dig into in the agentic AI guide.

Sharing a task, and what the link really carries

The second August 25 addition, and this one is open to the free plan: you can now share a task by link. The recipient opens the link, reads the instructions, adjusts them, and creates an independent copy in their own account. Nothing is pooled: not the results, not the original schedule.

The steps: Scheduled page, the task's three-dot menu, Share, then Copy link. Active and paused tasks can be shared, including event-triggered ones. Completed tasks can no longer be newly shared.

What matters is what the link carries, because the answer is subtler than "nothing personal".

The left-hand column hides a trap the documentation spells out: the title may appear in a link preview, and anyone who opens the link reads the instructions in full. OpenAI therefore explicitly asks that you put no health information, financial details, passwords or account numbers in the title or the instructions. That is worth reading before you share a task built around your personal data.

Three management details, finally. Editing the task does not refresh the link: you have to reopen the Share dialog and copy the link again. Deleting the link stops new openings but deletes neither the original task nor the copies other people already created. And if you ever close your account, delete your shared task links first: the documentation recommends it explicitly.

Three tasks to create today with the free quota

Three slots, one daily run each. Here is how I would split them. These are phrasings to copy, not measured results: the constraints they respect come from OpenAI's documentation, the quality of the answer will depend on your subject.

1. A watch that only speaks when something happens

The best use of a slot. The template phrasing:

Every morning, check whether [specific page or subject] has changed since your last check. Only tell me if there is something new, in three lines at most. If there is nothing, do not send a notification.

Two documented mechanisms work for you here: the memory of previous runs, which stops the task repeating itself, and conditional notification, which keeps your quiet days quiet.

2. A morning briefing, but bounded

This is the use OpenAI itself recommended to former Pulse users when it was sunset. It works, provided you bound it:

Every morning, give me a five-bullet round-up at most on [two or three named subjects]. One bullet per genuinely new piece of information. If a subject has nothing new, leave it out.

Without the bullet cap and the "genuinely new" clause, a daily briefing turns into an interchangeable digest you stop reading after ten days. The cap is what saves it.

3. A one-time task, which frees its slot

The third slot I would keep rotating. A one-time task — prepare a file the day before a meeting, follow up on a deadline, take stock of a project on a given date — finishes and makes its slot available again. It is the only one of the three that does not tie up the quota for long.

And if all three slots are taken, nothing is lost: pausing a task is enough to create another. A paused task keeps its instructions, but it still occupies a slot while it is active — it really is pausing or deleting that frees the counter.

The eight traps that make a task fail

All documented by OpenAI, and all enough to leave a task doing nothing without anything saying so.

1. Notifications are not allowed. The task runs, you never see it. On a computer, the browser permission comes on top of the ChatGPT one.
2. The task was created inside a project holding files. It will not reach them: the documentation is explicit, a task created in a project cannot access that project's files.
3. The associated chat was deleted. The task is automatically paused. Deleting the task, conversely, does not delete the chat.
4. The task was left unattended. Inactive tasks may pause automatically after a period of inactivity.
5. An action is waiting for your approval. As soon as a task wants to send a message or change external data, it pauses until you review it.
6. You are at the active-task cap. ChatGPT will then refuse to create a task and to resume a paused one.
7. You are expecting an exact time on a free plan. It does not exist: it is a window — morning, afternoon or night.
8. You received a "Weekly finances update" you never created. Not a bug: connecting a financial account to ChatGPT Finances may create that task automatically. Pause or delete it from the Scheduled page, without having to turn off all your ChatGPT notifications.

FAQ - Frequently asked questions

The questions that keep coming up since the August 25 note, with short, sourced answers.

What to take away

Three tasks is not much. It is still enough to change the relationship you have with an assistant: until now, ChatGPT waited for you to open it. With a scheduled task, it comes to you. On the free plan, that shift dates from August 25, 2026, and it fitted into three sentences of a release note.

The constraints are real and worth knowing before you get disappointed: one run a day, a window rather than a time, no event triggering, and a usage counter shared with the rest of your conversations. Those constraints rule out the timed reminder and professional automation. They get in the way of nothing about a daily watch, which happens to be the most useful format.

The real advice is one word: only. A task that speaks to you every day ends up switched off. A task that tells you only when something happens stays useful for months. With three slots, the trade-off makes itself.

That leaves the documentation gap, which will presumably close within a few weeks. Until then the rule is simple: on this subject, the English help page is current, the French one is not, and the French pricing page still shows "no" on a row where the release note says "yes". If you cannot see Scheduled yet, it is not you.

Want to follow this story? Subscribe to the EverydayAITech newsletter for weekly analysis in plain language.

---

About the author: Flavien Hue has been testing and analysing artificial intelligence tools since 2023. His mission: make AI accessible through practical, honest guides, without unnecessary jargon.