ChatGPT Work Tasks: Catch the Inquiry You Might Miss

A customer asks about availability while you're serving someone else. By closing time, the message is buried. ChatGPT Work's event-triggered tasks offer a way to surface that inquiry and prepare the next step for review. OpenAI announced the update on August 25, 2026. See the release notes.

A useful first goal is modest: notice the right message, summarize the request, and prepare a reply you can check.

How is this different from a daily reminder?

A calendar-based task runs at a chosen time. An event-triggered task reacts to supported activity, such as an incoming Gmail message or a Slack channel message. Setup in Work includes a trigger, condition, and prompt. App access and plan eligibility matter; actions requiring approval can pause the task. OpenAI's scheduled-task guide explains the controls.

Think of the trigger as “when something happens,” the condition as “which cases count,” and the prompt as “what to do with those cases.” You can write all three without learning programming.

What does a sensible first workflow look like?

Use a test mailbox or practice messages in an account you're authorized to connect. Give those messages a distinctive subject, such as “Workshop practice inquiry.” That makes the test easy to separate from real customer correspondence.

Try this instruction:

When a new Gmail message arrives with “Workshop practice inquiry” in the subject, summarize the sender's request in three bullets. Identify any requested date, service, and unanswered question. Draft a reply for me to review here in Work. Do not send a message, promise availability, or quote a price. If a detail is missing, ask for it in the draft.

Before enabling the task, review the displayed trigger and condition. A beautifully written prompt can't fix a trigger that watches the wrong inbox.

How can you test it in one sitting?

Send three clearly labeled practice messages to the connected mailbox from an account you control:

1. A straightforward request with a date and service.
2. An incomplete request that says only, “Can you help next week?”
3. An unrelated message without the test subject.

The first should produce a useful summary. The second should preserve the uncertainty. The third should fall outside your condition.

Open the task record and inspect what actually happened. Check the notification settings too. An automation that runs correctly but never gets your attention may still miss your business goal.

Can you share the setup with a teammate?

Eligible tasks can be shared as instructions for a recipient's independent copy. Their account uses its own app access; changes to your original don't automatically update that copy. OpenAI’s task-sharing documentation covers the details.

For a team exercise, have each person adapt the subject filter and draft instructions to a fictional role. Then compare what their tasks catch. This exposes vague conditions before they affect customers.

When is the workflow ready for real inquiries?

After you can explain why it triggered, trust the summary enough to check it quickly, and know where to pause it. Start with a narrow category of inquiries and continue reviewing the drafts.

Bring a routine follow-up problem to a JOSA.AI public class in Lakeland or online and practice describing it as a ChatGPT Work task. For staff who share an inbox or divide customer responsibilities, custom team training can work through the handoffs together.