Give it responsibilities, not prompts.
Dot is ChatGPT's always-on agent. Here is how to set yours up in about 15 minutes, keep it safe, and give it real responsibilities instead of one-off prompts. Every prompt is copy and paste ready.
Free. No email required, nothing to download.
OpenAI launched dots at DevDay on 29 September 2026. Right now they are only on ChatGPT Pro and Business Premium plans, in eligible markets. Your first dot is included in those plans at no extra cost. Plus and Free users do not have it yet, and OpenAI says more users are coming soon.
Not on Pro? Bookmark this page. The setup steps and prompts will still apply when it reaches your plan.
Dot is an AI agent that keeps working when you close the chat. Normal ChatGPT answers a request. A dot takes on a responsibility and keeps it moving.
The easiest way to think about it. Normal ChatGPT: "Do this task." A dot: "This is now your responsibility. Keep it moving." You can call your dot, give it something to work on, hang up, and the work carries on.
Create your dot in the ChatGPT desktop app or ChatGPT in a desktop browser. During setup you can give it a name, customise how it looks, connect apps and optionally connect your computer.
Once it is set up you can reach the same dot from your phone too. It keeps the same context everywhere, so it never starts from scratch.
We called ours Freddy. Naming it sounds superficial, but it reinforces the right mental model: this is your ongoing assistant, not another random chat.
This is where your dot starts being useful. Open its profile and connect the apps that matter for your work.
Your existing app permissions in ChatGPT apply to your dot, so you control what each connection can read or change.
Do not connect everything just because you can. Ask: does my dot need this to do a job I am giving it? If not, leave it off.
Your dot already has its own cloud computer and browser. It can research, create files, analyse information and keep working with your laptop switched off. That is ideal for overnight research, long-running analysis, document creation, checking websites and preparing work for the morning.
You can also connect your own computer. Do that when your dot needs local files, local apps, your own browser or an existing local Codex setup. Your computer has to stay on, with the ChatGPT desktop app open, while your dot is using it.
Simple rule: the cloud computer is the default. Connect your own only when your dot needs something that lives on your machine.
You do not have to keep opening ChatGPT to check in. You can reach the same dot through ChatGPT on desktop, web and mobile, voice calls, Slack and Microsoft Teams. Switching channels does not reset anything. Texting is coming soon.
Try telling it where different updates should go.
Keep normal updates inside ChatGPT. If something needs my decision today, message me in Slack.
If I'm away from my desk, I'll call you for an update. Keep working after we hang up.
This step matters most. Go to Settings, then Personalization, then Custom Rules, and decide how your dot handles different actions. There are four options.
Start conservative. Loosen the rules once you have seen how your dot works. A sensible starting setup looks like this.
Do not start firing random tasks at it. Your first conversation should get your dot to map itself against your actual work. That beats asking "what can you do?"
I want you to become my ongoing AI assistant rather than just responding to individual requests. Start by learning how I work. Review the context and connected apps you're allowed to access, then interview me about: - what I do - my main projects - recurring responsibilities - people I regularly work with - tasks I repeatedly forget or delay - work that takes me too long - information I regularly have to check - decisions I want to remain responsible for - actions I'm comfortable with you taking independently Once you understand that, identify the 10 best responsibilities I could delegate to you. Rank them by time saved and usefulness. Don't take any external actions yet. Show me your recommendations first.
After the interview, give your dot standing instructions. Think of this as its operating system.
Here's how I want us to work together going forward. Your job is to keep my work moving without me micromanaging you. Be proactive when you notice something relevant to the responsibilities I've given you. When possible: 1. identify what needs doing 2. gather the context you need 3. prepare or complete the work you're authorised to do 4. bring me the result, not just the problem Don't interrupt me with routine updates. Come to me when: - something needs my decision - an important deadline is at risk - you find something genuinely important - you're blocked - an action needs my approval When you need a decision, give me the context, your recommendation and the options, as briefly as possible. If you're unsure whether you have permission to do something consequential, ask first.
This is the biggest mindset shift. Your dot can decide when to pick up an assigned responsibility again, without you scheduling every follow-up.
Without this, any proactive assistant gets annoying fast.
Don't notify me just because something changed. Notify me only when the change needs an action, creates a risk, changes a decision or is genuinely important.
Cloud work keeps running while your devices are off. Hand your dot something at the end of the day and see what is waiting in the morning.
I'm finished for today. While I'm offline, research [TOPIC] properly. Use multiple credible sources, compare the main options and investigate anything that looks questionable. By tomorrow morning, prepare: - a one-page summary - the five most important findings - anything surprising - recommended next steps - the source material I should review Don't interrupt me unless you hit a blocker that stops the work.
Each one is a responsibility, not a one-off task. Copy the prompt, swap in your details, and let your dot run with it.
A short daily briefing with only what matters.
Keep track of my calendar, important email conversations and ongoing responsibilities. At the start of my working day, give me a short briefing with only: what matters today, anything I've forgotten, deadlines at risk and decisions I need to make. No generic summaries.
Better than a reminder. Your dot can say: "You promised James the proposal tomorrow. I have pulled the meeting notes and started the draft."
Keep track of important commitments I make in meetings and conversations. If I've promised someone something and look at risk of missing it, flag it early. Where possible, gather the context and prepare the work or reply so I can act straight away.
A client asks for a website change. Your dot reads the request, checks past discussion, starts a Codex task and brings you the result.
When an important client request comes in, work out what they're asking for, find the relevant project context and prepare the next piece of work. If it needs ChatGPT Work or Codex, start a task. Don't send anything externally until I approve it.
The goal is not "compare broadband plans". It is "save me money".
Review the bills, subscriptions and renewals I give you. Flag anything that looks expensive compared with current alternatives. Research better deals and work out the realistic saving. For each one worth chasing, draft the message I'd send my current provider to ask for a better price. Don't agree to a new contract, make a payment or cancel anything without my approval.
If a flight gets cancelled, your dot is not just summarising the email. It is already working on plan B.
Keep track of this trip and the booking details I've shared. If a flight, hotel or other important booking changes, work out the best alternatives straight away. Compare them on price and disruption and bring me the best choice. Ask before making any booking or payment.
The purest dot use case. You are not handing over a task. You are handing over an outcome to protect.
We're launching [PROJECT] on [DATE]. Help me keep the launch moving. Track deadlines, important emails, meetings, outstanding work and blockers. If something can be prepared or investigated without me, do it. Only interrupt me when a decision is needed or the launch is genuinely at risk.
This is the one we run ourselves. Much stronger than "send me AI news every day".
Track meaningful announcements from OpenAI, Anthropic, Google, Meta and other major AI companies. Ignore minor product updates and hype. When something genuinely useful for marketers, creators or everyday professionals appears: - verify the announcement - explain what actually changed - find practical use cases - spot anything other creators are missing - prepare a short-form video brief Only tell me when you think something is genuinely worth covering.
An enquiry lands. You open your dot to find the company researched, the opportunity summarised and a reply drafted.
When a legitimate new business enquiry arrives, research the person and company, work out what they're likely looking for, check any previous contact we've had, and prepare a short lead brief plus a suggested reply. Don't contact them until I approve it.
Not every change. Only the ones that move something.
Keep track of these competitors: [LIST]. Watch their websites, public launches, positioning and major announcements. Don't report every change. Alert me when something materially changes their product, pricing, positioning or marketing. When that happens, explain what changed, why it might matter to us and whether we should look into it further.
Do not remind me about the work. Help do the work.
After important meetings, pull together the decisions, commitments, owners and deadlines. Track the items assigned to me and help move them forward. If you already have what you need to prepare the next step, do that instead of just reminding me.
Your dot can split work across background agents and keep coordinating them while you do something else.
Research [PROBLEM] while I'm offline. Split the investigation into parallel workstreams where useful. Challenge obvious answers and verify important claims. By tomorrow, give me the recommendation, evidence, alternatives and what's still unknown.
Ongoing responsibility, proactive work and scheduled delivery in one job.
Own my weekly [SEO / sales / marketing / project] report. Use the connected sources I've approved. Keep the underlying information current through the week and prepare the final report every Friday. Flag unusual changes when they happen instead of waiting until Friday.
Final tip: do not judge your dot by asking it to write an email. Normal ChatGPT already does that. Give it something like "take responsibility for keeping this project moving, learn what is happening, do what you can without me, and come to me when you need a decision". That is when it clicks.
Tell us about your business and we will reply within 48 hours with what Karl would make for you. No sales call.