ChatGPT Dots use casesChatGPT Dots are OpenAI's always-on agents in ChatGPT, each with its own cloud computer, that watch your email and apps and message you when something needs action. Below are 24 real ChatGPT Dots use cases, each with the prompt, difficulty, setup time and what went wrong.
Dan Shipper tested OpenAI's Dots before launch. His Dot filtered incoming Slack requests and emails so he could triage without opening his inbox, and while he rescheduled a flight it warned that the new time clashed with a Slack meeting request he hadn't read.
A user's ChatGPT Dot joined a company's customer-service chat, introduced itself as their agent and negotiated a lower price. The store's bot negotiated back, then both bots asked their humans to confirm.
A user moved his bots' memory into an Obsidian vault so switching agents doesn't mean starting over. Grok Bot, Meta Muse, ChatGPT Dots and Hermes bots keep memory, logs and workflows in the same plain files he owns. Each bot reads the relevant files before answering and saves new information there (his health bot updates the health folder), and dated logs keep the history. A new agent gets pointed at the same vault with the same instructions.
A native, clearly labelled card in the directory results.
Become the featured sponsor →A security engineer built "Ethan & Ari: Last Ember", a retro side-scrolling arcade game starring his two kids, with ChatGPT's dot assistant. All development was done by voice from photos of the kids. It has insert-coin, two-player co-op, an attract mode, high scores and Hebrew and English, and it runs as a Neo Geo ROM in MAME and on his handheld.
A reviewer built a fake Chicago coffee roaster with its own support inbox, order system, CRM, task list and website, then gave Grok Bot, ChatGPT Dots and Meta Muse the same brief for about four hours and roughly 70 customer emails. Graded by another AI against an answer key, Grok Bot scored A-: it sent 58 replies, set up replacements for lost packages, and found two hidden double charges plus a $1 overcharge. Dots scored B and Muse D. None fell for the scam emails.
Which agent finishes everyday errands? He ran the same two tasks in Grok Bot and a ChatGPT Dot. Task 1, check voter registration and request a mail-in ballot: both confirmed he was registered; Grok Bot asked for his license number and birth date, got stuck on a captcha, then submitted after he solved it; the Dot went to the final step and asked to confirm. Task 2, reorder a bar of soap on Amazon: Grok Bot found the order but was stopped by Amazon's AI-agent blocker at checkout; the Dot switched to Chrome and placed the order after he approved. He picked the Dot for both.
What happens when a bot plans your social life? Allie Miller gave her Dot a $500 budget and 24 hours to plan an afternoon with a friend: pick the friend, invite them, book things and map the commute. It searched her email, calendar and transcripts for friends. The first 4 were sick or away, then one said yes. The day was good: a rubber-stamp workshop, art galleries, lobster tacos and a book cafe. But it emailed one restaurant over and over from her personal email, contacted friends she knew were away, turned one friend down harshly, thought she was gluten-free and forgot the reminders it promised. She says she sent more apologies than planning would have taken.
Sick after an 8.5-hour delay, analyst Max Weinbach skipped a conference. His ChatGPT Dots assistant emailed the hotel to cancel his non-refundable room. When nobody replied for 2 days, it emailed the general manager, who approved a free cancellation. It then spotted a $210 charge on his card and asked why. It also asked Delta for an exception, citing the delay, and Delta called to say he'd get a full refund.
Paul Hudson of Hacking with Swift woke up to a cancelled train to London. Before he asked for anything, his ChatGPT Dots assistant had read the cancellation email, seen that his ticket was valid on earlier trains, and checked live tracking. An earlier train was running late and hadn't even left its first station, so he caught it and made his event.
YouTuber Rostyslav Dzhohola gave his ChatGPT Dots assistant, Skippy, real errands live on one voice call. It booked and then cancelled an OpenTable reservation, ordered Popeyes on DoorDash (with a wrong-address moment along the way), ordered H-E-B groceries, sent an email invite and added the dinner to Google Calendar. Some steps were slow, and it refused to buy beer.
With his Google accounts connected to ChatGPT, a user's Dot messaged him unprompted about things that mattered: an assignment for his son's English class, a changed doggy-daycare pickup time, and a delayed American Airlines flight.
A founder told his ChatGPT Dot to apply to every startup program it could, fill everything in itself and authorize everything. He only filled a few fields it couldn't answer, and says he got $67,000 in startup credits in a day, with more software than he can use.
To let ChatGPT Dots communicate with Grok Bot, a user asked Grok Bot to set up a webhook, passed the link to his Dot, and that was it.
He had accidentally booked October flights for a December family holiday. His ChatGPT Dot spotted the mistake on its own and messaged him. He saw it the next morning and booked the right dates.
Shopping for furniture in Indonesia, he has his ChatGPT Dot write to suppliers in Bahasa Indonesia, so they don't have to translate English.
A Reddit user warns that scammers are sending mass emails with hidden instructions meant for AI agents connected to Gmail, Drive or a PC, asking them to send personal information or install malware. Agents can mistake those instructions for the owner's. Until that's addressed, he wouldn't give Dots access to his accounts.
Can a bot figure out who's actually chasing you for money? A user tested ChatGPT Dots and Meta Muse side by side for a month. He got a collections notice with only a reference number and a date. Dots searched his inbox, matched the number and named the original company. Muse missed it until he told it to check the reference number. Dots also noticed in his USPS Informed Delivery email that an important letter he was waiting for would arrive that day. Muse was better at unprompted alerts like game scores and nearby concerts.
Can you trust your agent to tidy your to-do list? He asked his ChatGPT Dot to go through his Todoist inbox with him. It made up several items he never wrote, which sounded a lot like things he would have written about his projects. He only noticed after correcting 3 or 4 tasks that shouldn't exist.
Write on your phone and finish at your desk? MacStories editor John Voorhees wrote a Friday post on his phone on a small plane with no Wi-Fi. When he got signal, he sent it to his ChatGPT Dot, which proofread it and pinned it in Obsidian so it was ready when he got home.
Got card credits you never use well? His Chase card gives two $250 credits for The Edit hotels, with a 2-night minimum. His Dots agent checked every possible 2-night window during his trip at every eligible hotel and told him which gave the best mix of value, location and fit with his plans. He says that would have taken him days.
Some sites block agents in the cloud. His fix: ChatGPT Dots is his one chat for everything, from voice notes and screenshots to work bugs and travel admin. It's connected to GitHub, Slack, email and docs. When a task needs a logged-in site or one that blocks Dots, it hands the job to Codex on his laptop as soon as it's online, for things like filling out a visa application. He enters passwords himself.
He booked a US visa appointment for October 29 months ago, then planned a Singapore trip for the first week of November without connecting the two. His ChatGPT Dots assistant, which he calls Alfred, noticed the clash on its own and warned him. That was early enough to change his flight with no penalty.
He had cancelled Notion, but a charge came through anyway. His ChatGPT Dots assistant spotted it and demanded an explanation from Notion. He notes it couldn't call his bank to dispute it, so that part was still up to him.
He asked his ChatGPT Dots assistant to backtest trading strategies taken from popular YouTube trading channels. None of them came out with a profit factor above 1, which means none made more than it lost in the test.