GPT DotFIELD GUIDE
Home/Workflows
Workflows

How to set up ChatGPT Dots workflows: copyable briefs for morning briefs, meetings, research watches, content and projects, and how to check a saved schedule (OpenAI Dots)

Set up a ChatGPT Dots (OpenAI Dots) workflow the way real users did: one job, one test run, a saved schedule with a time zone and end date, then a check under Scheduled. Copyable briefs for morning briefs, meetings, research watches, content and project follow-up.

GPT Dot · Updated
OpenAI Dots product demonstration
Image source

The short version: which Dot workflow to start with, how to set it up, and where saved schedules go wrong

In one line

A Dot workflow is worth setting up when you can name one job, the sources it should read and where the result should land, and it works best for jobs that read, compare and draft. If the job is one fixed prompt on a fixed clock, a plain ChatGPT scheduled task does nearly the same thing: BB gave the same 6:00 JST "five AI news items plus an X post" job to both and noticed little difference, and preferred the older task's wording, though the two sets of instructions weren't identical. A Dot earns its place when the job is a role with changes in between runs. たまじろー at SAMURAI ENGINEER handed his Dot two sales responsibilities over a deals sheet and a customer-feedback sheet; it saved its own weekday check at about 9:00, then reported an edit to the sheet before 9 without being asked, and asked "may I apply these three wording changes?" before touching the source files. The weak spots are the same in every workflow below: the saved schedule isn't always the one you asked for, a run marked completed doesn't prove the result reached you, and scheduled sends and writes are where runs get blocked. Start with one read-only job delivered to the Dot chat, and add more once a few runs have checked out.

Set it up in this order

  1. Pick one outcome from the list below and connect only the apps that brief needs. Each app is its own connection: reaching your Dot from Slack or Teams doesn't open your Gmail, Drive or Calendar (Connect computers and apps; Integrations). If you don't have a Dot yet, start with getting started.
  2. Describe the job as a responsibility with limits: what to keep track of, which sources to read, what to report, and what it must never do without asking. たまじろー's brief named two responsibilities, asked for the reason, the risk and a next step for every flagged deal, and listed four things that need his OK: customer email, data changes, publishing and deal-stage changes.
  3. Run it once by hand and open every link in the result. The Rundown's guide goes further and starts with a fictional practice brief that has an answer key, then one live run on approved sources, before anything is scheduled (ChatGPT Dot setup & practice guide).
  4. Set custom rules so sending, posting and calendar or file changes are on Ask before taking action or Hand off to you. Asking your Dot for a draft doesn't give it permission to send it (Control your dot).
  5. Ask for the schedule in the text chat, with the four parts OpenAI lists for recurring work: what to check, when to run (time zone and any end date), which changes deserve a notification, and where to deliver. Then ask the Dot to confirm what it saved (Tasks and memory). Set schedules in text rather than on a call: nkpryor gave detailed instructions for recurring calendar changes during a call, and the text chat later couldn't find the request.
  6. Open the saved task yourself. It sits in the Scheduled section of ChatGPT and under Recent activity in your Dot's profile; check its instructions, time zone, end date and destination (Getting started with your dot; Control your dot).
  7. Watch the first three runs. Note when each result arrives and compare it with the task's last run. OpenAI's guidance is to review the output even when a run is marked completed (Tasks and memory).

Common blockers and fixes

  1. The saved schedule isn't the one you asked for. Ramanpal's link-building task named a deadline in its text while the schedule said End repeat: Never (Promptslove), and Loïc Frissard's 8:00 watch didn't show in the Scheduled tab even though the Dot said it was saved; it was listed in the Dot's profile (Digitiz). Fix: open the task and check the end date and time zone, and look in the Dot's profile before setting the same job up again.
  2. You end up with two copies, or the Dot can't edit its own task. markifur's Dot couldn't see every scheduled task while a ChatGPT Work chat could (OpenAI Community), and OKJOIZJ's Dot got "You cannot modify this task." while the same edit worked in the task editor on the ChatGPT website eleven minutes later (openai/codex #51285). Fix: ask for a list of every scheduled task before creating one, and edit the existing task under Scheduled yourself.
  3. It arrives late or misses a source. uehara's 8:30 brief arrived at 8:44, and a request for "events" without the word "all" read only his main calendar (uehara on DEV). Fix: schedule a little before you need the result, name every calendar or source, and ask each report to start with when the run began, what it read and what it couldn't open.
  4. The task says completed, but nothing reached you. Future Stack Reviews couldn't tell whether a "No change" message came from its saved test task (Future Stack Reviews). Fix: a one-line heartbeat on every run, and a "last run" line so a missed run shows up the next time.
  5. It stops at suggestions. たまじろー's Dot first said the sales deck should be improved and went no further until he wrote "don't stop at suggestions; draft concrete improvements to the sales deck, FAQ and landing-page copy". Fix: say what you want back, a list of suggestions, a new draft or an edited file, and where it should go.
  6. A scheduled send or write is blocked. In early October, scheduled ChatGPT Work tasks that sent Gmail or wrote to Google Sheets ran on time and then had the write stopped by safety checks, in gaelto's case with no approval prompt (OpenAI Community). Fix: keep scheduled runs to reading and drafting, and ask the Dot to report any blocked send with the exact error.
  7. Alerts repeat old news. Jayprakash Suthar's six-hourly watch sent 2 repeats out of 8 notifications (AI Vidya Setu). Fix: say "don't repeat anything already reported" and give the condition for a message.
  8. Instructions given on a call get lost. Besides nkpryor's lost request, kevinkaiser found that the Dot could no longer read call transcripts at all and stopped using calls (OpenAI Community). Fix: set up and change recurring work in the text chat, and ask for the confirmation there. More on calls is on the voice calls page.
  9. A sign-in expires mid-series. 超级峰's evening posting stopped because the cloud computer was missing a Feishu credential (超级峰 on X), while Aruru Seisakusho signs in to each service themselves before the Dot checks it can reach every posting screen (Aruru Seisakusho on note). Fix: do the sign-ins yourself at setup, and have the Dot stop and hand over at a login or captcha instead of skipping the run.

What it can do, and what it can't yet

A Dot can run work at fixed times with a time zone and an end date, respond to a supported event when you ask it to watch for one, and decide on its own when to pause and wake up between runs, which is how たまじろー's Dot caught the sheet edit. It can list, change or cancel its scheduled tasks and read and edit your ChatGPT Work automations (Tasks and memory; what's new, October 5–9). Jobs that don't need your computer keep running on its cloud computer while yours is off.

It can't yet promise an exact minute, and its own between-run checks have no fixed time, so use a saved schedule when timing matters. Connecting Slack or another source doesn't create a watch by itself. Proactive research doesn't send messages or change your apps, so acting on a finding still goes through your approvals. Pause stops the Dot's current main task but doesn't cancel future scheduled runs; disable or delete the task under Scheduled to end it (Control your dot).

One message to copy

You're in charge of [job] until [end date]. Use these sources: [apps, pages, channels], and tell me if you can't open one. Do one run now so I can check the result. Then save it as a recurring task every [days] at [time] [IANA time zone] until [end date], delivered to [this chat or a Slack channel]. Start every report with one line: when the run began, which sources you read and which you couldn't. Message me only when [condition], and don't repeat anything already reported. Only read, summarize and draft; anything that sends, posts, deletes or changes a calendar or file waits for my OK. Before saving, check whether a task like this already exists. After saving, reply with the task name, days, time, time zone, end date, where results go and the next run time.

Pick an outcome

Choose the result you want back, then open the matching brief. Each page has its own short version, a copyable brief and real setups with sources.

  • Morning brief use case: from five decisions a day to overnight work reports, with copyable asks.
  • Morning brief workflow: setup plus real daily briefs built from calendar, email, news, weather and Slack.
  • Meetings: a brief before every call, scheduling rules, calendar-change alerts and follow-ups.
  • Project follow-up: a Dot as project manager, with trackers, stale-item checks and meeting action items.
  • Research watch: competitor prices and sites, search rankings, flight and hotel prices, news and release alerts, job listings, and your own systems and inbox.
  • Content pipeline: a webinar cut into a reel, blog drafts written overnight, scheduled posts under written rules, all for your approval.

Three setup references to start from

  1. OpenAI's own recurring task. The example in Tasks and memory asks the Dot to check a connected planning channel each weekday at 9 AM Pacific for four weeks, update a checklist when a deadline changes, message only when a deadline is at risk or a decision is needed, and confirm the schedule. Stopping active work and cancelling a schedule are separate actions.
  2. A content package with approval. Among the launch scenarios in Introducing dots, a creator's Dot finds clip moments in a new interview transcript, prepares show notes and drafts social posts for approval, then carries edits across the set. An early tester's Dot also noticed a forgotten invoice, prepared it and sent it only after approval.
  3. Practice, one live run, then a weekday schedule. The Rundown's setup guide runs a fictional practice brief with an answer key, then one live brief on approved sources, then a weekday 8:30 brief in your IANA time zone that reuses the checked rules and asks the Dot to confirm the task name, sources, time zone, next run and repeat schedule.

Talking to your Dot doesn't count toward ChatGPT message limits, but Work or Codex tasks it starts on a connected computer use those products' shared allowance (pricing).

Official references

Related

Copy this task brief

ChatGPT Dots / OpenAI Dots workflows: the setup order, where saved schedules go wrong, and copyable briefs for morning briefs, meetings, research watch, content and project follow-up.
Download Markdown ↓
Workflows

How to set up a ChatGPT Dot morning brief: schedule, delivery checks and 20 real daily-brief examples (OpenAI Dots)

Set up a ChatGPT Dot (OpenAI Dots) morning brief in seven steps: name every calendar, ask for a fixed time with time zone and end date, deliver to chat first, then check the task under Scheduled. Eight places briefs get stuck, with fixes, and 20 real daily-brief setups.

Workflows

How to set up ChatGPT Dots for meetings: a brief before every call, scheduling rules, calendar-change alerts and follow-ups (25 real setups, OpenAI Dots)

Connect every calendar plus the mail or Slack where meetings get agreed, make sends and invite edits ask first, run one brief by hand, then add scheduling rules and a change watch. Copyable message, common fixes and 25 real ChatGPT Dots (OpenAI Dots) meeting setups.

Workflows

How to use a ChatGPT Dot as a project manager: setup, status boards and 19 real project follow-ups (OpenAI Dots)

Yes, a ChatGPT Dot (OpenAI Dots) can follow a project for you. Start with the short version: what it does well, six setup steps, where it gets stuck and what worked, and one message to start with. Then 19 real project follow-ups, from an App Store review watched through to approval to boards the Dot keeps current, plus what to tell it before an overnight run.

Workflows

How to set up a ChatGPT Dots research watch: competitor, price and news alerts, with 50 real setups (OpenAI Dots)

What a ChatGPT Dot (OpenAI Dots) watch is good for, seven setup steps (sources, a baseline, what counts as a change, a real end date, where alerts go) and eight places watches get stuck, with fixes. 50 real setups, from 30 competitor sites and hourly rounds over 4,891 inns to price and news alerts.

Workflows

How to set up a content pipeline with ChatGPT Dots: blog drafts, scheduled posts and 38 real examples (OpenAI Dots)

What a ChatGPT Dot (OpenAI Dots) can do in a content pipeline, seven setup steps (one source, a short brief, drafts in a named place, publish rules, your own sign-ins, then a schedule) and eight places it gets stuck, with fixes. 38 real examples, from WordPress drafts approved before publishing to three Xiaohongshu posts a day.

Sources & further reading