ChatGPT Event-Triggered Tasks: How Gmail, Slack & GitHub Triggers Work

Until now, most people thought of ChatGPT tasks as something that happened at a particular time: every morning, every Friday, in two hours, or once a week. ChatGPT event-triggered tasks add a different workflow. The task can start because something changed in a connected app.

That difference matters. A fixed schedule asks ChatGPT to check at a time you chose. An event trigger can react to the thing you actually care about—for example, an important email arriving, a project message appearing in Slack, or new activity on a GitHub pull request.

This guide explains what the new triggers do, who can use them, how to set one up, when they are more useful than normal Scheduled Tasks, and how to avoid giving an automated workflow more access than it needs.

If ChatGPT Work itself is new to you, start with our beginner guide to ChatGPT Work. It explains the broader Work experience, permissions, projects, files, and human-review model.

What Are ChatGPT Event-Triggered Tasks?

A ChatGPT event-triggered task is a Scheduled Task in ChatGPT Work that begins when a supported event occurs in a connected app instead of waiting for a preset clock time.

OpenAI’s August 25, 2026 release notes describe these as webhook-triggered tasks. The first documented webhook sources are:

• Gmail: a new message arrives.
• Slack: a monitored channel receives a new message.
• GitHub: activity occurs on a pull request.

The important word is supported. This is not the same as being given a blank webhook URL where you can connect any service on the internet. OpenAI currently documents specific trigger types tied to supported connected apps.

That makes these workflows useful for work that should happen after new information arrives, not simply because the clock reached 9:00 a.m.

Who Can Use ChatGPT Event-Triggered Tasks?

OpenAI’s August 25 consumer release notes explicitly confirm webhook-triggered Scheduled Tasks for Plus and Pro users through ChatGPT Work on web, iOS, and Android. OpenAI’s separate ChatGPT Business release notes also confirm the feature for Business members.

The same consumer release says Free and Go users cannot create webhook-triggered tasks. Free users can create a limited number of ordinary Scheduled Tasks, but that should not be confused with Gmail, Slack, or GitHub event triggers.

Scheduled Tasks vs Event-Triggered Tasks

The easiest mistake is treating the two kinds of automation as interchangeable. They solve different problems.

A quick decision rule

If your sentence begins with “Every Monday…”, “At 8 a.m….”, or “Once a day…”, a normal Scheduled Task is probably the better fit.

If your sentence begins with “When I receive…”, “When someone posts…”, or “When this pull request changes…”, an event-triggered task may be the better fit.

How to Create an Event-Triggered Task in ChatGPT

The exact interface can change during rollout, but OpenAI’s current guidance establishes the core workflow: use ChatGPT Work, connect the relevant supported app, approve the requested access, and describe the event you want ChatGPT to respond to.

1. Open ChatGPT Work.
   Use ChatGPT on the web, iOS, or Android with an eligible plan and select the Work experience.
2. Connect the app you need.
   Connect Gmail, Slack, or GitHub through the supported ChatGPT app/plugin flow. Use the correct personal or work account.
3. Review the permissions.
   Check what ChatGPT is allowed to read and, where applicable, what actions it can perform. Do not approve broader access merely for convenience.
4. Describe the event precisely.
   Tell Work what must happen before the task should run. Avoid a vague trigger such as “watch my email.”
5. Describe the output.
   State exactly what ChatGPT should produce after the event: a summary, action list, draft, comparison, revision plan, or another clearly reviewable result.
6. Add your safety boundary.
   Tell ChatGPT what it must not send, edit, publish, delete, purchase, or change without your approval.
7. Test with a low-risk event.
   Confirm that the trigger, source, output, and approval behavior work as expected before depending on the automation for important work.

Gmail Trigger: Turn a New Email Into a Useful Next Step

The Gmail trigger is easy to understand because the event is familiar: a new email arrives. The useful part is deciding which emails deserve an automated response and what ChatGPT should do after one qualifies.

A weak automation would be:

That could create unnecessary runs and process messages that do not need AI involvement. A better workflow narrows the event and keeps the result reviewable.

When a new Gmail message arrives that is relevant to an active client project, review the message and prepare a concise triage note.

Include:
1. What the sender is asking for
2. Any deadline, date, or commitment mentioned
3. The next action required from me
4. Information that is missing
5. A draft reply if a response is clearly needed

Do not send the reply.
Do not change, delete, archive, forward, or label anything unless I explicitly approve it.
Flag uncertainty instead of guessing.

Where a Gmail event trigger is useful

• Customer-success teams triaging important inbound requests.
• Founders spotting messages that require a decision.
• Project managers extracting deadlines and requested actions.
• Sales teams preparing follow-up notes without automatically sending them.
• Operations teams turning recurring inbound notices into structured action lists.

For broader safety considerations around inbox access, read our AI Agent Safety Checklist before granting an AI agent access to work accounts.

Slack Trigger: React to a New Channel Message

Slack event triggers can be useful when important work happens inside a specific channel and you want ChatGPT to turn new activity into a structured follow-up.

OpenAI’s August 25 release notes specify an extra requirement: add @ChatGPT to each Slack channel you want the task to monitor.

That matters because “monitor Slack” is too broad. A better workflow names the channel, the kind of message worth reacting to, and the result ChatGPT should prepare.

When a new message in the approved project Slack channel contains a client decision, requested change, blocker, deadline, or explicit action item, create a short project update.

Return:
- What changed
- Who appears responsible
- Any deadline mentioned
- What is blocked
- The next recommended action
- Any question that still needs an answer

Ignore casual conversation and messages with no actionable project information.

Do not post a reply, message anyone, or modify another system without my approval.

Why filtering matters

Busy channels contain greetings, reactions, duplicate information, brainstorming, side discussions, and incomplete thoughts. Triggering meaningful work from every message would create noise instead of saving time.

The better target is a small class of high-value events: decisions, blockers, approvals, deadlines, customer feedback, or explicit requests.

GitHub Trigger: React to Pull-Request Activity

GitHub is the most technical of the first three documented event sources. OpenAI says ChatGPT Work can respond when a GitHub pull request changes, and its Learn documentation gives review feedback as an example.

A useful workflow is not “change my code automatically whenever someone comments.” A safer starting point is to summarize what changed and prepare a revision plan for human review.

When one of my pull requests in the approved repository receives new review feedback, summarize the new feedback and prepare a revision plan.

Include:
1. The reviewer concern
2. The file or area affected
3. Whether the feedback is blocking or optional when that is clear
4. A concise proposed fix
5. Tests or checks that should run after the change
6. Any ambiguity that requires human clarification

Do not merge, push, publish, or make consequential repository changes without the required approval.

This is a good example of where automation should reduce administrative effort while leaving the developer responsible for correctness.

Use the TAPS Framework Before You Automate

The fastest way to create a bad automation is to focus entirely on what ChatGPT can do and ignore what it should do.

Use this Designs24hr framework before creating any ChatGPT event-triggered task:

TAPS example

Best Real-World Uses for Event-Triggered Tasks

The best automation is usually repetitive, easy to evaluate, and triggered by a clearly recognizable event. It should remove busywork without quietly taking over decisions that still require judgment.

What event-triggered tasks are not good for

• Ambiguous decisions that require significant context.
• Workflows where a wrong action could cause financial, legal, employment, health, security, or reputational harm.
• Accounts containing unnecessary highly sensitive data.
• Tasks whose success is difficult for you to evaluate.
• Automation created only because automation is possible.

If the output becomes a list of actions you still need to organize manually, you can use the free AI Daily Task Planner to turn those follow-ups into a practical order for the day.

Privacy, Permissions and Human Approval

Event triggers make permission design more important because the workflow may run when you are not actively sitting in the chat.

That means the main safety question is not simply, “Is ChatGPT’s summary accurate?” You also need to ask:

• Which account is connected?
• Which messages, channels, repositories, or records can the task access?
• Can the connected app only be read, or can it also be changed?
• Can information be sent outside the original app?
• What happens when external content contains misleading instructions?
• Which actions require your approval?
• How will you disable the task or revoke access later?

Keep consequential actions behind review

OpenAI’s August 25 release notes say actions requiring approval pause until you review them. That is useful, but it should not replace your own workflow design.

For important tasks, explicitly tell ChatGPT not to perform consequential actions until you approve them.

SAFETY BOUNDARY

Use the connected app only for the task described above.

Do not:
- send or publish messages
- delete or move data
- change account or workspace settings
- purchase anything
- modify permissions
- make irreversible changes
- disclose private information to another destination

If an action could have a meaningful external consequence, stop and ask for my approval first.

If the source information is ambiguous, flag the uncertainty instead of guessing.

For a deeper permission review, use the 9-check AI Agent Safety Checklist. It covers least privilege, sensitive data, prompt-injection risks, approval gates, activity review, and revoking connections after the task ends.

Remember that incoming content is untrusted

An email, Slack message, or repository comment can contain text written by someone else. Automated systems should not blindly treat every instruction inside that external content as an instruction from you.

Your task prompt should separate source content from your actual instructions. For example, ask ChatGPT to summarize what a message says rather than obey commands contained inside the message.

Important Limits to Know

1. Only specific webhook sources are currently documented

OpenAI currently lists new Gmail messages, Slack channel messages, and GitHub pull-request activity. Do not assume that every ChatGPT-connected app automatically supports an event trigger.

2. This is not a universal webhook builder

The August 25 announcement uses the word webhooks, but that does not mean the consumer interface gives you an arbitrary webhook endpoint for every third-party service.

Use the supported trigger options shown in ChatGPT and documented by OpenAI rather than assuming generic webhook compatibility.

3. Slack requires channel setup

OpenAI says to add @ChatGPT to every Slack channel that the task should monitor.

4. Connected-app permissions still apply

A task cannot responsibly be treated as having unlimited access simply because the app is connected. Your account permissions, ChatGPT permissions, and workspace administrator controls can all affect what is available.

5. Rollout can differ from documentation

New ChatGPT features can roll out gradually. An eligible account may not always see the same controls at the same moment as another user.

What to Do If an Event Trigger Is Not Working

Check 1: Make sure you are using ChatGPT Work

The August 25 announcement specifically describes webhook-triggered Scheduled Tasks in ChatGPT Work. If you are using a normal Chat conversation, switch to Work before assuming the feature is unavailable.

Check 2: Confirm your plan

Plus, Pro, and Business are currently explicitly confirmed in OpenAI’s August 25 documentation. Free and Go cannot create webhook-triggered tasks.

Check 3: Confirm the app is connected

The relevant Gmail, Slack, or GitHub connection needs to be available and authorized for the account or workspace you are using.

Check 4: Review workspace controls

On organizational accounts, an administrator may control which apps or actions are available. A feature can exist in ChatGPT while still being disabled for your workspace.

Check 5: For Slack, add @ChatGPT to the channel

OpenAI explicitly requires @ChatGPT in each monitored Slack channel. If the task depends on a channel where it has not been added, fix that before rebuilding the entire workflow.

Check 6: Make the trigger specific

Instead of asking ChatGPT to “watch everything,” describe the supported event and desired result clearly. Narrow tasks are easier to create, test, and verify.

Check 7: Test with a simple event

Use a low-risk test email, Slack message, or pull-request event and confirm that the workflow fires and produces the expected output.

Check 8: Review paused approvals

If the workflow reaches an action that requires approval, it may stop until you review it. A paused task is not necessarily a broken task.

Can You Share an Event-Triggered Task?

Yes, OpenAI’s August 25 release notes also added sharing for Scheduled Tasks.

For personal plans, OpenAI says Free, Go, Plus, and Pro users can share Scheduled Tasks. The recipient gets an independent copy that they can review and customize. If the shared task uses a webhook trigger, the recipient still needs access to ChatGPT Work and must connect their own app.

OpenAI separately says ChatGPT Business members can share scheduled or webhook-triggered tasks with other members of the same workspace, with each person using their own connected-app permissions.

When Event Triggering Is Actually Worth Using

ChatGPT event-triggered tasks are most useful when four conditions are true:

• The event happens often enough that checking manually wastes time.
• The event can be identified clearly.
• The output follows a repeatable structure.
• A person can quickly verify whether the result is correct.

If those conditions are missing, a normal ChatGPT conversation or a time-based Scheduled Task may be simpler and safer.

The goal is not to automate the maximum number of actions. It is to automate the predictable handoff between new information and a useful next step.

Frequently Asked Questions

Final Takeaway

ChatGPT’s new event triggers change Scheduled Tasks from something that only waits for a clock into something that can react to supported work events.

The strongest use cases are not giant autonomous workflows. They are narrow, predictable handoffs: an important message arrives, ChatGPT extracts the action; project feedback appears, ChatGPT structures the next steps; pull-request feedback changes, ChatGPT prepares a revision plan.

Use the TAPS framework—Trigger, Access, Prompt, Safeguard—before turning on any automation. Define exactly what starts it, connect only what it needs, specify a reviewable output, and keep consequential decisions under human control.

For a broader explanation of the product around these workflows, read What Is ChatGPT Work?. For permission and security checks before connecting accounts, use the AI Agent Safety Checklist.

Official Sources