4 October 2026 · 10 min read
10 Insane ChatGPT Dots Use Cases
Ten jobs worth handing a ChatGPT dot: what each one gives you back, how far it's proven, and the instructions to copy.
By Manu Jayawardana, TAI LabsHow we researched this
On this page
What is a ChatGPT dot?
A ChatGPT dot is an AI assistant you give a job, not a question. It keeps doing that job after you close the chat: it works in the apps you connect, on a schedule or when something changes, and messages you when it has a result or needs a decision.
- Included with
- ChatGPT Pro (from $100 a month) and Business Premium
- In the UK and Europe
- Business Premium only, for now
- On Enterprise
- A beta your workspace admin can turn on
- Not yet
- Free, Go and Plus, ChatGPT's cheaper plans
OpenAI launched dots on September 29, 2026. Sources: OpenAI's Help Center, announcement, docs and DevDay recap. The full table is in Can you get a ChatGPT dot yet?
A real example
Sam Altman gave up on a search. His dot kept looking.
His dot
One day
Altman searches Slack for something he's sure he saw. Nothing comes up, so he lets it go.
Overnight
His dot keeps looking. The thing was a screenshot, so it goes through the images.
Next day
It tells him it found it.
“I kept thinking about it overnight and I found it for you.”
A chat helps when you ask. A dot keeps the job after you leave.
How a dot works
You give it a job
What to do, which apps it may use, and what it may do without asking you.
It works on its own
On its own computer in the cloud, in the apps you connect. It keeps going with your laptop off.
It comes back to you
It messages you in ChatGPT, Slack or Teams, or calls you, when it has a result or a question.
If it decides by itself to look into something, it can only read. Anything that sends, changes or pays follows the rules you set.
The 10 ChatGPT Dots use cases
1. It spotted a clash that wasn't on the calendar
Dan Shipper had his dot research a flight. It warned him the new time might clash with a colleague's video recording, which wasn't on Shipper's calendar. The dot had read about it in a Slack discussion he hadn't seen.
Good to know: It only reads the channels you connect, and early testers hit glitches: Shipper had dropped messages and permission problems, and suggested waiting a week or two.
Tried by a reviewer: Dan Shipper, who runs the newsletter Every, in his review.
Before you book that flight
- The new time may clash with the video recording discussed in Slack.
- It isn't on your calendar.
- Want me to look for another flight?
Set it up: the instructions to copy
How it runs
- You ask about a change. A flight, a meeting, a deadline.
- It reads around it. Your calendar and the Slack channels it can see.
- It spots a possible clash. Including plans that only came up in Slack.
- It tells you first. With the message it came from.
The instructions to paste in
Before I commit to a time, a trip or a deadline, check it against my calendar and the Slack channels you can read. If anything might clash, including plans that only came up in Slack, tell me what and where, and suggest one fix. Don't message anyone else.
- Connect
- Google Calendar, Slack
- Setting to choose
- Does it on its own (OpenAI’s setting: “Take action without asking”)
2. It noticed you never sent the invoice
A tester did some work for a publication and forgot to bill for it. His dot noticed, prepared the invoice, and sent it once he approved. It's one of OpenAI's own launch stories.
Good to know: Make it ask you before it sends anything. An invoice to the wrong client is worse than a late one.
OpenAI's example: OpenAI's launch announcement. OpenAI doesn't say how the dot spotted it. In the instructions below, it checks every Friday.
One invoice still to send
- Work for the publication, delivered
- No invoice sent for it
- Draft invoice ready. Send it?
Set it up: the instructions to copy
How it runs
- Every Friday. It checks the week's delivered work.
- Matches invoices. Against your sent mail and invoice folder.
- Drafts what's missing. From your last invoice to that client.
- Waits for your yes. Then sends it.
The instructions to paste in
Every Friday, compare the work I delivered this week (email and Drive) with the invoices I've sent. If something has no invoice, draft one from my last invoice to that client and show it to me. Send it only after I approve it.
- Connect
- Gmail, Google Drive
- Setting to choose
- Asks you every time (OpenAI’s setting: “Ask before taking action”)
3. It answered all but two questions on an insurance form
Casey Newton gave his dot an office insurance form. It worked through his lease, city websites and budget documents, and left him just two questions to answer.
Good to know: Check each answer against the source it cites. That's where a confident mistake would hide.
Tried by a reviewer: Casey Newton, who writes the Platformer newsletter, in his report.
Office insurance application
- Answers from the lease
- Answers from city websites
- Answers from budget documents
- Two questions left for you
Set it up: the instructions to copy
How it runs
- You hand it the form. A link or a PDF.
- It hunts for answers. Your documents and official websites.
- It fills in what it can. With the source of each answer.
- It hands back the rest. The questions only you can answer.
The instructions to paste in
Fill in this form: [link]. Use [folder] and official public websites, and note where each answer came from. Leave anything you're unsure of blank and list those questions for me. Don't submit it.
- Connect
- Google Drive
- Setting to choose
- Asks you every time (OpenAI’s setting: “Ask before taking action”)
4. It catches a launch email that still promises a dropped feature
OpenAI's docs describe it: before the launch meeting, a dot notices the customer email still promises a feature that's no longer planned, checks the latest product decisions, and suggests what to change.
Good to know: It only knows decisions someone wrote down.
OpenAI's example: An example in OpenAI's dots documentation.
Launch email, latest draft
- Out of date:
Offline mode is included at launch. - Offline mode was cut in the latest product review
- Suggested: Suggested: drop the line, or say it's coming later
Set it up: the instructions to copy
How it runs
- On launch-meeting mornings. On a schedule you set.
- Reads the drafts. Emails, pages, release notes.
- Checks the decisions. The latest specs and meeting notes.
- Suggests edits. With the decision each one comes from.
The instructions to paste in
On the morning of each launch meeting, read every draft that goes to customers (emails, pages, release notes) and check each promise against the latest decisions in [doc or channel]. List anything out of date, the decision it conflicts with, and a suggested rewrite. Don't edit the drafts.
- Connect
- Google Drive, Slack
- Setting to choose
- Does it on its own (OpenAI’s setting: “Take action without asking”)
5. One rough idea, five or six versions to choose from
Sam Altman says that in spare moments he gives his dot an idea for a feature, and it builds five or six versions for him to review.
Good to know: The first month doesn't count toward your plan's usage limits. After that, every version does (release notes).
OpenAI's example: Sam Altman, OpenAI's CEO, on Every's podcast.
- AOne-screen checklist
- BGuided tour
- CSample project
- DShort video
- EProgress bar
Set it up: the instructions to copy
How it runs
- You describe the idea. Typed, or out loud.
- It builds variations. Each one a different approach.
- It sums them up. What each does, and its trade-off.
- You pick. Or ask for another.
The instructions to paste in
Idea: [one paragraph]. Build five prototypes that each take a different approach. For each one, give me a one-line summary, the trade-off it makes and a link. Don't publish anything or spend money.
- Connect
- Its own computer
- Setting to choose
- Asks you every time (OpenAI’s setting: “Ask before taking action”)
6. A watchlist that only speaks when something changes
Point a dot at pages you keep re-checking, like a competitor's pricing or a regulator's notices, and it messages you only when one of them changes.
Good to know: Some websites block it. In one review, a restaurant's ordering page did, and an IKEA login trapped it in a security check (review summary). If it stays quiet, check its activity log to make sure it really looked.
From OpenAI's docs: OpenAI's docs say you can tell a dot what to check, when, and what's worth a message.
Two changes since yesterday
- Competitor A added a Team plan
- Competitor B removed its annual discount
- Every other page unchanged
Set it up: the instructions to copy
How it runs
- On a schedule. Weekdays at 8, in your time zone.
- Reads each page. In its own browser.
- Compares with last time. Against the notes from its last run.
- Messages only on change. With the link and one line on why.
The instructions to paste in
Every weekday at 8am [time zone], check these pages: [links]. Compare each with what you saw last time. Message me only when something changed, with the link and one line on why it matters. Otherwise stay quiet.
- Connect
- Its own browser, Slack
- Setting to choose
- Does it on its own (OpenAI’s setting: “Take action without asking”)
7. The Monday numbers, flagged before the meeting
In OpenAI's examples, a dot tracks product usage and revenue, flags what changed and refreshes the slides. Put that on a schedule and it's ready before Monday's meeting.
Good to know: Check the numbers before the meeting. One tester's spreadsheet formulas broke when opened in Google Sheets.
OpenAI's example: OpenAI's examples, as reported by Tom's Guide. In the instructions below, it runs every Monday.
Weekly metrics, refreshed
- Every changed number, listed at the top
- Sign-ups dipped mid-week: a first guess at why, on slide 4
- Slides 3 to 5 updated, the rest untouched
Set it up: the instructions to copy
How it runs
- Every Monday, 7am. Before anyone's in.
- Pulls last week's numbers. From the sheets and tools you connect.
- Flags what moved. Anything past your threshold.
- Refreshes the slides. And lists every number it changed.
The instructions to paste in
Every Monday at 7am, pull last week's [metrics] from [source]. If anything moved more than [10%], give me your best explanation and the evidence. Refresh slides [3 to 5] of [deck] and list every number you changed.
- Connect
- Google Drive
- Setting to choose
- Does it when you've asked (OpenAI’s setting: “Take action when you say so”)
8. A proposal that catches up by the end of the day
Deals change on calls and in email, and the proposal falls behind. One of OpenAI's launch examples is a dot that keeps a sales proposal current.
Good to know: The least proven job here: we found no public report of anyone doing it yet. OpenAI's dot docs don't name any sales software, so check yours connects to ChatGPT first.
OpenAI's example: OpenAI's launch examples (announcement).
Updated after today's call
- Out of date:
Start date as agreed last month - Suggested: New start date, from today's call notes
- Change logged, with its source
Set it up: the instructions to copy
How it runs
- Each evening. On the deal you name.
- Reads what changed. Call notes and the deal's email.
- Updates the proposal. Scope, price, dates.
- Logs the change. What changed, why, and the source.
The instructions to paste in
You own [proposal doc]. Each evening, check the call notes and email for [deal]. If the scope, price or dates changed, update the doc and add a line at the top: what changed, why, and the source. Never send it to the client.
- Connect
- Gmail, Google Drive
- Setting to choose
- Does it when you've asked (OpenAI’s setting: “Take action when you say so”)
9. The clips from the interview finally get made
In OpenAI's example, when a new interview transcript arrives, the dot picks the best moments for short clips, writes the show notes and drafts social posts, then learns from your corrections.
Good to know: Your voice takes a few rounds. Don't post the first week's drafts unread.
OpenAI's example: OpenAI's launch examples (announcement).
New episode: drafts ready
- Three clips, with timestamps
- Show notes
- Three posts in your voice, waiting for your edits
Set it up: the instructions to copy
How it runs
- Each morning. It checks your transcripts folder.
- Finds the best moments. With timestamps.
- Drafts the rest. Show notes and posts in your voice.
- Learns your edits. Next week's drafts start closer.
The instructions to paste in
Each morning, check [folder] for new interview transcripts. For each one, send me the three strongest clips with timestamps, show notes and three posts in my voice. Use my past posts in [doc] as the style guide and learn from my edits. Don't post anything.
- Connect
- Google Drive
- Setting to choose
- Does it on its own (OpenAI’s setting: “Take action without asking”)
10. A bug report in Slack, a fix waiting for review
A bug report lands in Slack. In OpenAI's examples, the dot looks into it, builds and tests a fix, and leaves it waiting for an engineer to approve.
Good to know: For software teams. An engineer still reviews every fix, and the coding work counts toward your plan's coding limits (docs).
OpenAI's example: Examples in OpenAI's dots documentation and announcement. Its docs on apps name GitHub as where the fix goes.
OpenFix: export fails on large files
- Found the problem from the Slack report
- Fix, plus a test so it can't come back
- Waiting for an engineer to approve
Set it up: the instructions to copy
How it runs
- A bug lands in Slack. In the channel you connect.
- It looks into it. The report and the code.
- Builds and tests a fix. In a separate copy of the code.
- Asks for review. An engineer checks it and decides.
The instructions to paste in
Watch the [bugs] Slack channel. When a new report comes in, try to reproduce it. If you can, write a fix with a test and open a pull request that links the report. Never merge it. Tag me in Slack when it's ready.
- Connect
- Slack, GitHub
- Setting to choose
- Does it when you've asked (OpenAI’s setting: “Take action when you say so”)
Where to start
Try first
Jobs that repeat often, where a mistake is cheap to catch.
Try next
Jobs that repeat, but check each result.
When it comes up
One-off jobs. Try one when it comes up.
The real bottleneck is you
Casey Newton's estimate, across a few jobs
- The dot's work120 min
- Newton's time15 min
The two hours were the dot's. The fifteen minutes were his, and that's the part that stays yours: writing the instructions and checking what came back.
That was the most common complaint when dots launched, according to a round-up of the launch discussion: the limit is people's time to check the work, not the computer. So count the checking time before you call it a saving.
Can you get a ChatGPT dot yet?
| Plan | Most countries | UK and Europe |
|---|---|---|
| ChatGPT Pro | Most countriesYes. One dot included | UK and EuropeNo. Not at launch |
| Business Premium | Most countriesYes. One dot included | UK and EuropeYes. One dot included |
| Enterprise, Edu, Healthcare | Most countriesPartly. Beta, once an admin turns it on | UK and EuropePartly. Beta, once an admin turns it on |
| Other Business plans | Most countriesNo. Not named at launch | UK and EuropeNo. Not named at launch |
| Free, Go, Plus | Most countriesNo. Not at launch | UK and EuropeNo. Not at launch |
Pro starts at $100 a month (OpenAI, CNBC). For now it's one dot each; OpenAI says you'll be able to add more later (announcement). For the first month, using your dot doesn't count toward your plan's usage limits (release notes).
How to set up your first dot
Start on a computer
Create it in the ChatGPT desktop app or on the ChatGPT website. OpenAI says you'll be able to talk to it on your phone once mobile access arrives (Help Center).
Connect only the apps this job needs
Any app you connect here is connected for all of ChatGPT, not just this dot (safety FAQ). So connect only the two or three this job needs.
Set two rules
Make it ask you before it sends any message, and hand anything that spends money back to you. OpenAI calls these Custom Rules (controls).
Give it one job
Pick a job from the Where to start list above and paste its instructions, or fill in the template below.
Correct it on day two
Open its activity log, fix one thing and say why. Dots learn from feedback (announcement).
What it may do without you: the four settings
Does it on its own
Reading, searching and drafting.
OpenAI’s setting: “Take action without asking”
Does it when you've asked
The job you gave it, like updating the Monday slides.
OpenAI’s setting: “Take action when you say so”
Asks you every time
Emailing a client, publishing, deleting.
OpenAI’s setting: “Ask before taking action”
Always hands it back to you
Paying for things or changing a password.
OpenAI’s setting: “Hand off to you”
Write your own instructions: fill in these seven lines
- The job
- Watch our three competitors' pricing pagesPhrase it as a job a person could own.
- Done looks like
- One Slack message, only when something changedWithout it, you get an essay.
- It may read
- Those three pages and our #pricing channelConnect only what the job needs.
- When it runs
- Weekdays at 8am, London timeScheduled jobs take a time zone and an end date.
- Never without asking
- Send, publish, delete, buyThe short list that keeps you out of trouble.
- Bring me in whenOften forgotten
- A price changes by more than 10%Your rule for what deserves a human.
- Report backOften forgotten
- The link, what changed and why it matters, in two linesThis is what makes checking it quick.
What doesn't work yet
Paying for things was one of the jobs dots couldn't do well in their first week. A reviewer at AI Agents Library, a site that tests AI agents, tried dots on ten business tasks: they were best at building web pages and planning content, and worst at spreadsheets, sorting an inbox and following up after a meeting.
Paying
Who hit it: One tester's verdict, reported by Yahoo Tech: payments don't really work yet.
What to do: Let it compare prices and do the paying yourself. Set a rule that hands anything that spends money back to you.
Sending before you've looked
Who hit it: Corey Noles, reviewing dots for The Neuron newsletter, had one send an email before he'd checked it.
What to do: From day one, make it ask you before it sends anything.
Logins and security checks
Who hit it: An IKEA login trapped one reviewer's dot in a security check (review summary).
What to do: Expect sign-ins and 'are you human' checks to come back to you.
It can stall
Who hit it: At the launch event, a dot asked on stage to catch up on last night's user testing went quiet (Futurism).
What to do: Check its activity log every day for the first week.
Hidden instructions
Who hit it: Our caution: a web page or email could hide text telling your dot to do something. OpenAI says it double-checks certain actions before they run (safety FAQ).
What to do: Make it ask you first before anything that sends, deletes or pays.
If your company is on ChatGPT Enterprise
An admin switches it on
On Enterprise, Edu and Healthcare plans, dots are an early (beta) version that stays off until a workspace admin turns it on (admin guide).
Admins decide how far it reaches
Separate settings cover putting dots in Slack and Teams, letting them use an employee's own computer, and the rules for what dots may do alone (admin guide).
Connecting an app is a company decision
App permissions are shared between dots and OpenAI's other ChatGPT tools (safety FAQ), so treat connecting a client folder like any other data-access decision.
Pilot three jobs, not thirty
Our advice: start with the Try first jobs, give each job one owner, and time the instructions and the checking for two weeks before you widen it.
ChatGPT Dots FAQ
What are ChatGPT Dots?
AI assistants inside ChatGPT, launched by OpenAI on September 29, 2026, that keep working on a job after you close the chat. Each runs on an OpenAI model called GPT-6 Astra, with its own cloud computer, works in the apps you connect, and messages you when it has results or needs a decision.
How are dots different from normal ChatGPT?
A chat responds when you ask, then stops. A dot holds a job: it works on a schedule or when something changes in a connected app, keeps its own notes between runs, and gets in touch when it has a result or a question.
Can a dot work while I'm away?
Yes. Its cloud computer can keep working when yours is off, and scheduled jobs run at the times you set. When it researches on its own initiative, OpenAI limits it to reading: it can't send messages, change what's in your apps or control your browser or computer.
How much do ChatGPT Dots cost?
One dot is included with ChatGPT Pro, which starts at $100 a month, and with a Business Premium seat. Pro doesn't include one in the UK, the European Economic Area or Switzerland at launch. Enterprise workspaces can turn on an early (beta) version. Free, Go, Plus and the other Business plans weren't named at launch.
Can I use ChatGPT Dots in the UK or Europe?
Not on Pro at launch, which leaves out the UK, the European Economic Area and Switzerland. Business Premium includes dots in every supported region, so it's the route for UK and European teams for now.
Can a dot send email or spend money without asking?
Whether it sends on its own depends on the rules you set, and one reviewer's dot sent an email before he'd looked. OpenAI says the most sensitive actions, like changing a password or transferring money, need you to take over. For anything else that spends money, set a rule that hands it back to you.
What should I give my first dot?
A job that comes back often and is cheap to get wrong: a watchlist of pages, a clash check on your plans, or first drafts of clips from an interview transcript. Make it ask you first before anything that sends, publishes or pays.
Sources
OpenAI
- Introducing dotsopenai.com
- DevDay 2026 recapopenai.com
- Your dot: setup and availability (Help Center)help.openai.com
- Dots privacy, security and safety FAQs (Help Center)help.openai.com
- Manage dots in ChatGPT workspaces (Help Center)help.openai.com
- ChatGPT release noteshelp.openai.com
- Meet dots (docs)learn.chatgpt.com
- Control your dot (docs)learn.chatgpt.com
- Tasks and memory (docs)learn.chatgpt.com
- Channels (docs)learn.chatgpt.com
- Computers and apps (docs)learn.chatgpt.com
Reviews and reporting
- Every: How Sam Altman uses dots to take back his timeevery.to
- Every: Dan Shipper's vibe check of dotsevery.to
- Platformer: Casey Newton on dotsplatformer.news
- The Neuron: My first day with OpenAI Dotstheneuron.ai
- AI Agents Library: I tested 10 business tasksaiagentslibrary.com
- Summary of a hands-on review: errands, website, videozetik.com
- Yahoo Tech: a tester on dotstech.yahoo.com
- Tom's Guide: what the new agents can dotomsguide.com
- CNBC: OpenAI follows Meta into personal agentscnbc.com
- Futurism: the live demo that stalledfuturism.com
- DEV Community: practitioners shipped skepticismdev.to
Community
- Hacker News launch threadnews.ycombinator.com
How we researched this
- We relied on search summaries of OpenAI's announcement, Help Center articles, docs and release notes on dots, as they appeared on October 4, 2026; we couldn't open the pages themselves. Product facts come from there first, and setting names are OpenAI's own.
- We saw every hands-on review and community thread through search results too. One review we saw only through a third-party summary, and that summary is the link we cite.
- Each use case says where its proof comes from: a reviewer who tried it, OpenAI's own example, or OpenAI's docs. The "what you'd get" panels are illustrations, not screenshots.
- We haven't run a dot of our own for this article, so nothing here is presented as our test. Where to start is our judgement and says so.
- AI tools helped research and draft this article. Claims in the body link to their sources, and the FAQ restates them. Spot something out of date? Tell us.