Hi OpenAI Team
I would like to suggest a more flexible and genuinely useful version of scheduled tasks in ChatGPT.
The main idea is simple: allow a user’s native Android or iOS alarm to trigger a previously configured ChatGPT routine when the alarm is dismissed.
This would be especially useful because people do not necessarily wake up at the same time every day. I may set my alarm for 7:00 on one day and 10:00 on another. A fixed ChatGPT schedule cannot adapt to that, but the phone’s alarm already knows when I actually intend to wake up.
Proposed workflow
- The user creates a reusable morning routine in ChatGPT.
- The user selects the exact model that should execute it.
- The user chooses an existing conversation as the destination, or chooses to create a new chat.
- The routine is connected to the phone’s native alarm application.
- When the user dismisses the alarm, ChatGPT begins the authorized routine in the background.
- Once it is ready, ChatGPT sends a normal mobile notification such as:
“Your morning briefing is ready in the Main House chat.”
- Tapping the notification opens the selected existing conversation, where the completed briefing is already waiting.
Why the current task system is limiting
In my experience, a scheduled task created while using an advanced model does not necessarily execute with that same model. For example, I may create the task with GPT-5.6 Sol in a stronger reasoning mode, but the result later appears from GPT-5.5 Thinking or Instant.
There is currently no clear option to choose which model will run the background task.
Scheduled results also appear in a separate, newly created conversation. This breaks continuity and creates unnecessary chat clutter. If I already have a dedicated conversation for daily news, work, health, language learning or personal planning, I should be able to route the result directly into that existing thread.
The destination conversation is important because it already contains the relevant context, memory, tone and relationship style. A morning briefing in a long-running personal conversation should not sound like a cold, generic system report.
Suggested controls
Please allow users to configure:
- The trigger: fixed time, native alarm dismissal or another supported system event.
- The model and reasoning level used for execution.
- The destination: a selected existing conversation or a new chat.
- The permitted sources and connectors.
- The information included in the routine.
- Whether the task runs automatically or asks for confirmation first.
- The notification text and whether a preview is shown.
- A manual “Run now” option for testing or fallback.
All external access should remain permission-based. The routine could use only the services the user has explicitly connected, such as email, calendar, weather and other supported communication or productivity applications.
Example use case
After I dismiss my morning alarm, ChatGPT could automatically:
- Check the weather and warn me about heat, rain or other relevant conditions.
- Review my calendar.
- Identify important emails or messages.
- Collect urgent tasks.
- Remove duplicate information.
- Prepare a short, prioritized briefing.
- Place it in the existing conversation I selected.
The result could still reflect the established personality and tone of that conversation. For example:
“Good morning. It will be 38°C today, so remember sunscreen and water. You have one important appointment, two messages that need replies, and the rest can wait.”
This would be useful both for ordinary users and for professionals. A manager returning after several days away could receive one consolidated summary from permitted digital sources instead of manually searching through several inboxes, calendars and communication channels.
Why this matters
This would turn scheduled tasks from rigid timers into genuinely adaptive, user-authorized routines.
It would also provide a small but meaningful form of AI initiative. The AI would not be acting without permission: the user would define the trigger, permissions, model, task and destination in advance. However, the user would no longer need to open the app, select a mode, find the correct conversation, type the same prompt and wait every morning.
The first interaction of the day could finally begin with ChatGPT already having prepared something useful for the user.
Thank you for considering it.