ChatGPT dots: 10 practical uses, with a free guide for each

By Kitson Workshop · 3 October 2026 · 9 minute read

OpenAI launched dots on 29 September 2026. A dot is what OpenAI calls an "always-on agent": it lives in ChatGPT and keeps working between your conversations, on its own cloud computer and browser, so it carries on even when your computer is off. It can use apps you connect, such as your calendar and email, and you can message it in ChatGPT, Slack or Teams.

That's powerful, and it's also exactly why the first job you give it matters. Below are the ten jobs an expert panel we put together (a small-business consultant, a productivity coach, an AI-safety and privacy specialist, a consumer-protection educator and a teacher of adult learners) rated most useful and safest, based on OpenAI's own pages as of 2 October 2026.

Before you start

• Plan and age. At launch, dots are for ChatGPT Pro users aged 18 or over outside the EEA, UK and Switzerland, and for Business Premium in all supported regions. Enterprise, Edu and Healthcare workspaces need an admin to turn them on. Rollout is gradual.
• Set it up on a computer. You create your dot in the ChatGPT desktop app or a desktop browser; OpenAI says mobile web isn't supported.
• Cost. Your first dot is included in your plan, with extended limits for the first month. OpenAI hasn't published prices after that.
• One job at a time. You get one primary dot. OpenAI's own lab tests found more mix-ups between tasks as more tasks piled up, so add one job, check it for a week, then add another.

The four rule levels

Custom Rules tell your dot how much it may do on its own. There are four levels: take action without asking, take action when you say so, ask before taking action and hand off to you. OpenAI describes rules as instructions the dot tries to follow, and says it can make mistakes, so treat them as a safety net. Rules apply to the whole dot, not just one task. The panel's baseline is on our ChatGPT dots page; in short, anything that sends, shares or changes things is "ask", and buying, paying, deleting and personal forms are "hand off to you".

The 10 uses

1. A morning briefing

Every weekday at a time you choose, the dot reads today's calendar and recent email and sends you one short message: what's on, any clashes or travel time, what to prepare, and the emails most likely waiting on you. It only reads, and it only writes to you, which makes it the best first job. It's close to OpenAI's own example.

Sample instruction: "Every weekday at 8 am Eastern time for the next four weeks, look at my calendar for today and my email from the last 24 hours. Send me a single short message: what's on today, any clashes or travel time, what I need to prepare, and the three emails most likely waiting on me. Don't reply to anyone or change anything. Confirm the schedule."

Check it against your real calendar every day in the first week, including the time zone. Free ebook · video

2. A weekly inbox tidy-up

Once a week it sorts mail into "needs a reply", "waiting on someone", "newsletters and receipts" and "probably junk", drafts replies to the first pile and suggests what to archive. Drafts: take action without asking. Sending: ask first. Archiving: when you say so. Deleting or unsubscribing: hand off to you. Read every draft for names, dates and promises before it goes. Free ebook · video

3. A forgotten-invoice catcher

For freelancers and small firms: every Monday it compares finished jobs (from email, calendar or a jobs sheet) with invoices already sent, lists anything not yet billed, and prepares draft invoices for you to approve. OpenAI's launch story describes an early tester whose dot spotted a missed invoice and sent it only after he approved. Check amounts, tax lines, names and invoice numbers, and never give it bank logins; it doesn't need them to make an invoice. Free ebook · video

4. A renewals, free trials and bills radar

Once a week it scans your email for renewals, trials ending, price rises and bills due, keeps one list with dates and amounts, and warns you seven days ahead. It never cancels, pays or clicks links in those emails; you do that yourself on the company's own site. Not every subscription emails you, so look for gaps. Free ebook · video

5. A study planner and deadline keeper (18+)

From course outlines you share, it builds a deadline list and sends a plan each week. It plans and quizzes; it doesn't do assessed work, which may break your institution's rules. Check every date against the official course page. Free ebook · video

6. A customer-enquiry first responder

A few times a day it sorts new enquiries into quotes, bookings, complaints and spam, drafts a reply for each from your price list and FAQ, and flags urgent ones. Sending to customers: ask first (OpenAI's own example). Refunds, discounts and price promises: hand off to you. Customer details enter its memory, so check your own privacy duties first. Free ebook · video

7. A family logistics keeper

Twice a week it reads school and club emails and keeps one family list: dates, forms to sign, money to pay, kit to pack. It suggests calendar entries and asks before adding them. Payments and consent forms stay with you, on the official portal. It works from your inbox only; it is not for watching anyone else's messages. Free ebook · video

8. A job-search tracker

Each morning it checks the job pages you name, adds matching roles to a sheet, drafts cover-letter starting points from your CV and reminds you to follow up after a week or so. You submit every application and send every email yourself. Check each job is real and still open on the employer's own site; some job sites block cloud browsers. Free ebook · video

9. Compare before you buy

Give it a need and a budget. It researches options on the open web, builds a comparison of price, delivery, reviews, returns and total cost, and re-checks prices for a few days. You make any purchase yourself. Leave purchases at "hand off to you", and check price and stock on the retailer's own site before buying. Free ebook · video

10. From recording to show notes

When a new transcript appears in a folder (or on a schedule), it drafts show notes, clip ideas with timestamps and a few social posts in your style. You review and publish; posting stays at "hand off to you". Check quotes, names and timestamps, and make sure people in the recording are happy to be quoted. Free ebook · video

The honest limits

• Memory. What your dot reads, it can remember. You can't view, correct or delete single memories, only the whole dot, and disconnecting an app doesn't erase what it already learned. Connect only what the job needs.
• Schedules. Pausing the dot stops the current task, not scheduled runs. Turn schedules off in Scheduled. Ask your dot each week: "List everything you're scheduled to do."
• Training. Business, Enterprise and Edu content isn't used for training by default. On a personal Pro plan, check "Improve the model for everyone" in ChatGPT's data controls.
• Hidden instructions. Emails and web pages can contain instructions meant to trick AI agents. OpenAI says its protections reduce this risk but don't remove it.
• Never your bank. Don't give a dot bank logins, card numbers, ID numbers or passwords in a chat or document. OpenAI says money transfers and password changes are always handed back to you.

What we left out, and why

The panel rejected paying bills or moving money automatically, watching a partner's or teenager's messages, medical decisions, buy/sell alerts for stocks or crypto, and "one dot that runs your whole life". Each one either needs access a dot shouldn't have, watches someone without consent, or piles up exactly the kind of mixed responsibilities where OpenAI measured more mistakes.

Kitson Explains is independent: it is not made, sponsored or endorsed by OpenAI. "ChatGPT" and "dots" are OpenAI's names. Facts here come from OpenAI's own pages as read on 2 October 2026; features, plans and screens may change. For educational purposes only: not financial, legal, medical, tax or careers advice.