OpenAI's Team Tasks can run recurring work for a ChatGPT Business or Enterprise team. This guide follows OpenAI's documented weekly project update example: prepare an update from selected project documents, run it once, inspect the source references, then adjust or pause future runs. Team Tasks are rolling out gradually, and the controls available to a person depend on workspace and team permissions. The steps below are based on OpenAI's documentation, not a BIG CHANGE test.
The big change
- What changed: Eligible Business and Enterprise teams can share ownership of recurring ChatGPT tasks that run in the cloud under a team service account.
- Why it matters: Teammates with permission can maintain future runs while workspace controls and connected account permissions still govern what a task can do.
- What to watch: Before keeping a recurring task active, check its connected account and source access, then inspect a manual run in Previous runs. Permissions and the saved instructions govern future work.
Check the owner and access before creating the task
Open the Teams list in the intended ChatGPT workspace and select the team that owns the project work. The team must already exist, and you must belong to it. The task will run in the cloud under the team's service account, using its saved instructions and team connections. It will not become a task in its creator's personal account. The creator can close ChatGPT without stopping a scheduled run.
Workspace permissions and team membership are separate checks. Enterprise administrators can use role-based controls for creating teams and creating or updating Team Tasks. A workspace permission does not add someone to a team or grant access to every task or run. Team members can view, run, edit, pause, resume, or delete tasks only where their permissions allow each action. The team owner alone can delete the team.
For the weekly update, identify the exact project documents and decide what the draft should contain. Ask the workspace administrator to make the required app connections available. The team owner then enables the eligible connections for this team. Inspect the provider account behind each connection: its permissions determine which documents the task can read and which destinations it can write to. If the update will be posted to an approved Slack channel, that account also needs permission to post there. App sign-in must be completed before an unattended run; a run cannot perform a new sign-in for you.
Plugins are a separate setup choice. Add a plugin if the task needs its tools or instructions, and review what it provides. Sharing a plugin does not share a personal app connection or grant the provider account more access. Team members do not each need to connect the same provider account when the team has an enabled workspace connection.
Create the weekly update
- In the owning team, open Scheduled and start a new task. Confirm the Team and enter a Task name.
- In the saved instructions, link the selected project documents. Ask for a weekly draft covering progress, blockers, and next steps, with a source link for each update. State the output format and what teammates should check before using it. Put essential guidance here: a shared Space does not automatically become task instructions, and Team Tasks do not inherit the creator's personal memory, Custom Instructions, or chat history.
- Choose Trigger → Schedule and set the weekly timing. Check the task's saved time zone. Date and time is for a one-time run; event triggers depend on supported app activity and the team's available connections.
- Review Plugins and Advanced, including Model, then select Create. Choose a model available to the team's service account; availability in your own account does not establish team access.
OpenAI documents the weekly project update as a draft from selected documents. Its example also describes posting an update to Slack, which requires an approved destination connection and posting permission. Check the intended destination separately if you configure that action.
Run once and inspect the result
Open the task details and select Run now. Then open Previous runs, select the run, and read its conversation and result. Check whether it used the selected documents, whether every reported update has the requested source link, and whether the draft contains the requested progress, blockers, and next steps. A teammate's ability to open the run does not automatically grant access to linked documents or files in the connected app; that app's sharing rules still apply.
If a permitted teammate needs to refine this result, they can reply in the run conversation. To change future runs, select Edit on the task, change its saved instructions or settings, save, and run it again to inspect the new result. If the task writes a file or sends a message, inspect that destination as well as Previous runs. A schedule marked Completed has no future time-based runs; the label does not prove that an individual run did the requested work.
Maintain the schedule and recover from errors
Use Pause to suspend future scheduled or event-triggered runs, Resume to enable them again, Edit to change instructions or settings, and Delete to remove the task. Check any run already in progress separately: pausing or deleting the task is not a reliable way to interrupt that run.
For a missing or failed result, first confirm the workspace and team, then check Previous runs for an error and compare the saved schedule, time zone, or event conditions with the expected run. For an app access error, inspect the team's connection and the connected provider account's permissions. For a model availability error, OpenAI says the task pauses; choose an available model under Advanced, save, select Resume, and check the next run.
Some temporary execution errors are retried automatically, but a failed app action does not necessarily receive an automatic retry. Before manually rerunning a task that posts a message or changes a file, check what the earlier run already did. If a usage or spending-limit error appears, record it and ask the workspace administrator to review the applicable workspace controls. Personal scheduled-task limits do not set the spending limit for shared Team Tasks. OpenAI's cited Team Tasks guide does not specify a task price, credit allowance, service guarantee, or a universal trigger list.
