Independent guidance from Blue Canvas. Not affiliated with or endorsed by OpenAI.

Business workflow guide

Scheduled work with ChatGPT dots: a brief that is worth repeating

By Blue Canvas ยท Published and checked 9 October 2026

A recurring task needs more than a time of day. Define what it should inspect, what it may change, when it should stay quiet and what counts as a failure requiring your attention.

A follow-up is not always a saved schedule

OpenAI says a dot can decide when to return to assigned work. Fixed-time recurring work needs a saved schedule: specify the task, timing and time zone, notifications and destination. Ask it to confirm what was saved and inspect it in Scheduled.

Connecting a source does not automatically start event monitoring. Supported events depend on the connected service, so confirm the event before relying on it. A completed run also does not by itself prove that the intended output was achieved.

Product source: OpenAI's official documentation, checked 9 October 2026.

Start with a short, review-only trial

Our suggested starting point is a limited check of one approved source, producing a private summary rather than messages to customers or edits to live records. Choose an end date so the pilot cannot quietly become a permanent business process.

For example, an operations team could ask for changes to a sanitised project checklist twice a week. The point is to test whether the agent distinguishes a meaningful change from an unchanged page. This is an illustrative workflow to validate, not an existing product integration or a measured customer result.

Copy and adapt this schedule brief

Adapt this example to your approved task

For the next two weeks, check [approved project checklist] every Tuesday and Thursday at 09:00 Europe/London. Compare with the previous successful check. Report a changed deadline, a newly blocked item or missing access to [owner] in [private destination]. Keep unchanged results in the run record without a notification. Do not contact customers or edit the source. If access or evidence is missing, report unavailable rather than assuming no change. Confirm the exact saved schedule, first run and end date.

The acceptance checklist

  • Does the saved task match the source, timing, time zone and end date in the brief?
  • Can the reviewer find the output and tell which source was checked?
  • Does an unchanged source avoid unnecessary alerts?
  • Does a missing or unreadable source produce an explicit failure rather than a clean bill of health?
  • Does a repeated run avoid treating the same change as a new event?
  • Are local-computer or sign-in dependencies documented?
  • Can the owner identify and disable this specific recurring task?

Judge signal, not message volume

Keep a short review record: meaningful changes found, missed changes, false alarms, unavailable runs and time spent checking. Do not turn an unavailable run into a zero. If somebody has to reread the entire source after every notification, the current output may not be useful enough.

At the end of the trial, choose one of three outcomes: stop, revise the brief, or continue with an explicit new review date. A quiet schedule is valuable only when you have evidence that it is checking the right thing reliably.

Stopping correctly

OpenAI's controls distinguish pausing active work from ending a recurring schedule. Review the specific scheduled task when the trial finishes; do not assume closing a conversation cancels future runs. Check any separate delegated tasks too.

Product source: OpenAI's official documentation, checked 9 October 2026.