+170 XP

ChatGPT · Scheduled tasks and automations

ChatGPT can wake up on its own, run a prompt on a schedule, and message you the result. "Every Monday, summarize this week's industry news" becomes a recurring job instead of something you remember to type. The feature is called Scheduled Tasks (usually just "Tasks"). It turns ChatGPT from a thing you poke into a thing that reaches out to you.

This lesson covers how Tasks work, how to set one up well, where they break, and when to move to the API instead.

What a scheduled task actually is

A Task is a saved prompt plus a schedule plus a delivery mechanism. When the schedule fires, ChatGPT runs that prompt in a fresh conversation on OpenAI's servers, so you do not need the app open. It then notifies you with the output in the app and, if you've enabled it, by email or push.

Three details are easy to get wrong:

  • Tasks run without you in the loop. Nobody is there to answer a clarifying question mid-run, so the prompt has to be complete and unambiguous on its own.
  • Each run starts a clean conversation. It still has access to your account-level memory, your custom instructions, and any apps (formerly called Connectors) you've authorized.
  • You can only have so many active Tasks. The limit varies by plan and has changed over time, so check your account rather than assuming a number.

The official reference lives at help.openai.com. For the wider agent feature set, see openai.com.

Setting up the Monday news brief

Here is a real example. The goal: every Monday at 8:00 AM, get a tight summary of the past week's developments in your industry that you can scan in 60 seconds.

Step 1: Write the prompt as a standalone job

With no follow-up possible, the prompt carries all the structure. Open a normal chat and type something like this:

Every Monday at 8am, search the web for the most significant developments from the past 7 days in enterprise AI tooling. Return exactly 5 bullets, each one sentence, ordered by importance. After each bullet, add a source link in parentheses. End with one line: "Worth watching:" and a single forward-looking item. If nothing notable happened, say so plainly instead of padding.

What's doing the work:

  • A real trigger phrase ("Every Monday at 8am"). ChatGPT detects the scheduling intent and offers to create a Task.
  • An explicit tool instruction ("search the web"). Tasks can use web search, but only reliably when the prompt asks for it.
  • A fixed output contract: 5 bullets, one sentence each, sources, one closing line. Output you skim every week should look the same every week.
  • A failure instruction. Telling it what to do on a slow news week prevents invented filler.

Step 2: Confirm and refine the schedule

When ChatGPT proposes the Task, it shows the parsed schedule. Fix the time or recurrence if it guessed wrong. You can also create, edit, and pause Tasks from the Tasks view in your profile menu, which lists every active and paused Task.

Plain English works for cadence ("the first Monday of each month", "every weekday at 7am"). ChatGPT translates it; you verify it.

Step 3: Scope it inside a Project (optional but smart)

If the brief is for a specific client or domain, create the Task inside a Project so it picks up that Project's instructions and files. A Task created in a Project called "Acme competitive intel" can weight Acme's competitors without you repeating it in the prompt.

Reviewing and tuning the output

The first run is a draft. Treat the first two or three Mondays as calibration.

What to check:

  • Did it use the tool you expected? If the brief reads like training data rather than this week's news, web search didn't fire. Make the instruction more forceful and add a date anchor ("only items from the last 7 days").
  • Is the format stable? Drift in length or structure means your contract is too loose. "Exactly 5 bullets" beats "a few bullets".
  • Are the links real? Spot-check them. If sources are weak, tell it to prefer primary sources and skip any item it can't link.

To edit, open the Task and revise the prompt directly. There's no need to recreate it. Pausing is handy during a holiday; resume when you're back.

A clean tuning loop:

  1. Read Monday's output.
  2. Note the single biggest flaw.
  3. Edit the prompt to fix that one thing.
  4. Wait for the next run and repeat.

Don't rewrite the whole prompt at once. You won't know which change helped.

How to Use ChatGPT Scheduled Tasks

Watch on YouTube

Tasks vs. pulse

OpenAI also offers ChatGPT Pulse, which prepares a daily set of updates on its own based on your chats, memory, and connected apps. Pulse decides what to show you. A Task does exactly what you wrote, on exactly the schedule you set. If you want a specific brief with a fixed format, use a Task. If you want a loose morning digest you don't have to design, try Pulse. Availability depends on your plan.

Making Tasks more powerful: apps and the agent

A web-search brief is the easy case. Tasks get much more useful when they touch your own data.

Apps (formerly Connectors)

Apps link ChatGPT to outside services such as Google Drive, Gmail, GitHub, and Outlook. What's available depends on your plan and region. OpenAI renamed Connectors to apps in late 2025, so older guides and screenshots may still use the old name. A Task can read from a connected app at run time:

Every Friday at 4pm, check my Google Drive folder "Weekly reports", summarize any document modified this week into 3 bullets each, and flag anything mentioning "delay" or "blocked".

Now the Friday digest is grounded in your actual files, not just the public web. Connect the app once in settings and the Task uses it on every run.

The ChatGPT agent

For multi-step jobs, the ChatGPT agent can browse, click, fill forms, and operate a virtual computer to finish a task end to end. You can schedule agent runs too, which handles things a single prompt can't, like working through a site with no usable search.

The tradeoff: agent runs are slower, can hit login walls or CAPTCHAs, and may stop to ask for confirmation before sensitive actions. For an unattended job that needs zero human input, a focused prompt with web search is usually more reliable. Use the agent when the job needs interaction, not just retrieval.

Knowledge check

1. Why must a scheduled task's prompt be complete and unambiguous on its own?

2. What best describes what a scheduled task actually is?

3. Each task run happens in a fresh, isolated conversation. What does this imply for how you should write the prompt?

MULTIPLE CHOICE

4. Select ALL resources a scheduled task can still access even though each run is a fresh conversation.

Select all the correct answers.

MULTIPLE CHOICE

5. Select ALL statements that are true about how scheduled tasks operate.

Select all the correct answers.

When to leave ChatGPT and use the API instead

Scheduled Tasks are good for personal, human-readable automations delivered to you. They hit a ceiling once you need any of these:

  • Delivery somewhere other than ChatGPT (post to Slack, write to a database, trigger another system).
  • Strict, machine-readable output that downstream code can parse every time.
  • Volume: many runs or many users.
  • Version control and testing of the prompt itself.

At that point you move the schedule into your own infrastructure (a cron job, a serverless function, a workflow tool) and call the OpenAI API on each tick. You handle the scheduling; the API does the thinking.

Here's the Monday brief as an API job using the Responses API with the built-in web search tool and structured outputs, so the result is JSON with a guaranteed shape:

python
import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

schema = {
    "type": "object",
    "properties": {
        "items": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {
                    "headline": {"type": "string"},
                    "source_url": {"type": "string"},
                },
                "required": ["headline", "source_url"],
                "additionalProperties": False,
            },
        },
        "worth_watching": {"type": "string"},
    },
    "required": ["items", "worth_watching"],
    "additionalProperties": False,
}

response = client.responses.create(
    model="gpt-5",
    tools=[{"type": "web_search"}],
    input=(
        "Find the 5 most significant developments from the past 7 days "
        "in enterprise AI tooling, ordered by importance. Include a source "
        "URL for each. Add one forward-looking item under worth_watching."
    ),
    text={
        "format": {
            "type": "json_schema",
            "name": "weekly_brief",
            "schema": schema,
            "strict": True,
        }
    },
)

print(response.output_text)

Wrap that in your scheduler and send response.output_text to Slack or email. With strict: True, the output always matches your schema. Model names change often, so check the models page for the current recommended model, and see the OpenAI API docs for the full reference.

If the automation grows into several coordinated steps (fetch, summarize, fact-check, format, route), look at the Agents SDK. It's built for orchestrating tool-using agents in code, so you don't have to cram everything into one prompt.

A quick decision rule

  • Just for you, readable, low volume: a Scheduled Task.
  • Needs your private files: a Scheduled Task plus apps.
  • Needs to click through a website: a scheduled ChatGPT agent run.
  • Feeds other software, needs a strict format, or runs at scale: your own scheduler plus the API with structured outputs.
  • Multi-step, multi-tool pipeline: the Agents SDK.

Pick the lowest-effort option that meets the need. Most people jump to the API when a Task would have done the job in two minutes.

Key Takeaways

  • Write Tasks as complete, standalone prompts. Specify the tool, the exact output format, and what to do on a quiet week, because nobody is there to answer questions.
  • Calibrate over the first few runs and fix one flaw at a time. Check that web search fired and that the links are real.
  • Use apps (formerly Connectors) and Projects to ground recurring briefs in your own data. Pulse is the hands-off alternative when you don't need a fixed format.
  • Move to the API with structured outputs once you need machine-readable results, delivery outside ChatGPT, or scale. Use the Agents SDK for multi-step pipelines.

Frequently asked questions

What is a scheduled task in ChatGPT?

A ChatGPT Scheduled Task is a saved prompt attached to a schedule and a delivery method. When the schedule fires, ChatGPT runs the prompt in a fresh conversation on OpenAI's servers without the app being open, then notifies you with the result in the app and, if enabled, by email or push.

How do I create a recurring ChatGPT task, like a Monday morning news brief?

Type the request in a normal chat with an explicit trigger phrase such as "Every Monday at 8am, search the web for..." and ChatGPT detects the scheduling intent and offers to create the Task. It then shows the parsed schedule so you can correct the time or recurrence. All active and paused Tasks are listed in the Tasks view of your profile menu, where you can edit or pause them.

Why does my scheduled ChatGPT brief read like old training data instead of this week's news?

Web search probably didn't fire. Tasks can use web search, but only reliably when the prompt asks for it explicitly, so make the instruction forceful and add a date anchor such as "only items from the last 7 days". Spot-check the links too: if sources look weak, tell it to prefer primary sources and skip any item it can't link.

What's the difference between ChatGPT Tasks and ChatGPT Pulse?

A Task does exactly what you wrote, on exactly the schedule you set. Pulse prepares a daily set of updates on its own, based on your chats, memory and connected apps, and decides what to show you. Use a Task when you want a specific brief in a fixed format, Pulse when you want a loose digest you don't have to design. Availability depends on your plan.

Can a scheduled ChatGPT task read my Google Drive or Gmail?

Yes, through apps (renamed from Connectors in late 2025), which link ChatGPT to services like Google Drive, Gmail, GitHub and Outlook. You authorize the app once in settings and the Task reads from it at every run, for example summarizing documents modified this week in a given Drive folder. What's available depends on your plan and region.

When should I stop using ChatGPT Tasks and move the automation to the OpenAI API?

Move to the API when you need delivery outside ChatGPT (Slack, a database, another system), machine-readable output that downstream code can parse every time, volume across many runs or users, or version control and testing of the prompt. You then run the schedule in your own infrastructure, a cron job or serverless function, and call the Responses API with structured outputs on each tick. For multi-step pipelines that fetch, summarize, check and route, look at the Agents SDK.

What to do, from this lesson

These actions are compiled in the role's Playbook.

  • Prompt agents for a deliverable with exact columns and flag-unknowns rules
  • Write scheduled Tasks as complete standalone prompts specifying quiet-week behavior
See the full action playbook →