Hi, I’m Keshav Chouhan. I spend a lot of my time testing AI prompts and looking at a simple question, why does one instruction give a useful result while another gives something generic?
Most of my prompt testing has involved ChatGPT, Gemini, AI image generation, content work, and everyday digital marketing tasks.
When OpenAI introduced dots on September 29, 2026, I noticed that writing a good dot prompt requires a slightly different approach.
A normal ChatGPT prompt usually asks for something now.
For example:
Give me five content ideas for Instagram.
A dot can take responsibility for something that continues after that first conversation.
For example:
Every Friday, check these three competitor websites, find what they published this week, compare those topics with my website, and show me three useful content gaps.
That changes how I think about the prompt.
I am no longer asking only:
What output do I want?
I also need to decide:
How often should the task run? What information should it use? When should it contact me? What can it do without asking me? When should it stop?
In this guide, I’ll explain how I would structure a dot prompt and share 7 ChatGPT dots prompts you can customize for content research, email, learning, price tracking, competitor research, and lead follow-ups.
What Is a ChatGPT Dot?
A ChatGPT dot is an always-on AI agent designed to handle ongoing responsibilities.
OpenAI introduced dots at DevDay 2026. They are powered by GPT-6 Astra, have their own cloud computer, can work with connected apps, remember useful context, and continue making progress between conversations. OpenAI
This is different from a normal ChatGPT conversation.
If I ask regular ChatGPT:
Research my competitors.
I am mainly asking for a task now.
With a dot, I could instead say:
Check these competitors every Friday and tell me when you find a useful content gap.
The second instruction describes an ongoing responsibility.
That is the main idea I would understand before copying any of the prompts below.
ChatGPT Dot vs Normal ChatGPT Prompt
| Normal ChatGPT Prompt | ChatGPT Dot Prompt |
| Usually asks for an answer now | Can define ongoing work |
| Often focuses on one output | Can continue between conversations |
| “Research these competitors” | “Check these competitors every Friday” |
| “Summarize my schedule” | “Summarize my schedule every weekday morning” |
| Mainly defines the requested output | Also needs timing, permissions and reporting rules |
| May not need an end date | Recurring work may need a stopping condition |
I would not simply take every old ChatGPT prompt and give it to a dot.
For ongoing work, the prompt needs to explain the responsibility more clearly.
Who Can Use ChatGPT Dots?
At the time I updated this article, OpenAI says dots are gradually rolling out to ChatGPT Pro users in eligible markets outside the European Economic Area, Switzerland and the UK.
Dots are also available to Business Premium users in supported ChatGPT regions.
Enterprise users can try the beta when their workspace administrator enables it. OpenAI’s DevDay announcement also identifies Edu and Healthcare users as eligible for the Enterprise beta when enabled by their workspace administrator. OpenAI Help Center
Because the rollout is gradual, an eligible account may not receive the feature immediately.
OpenAI also says usage will not count toward eligible Pro, Business and Enterprise plan allowances for the first month of rollout. Usage terms after that introductory period may change, so I would check OpenAI’s current help page rather than relying on an old article for future limits. OpenAI Help Center
How to Create Your First ChatGPT Dot
At the moment, you create a dot using the ChatGPT desktop app or ChatGPT on desktop web.
You cannot currently create a dot from the mobile app, and dots are not supported on mobile web. After creating one, you can talk to it through the ChatGPT mobile app when mobile access is available. OpenAI Help Center
During setup, you can give your dot a name and choose its appearance.
Your dot already has its own cloud computer.
Giving it access to your local computer is optional and starts turned off. I would personally leave local access disabled unless the task genuinely needs files or applications from my computer. OpenAI Help Center
You can also connect supported apps.
OpenAI says dots can work with its plugin ecosystem of more than 4,000 apps. OpenAI
Before connecting email, files, Slack, or another account, I would read the requested permissions rather than connecting everything simply because it is available.
The 5 Things I Add to a Good Dot Prompt
After looking at how dots work, I would structure their prompts differently from many of my one-time ChatGPT prompts.
These are the five things I would think about first.
1. Goal
What responsibility are you actually giving the dot?
A weak instruction would be:
Help me with my blog.
That is too open.
A clearer instruction would be:
Check these three competitor blogs every Friday and identify useful topics they published that my website has not covered.
Now the dot knows what job it owns.
2. Context
Give the dot information that changes its decisions.
If I am asking for content ideas, I might tell it:
- My website topic
- My audience
- My target country
- What I already publish
- What type of content I do not want
Without that information, the AI has to make more assumptions.
3. Sources
Tell it where it should look.
Instead of:
Find AI trends.
I could say:
Check these specific sources and give me only trends relevant to my audience.
For research tasks, source selection can matter as much as the wording of the final report.
4. Boundaries
This becomes much more important when an AI can work on an ongoing task.
Reading information is one thing.
Sending an email, publishing a post, changing a file, or completing a purchase is another.
I want the prompt to make it clear what the dot can do and what should come back to me for approval.
5. Check-ins and Stopping Conditions
I also tell the dot when I actually want to hear from it.
For temporary work, I add an end point.
For example:
Run this every Sunday until November 8, then stop.
Without a useful schedule or stopping condition, an ongoing task can continue longer than you intended.
With those five parts in mind, here are the prompts.
1. Daily Trend Watcher for Content Creators
I would use this when I want to monitor the same sources regularly without manually searching them every morning.
Copy This Prompt
I run an Instagram page about [TOPIC] for [AUDIENCE AND LOCATION].
Every morning at 8 AM IST, research current trends related to [TOPICS].
Check [SOURCES YOU WANT USED].
Send me the five most relevant ideas.
For each idea, include:
- Trend or topic
- Source link
- Why it appears relevant now
- One content idea for my page
Compare the suggestions with topics I have covered during the previous 14 days and avoid obvious repeats.
Do not publish, comment, message anyone, or change anything on my accounts. Research and report only.
If there is nothing genuinely useful that day, tell me that instead of filling the report with weak ideas.
Why I Wrote It This Way
The last instruction is one I would keep.
If I tell an AI:
Give me five ideas every day.
I have already created pressure for it to return five answers, even on a day when there may not be five useful developments.
For an ongoing research task, I would rather receive:
Nothing important enough to act on today.
than five filler ideas.
That small change makes the report more useful to me.
2. Festival Content Calendar Dot Prompt
I would use this type of prompt for a campaign that has a clear beginning and end.
Copy This Prompt
Help me manage content for [FESTIVAL OR EVENT] for my website [URL] and social media account [ACCOUNT].
My audience is [AUDIENCE].
Build a content calendar from [START DATE] to [END DATE] with three planned posts each week.
For every content idea include:
• Working title
• Main topic or keyword
• Content format
• Intended reader
• Suggested publishing date
• Reason for choosing that date
Before suggesting a new article, check the existing content I give you and avoid near-duplicate topics.
Every Sunday at 7 PM IST, send me the coming week’s plan and mention any important new development that could justify changing it.
Do not publish or edit my website without my approval.
Stop this recurring work after [END DATE].
Why I Added an End Date
Festival campaigns are temporary.
If I create a Diwali campaign, I do not want the same recurring task quietly continuing weeks after Diwali.
The stopping condition makes the responsibility clearer.
I also included duplicate checking because an ongoing content planner can easily suggest slightly different versions of topics already published.
3. Morning Email and Calendar Summary
This is an example where I care more about permissions and filtering than clever prompt wording.
Connect only the email and calendar sources the task actually needs.
Copy This Prompt
Every weekday at 7:30 AM IST, review my connected calendar and email.
Send me one morning brief containing:
• Today’s meetings in chronological order
• Emails that appear to need a reply today
• Messages received during the last 24 hours from [IMPORTANT PEOPLE OR CLIENTS]
• Any deadline you can identify for today
Keep the complete brief under 150 words.
Do not send, delete, archive, or forward any email.
If I ask for help replying to something, prepare a draft for my review.
Never send the draft unless I explicitly approve the action.
If you are unsure whether an email needs action, mark it for my review instead of guessing.
Why I Added a 150-Word Limit
Without a limit, a morning summary can become another inbox.
That defeats the purpose.
I want this task to reduce information and show me what deserves attention.
The word limit forces the report to prioritize.
4. Weekly Competitor Blog Tracker
As someone working with websites and SEO, this is one of the dot use cases I find most interesting.
But I would not ask an AI simply to copy whatever competitors publish.
Copy This Prompt
Every Friday, review these competitor websites:
[COMPETITOR URL 1]
[COMPETITOR URL 2]
[COMPETITOR URL 3]
Find relevant new articles they published during the previous seven days.
For each article, give me:
• Article title
• URL
• Main topic
• Two-sentence summary
Then compare those topics with my website:
[MY WEBSITE URL]
Suggest three useful topics the competitors covered that my website has not covered well.
Do not suggest a topic only because a competitor published it.
For every recommendation, explain why the topic could be useful to my audience and what original value my article could add.
Do not contact competitors, subscribe to newsletters, leave comments, or create accounts.
If there is no meaningful content gap that week, say so instead of forcing three recommendations.
The Line I Would Not Remove
This one:
Do not suggest a topic only because a competitor published it.
Competitor research can easily become competitor copying.
That is not what I want.
I want to know whether another website has revealed a useful gap for my readers.
If the topic is irrelevant to my audience, I do not care that three competitors wrote about it.
This is also how I would use AI for SEO research. AI should help identify opportunities, not become a machine for cloning another publisher’s content plan.
5. Product Price Watch Dot Prompt
This example shows why a recurring prompt needs a clear alert condition.
Copy This Prompt
I want to buy [EXACT PRODUCT AND MODEL].
My maximum final price is [AMOUNT].
Once per day, check prices from:
[STORE 1]
[STORE 2]
[STORE 3]
Notify me only if:
• The normal price falls below my target, or
• A clearly available offer brings the final price below my target.
When you notify me, include:
• Store
• Exact product and variant
• Listed price
• Relevant offer conditions
• Final calculated price
• Direct product link
Make sure it is the exact model, storage size, configuration, color, or other variant I requested where those details matter.
Do not purchase the product, enter payment information, or complete checkout.
Hand the purchase decision back to me.
Why “Exact Model” Matters
A price alert is useless if the AI compares the wrong products.
For example, it could compare:
- 128 GB with 256 GB
- New with refurbished
- Base laptop with upgraded RAM
- One product generation with another
The exact model and configuration are part of the success condition.
They are not minor details.
I also keep the final buying decision with myself.
6. Four-Week Learning Companion
A dot does not have to be used only for monitoring prices, emails, or websites.
Ongoing context can also be useful for learning.
Copy This Prompt
I want to learn [TOPIC] over the next four weeks.
My current level is [BEGINNER / INTERMEDIATE / ADVANCED].
My goal is [WHAT I WANT TO BE ABLE TO DO].
Every Monday, send me a learning brief containing:
• Three to five concepts to learn that week
• Two reliable sources to study
• One practical exercise
• Three questions to test my understanding
Build each week on what I learned previously.
At the end of each week, ask me what I found difficult or unclear.
Use my answer to adjust the following week’s material.
Do not assume I understood something simply because you explained it earlier.
Stop after week four and give me a final review covering what I learned and the areas I should continue studying.
Why This Works Better as an Ongoing Task
The interesting part is not:
Teach me [TOPIC].
Regular ChatGPT can already help with that.
The useful part is:
Use what I struggled with this week to change what you teach me next week.
That creates continuity.
Instead of receiving four disconnected lessons, the learning plan can respond to my progress.
7. Small Business Lead Follow-Up Assistant
This is another task where I would be careful about giving an AI too much freedom.
I want help organizing enquiries, but I do not necessarily want an agent speaking to customers without me.
Copy This Prompt
I run a [BUSINESS TYPE] in [CITY].
When a new customer enquiry appears in my connected email, record:
• Name
• Date received
• Contact information provided
• Product or service requested
• Quantity, if mentioned
• Budget, if mentioned
• Required date, if mentioned
• Current follow-up status
If I have not replied within 24 hours, remind me.
You may prepare a short reply draft using information available in the enquiry, but clearly mark anything you are unsure about.
Never invent a price, delivery promise, stock status, product specification, or business policy.
Never send the customer a message without my approval.
Every Saturday evening, give me a summary of:
• New enquiries
• Unanswered enquiries
• Leads waiting for follow-up
If information is missing from an enquiry, mark it as missing rather than guessing.
Why I Added “Never Invent”
This comes from a practical problem with AI-generated customer communication.
A reply can sound professional and still contain incorrect information.
If an AI does not know my current price, stock level, delivery date, or business policy, I would rather receive:
Price needs confirmation.
than a confident but invented answer.
For business tasks, knowing when not to guess is part of a good prompt.
Custom Rules I Would Set Before Using a Dot
A good prompt is only part of the setup.
OpenAI also provides custom rules for actions a dot can take.
The currently documented options include:
- Take action without asking
- Take action if pre-approved
- Ask before taking action
- Hand off to you
OpenAI explains that “pre-approved” means you explicitly requested the action in your prompt. OpenAI Help Center
I would not give every action the same permission.
Here is how I would personally start:
| Action | My Starting Preference |
| Research public information | Allow |
| Read connected information required for an approved task | Allow where needed |
| Prepare a draft | Allow |
| Send an email or message | Ask me first |
| Publish content | Ask me first |
| Make a purchase | Hand off to me |
| Delete important information | Keep human review |
| Change sensitive account information | Keep human review |
These are my suggested starting preferences, not rules required by OpenAI.
The right setup depends on the job and what you are comfortable allowing an AI agent to do.
OpenAI also warns that dots can make mistakes, including when following custom rules. Important information and actions should still be reviewed. OpenAI Help Center
Privacy Details I Would Check Before Connecting Apps
This is something I would not skip just to get to the prompts faster.
According to OpenAI’s current documentation, connected apps give a dot information it can use for tasks.
A dot can also proactively review information from connected apps and form memories from it, even when you have not asked a specific question about that information. OpenAI Help Center
That means I would not connect every available app automatically.
Before connecting something, I would ask:
Does this dot actually need this information to perform its job?
A competitor research dot probably does not need my personal email.
A learning companion probably does not need access to customer files.
A morning inbox assistant does need email access if reviewing email is its job.
Give the dot the access required for the task, rather than access simply because it is available.
There is another important point.
Disconnecting an app does not automatically delete information the dot has already obtained from it. OpenAI says deleting/resetting the dot is required to delete the dot’s own stored information. OpenAI Help Center
That is something I would want to know before connecting a sensitive source.
5 Mistakes I Would Avoid With ChatGPT Dot Prompts
1. Giving One Dot Too Many Unrelated Jobs
I would avoid a prompt like:
Manage my email, grow Instagram, research competitors, track laptop prices, organize my meetings, and teach me SEO.
Technically, an agent may be capable of doing many things.
That does not mean putting everything into one vague responsibility is a good idea.
I prefer a clear goal with clear reporting rules.
2. Forgetting When It Should Contact You
For monitoring tasks, define what deserves an alert.
If I am monitoring a ₹50,000 product and only care when it falls below ₹45,000, I do not need:
Price unchanged today.
every morning.
I want a message when the condition I care about becomes true.
3. Forgetting the Stopping Condition
Some responsibilities should continue indefinitely.
Others should not.
A festival campaign, four-week learning plan, temporary project, or product search may need an end date or completion condition.
Define it.
4. Connecting Apps the Task Does Not Need
More access does not automatically create a better agent.
If the task only needs public websites, I would not connect private files or email without a reason.
Keep the responsibility and access aligned.
5. Treating Background Work as Automatically Correct
An agent continuing to work while I am away does not mean its work becomes automatically correct.
I would still check:
- Important numbers
- Links
- Prices
- Customer information
- Business decisions
- External actions
- Anything that could have a meaningful consequence
Automation can save time, but review still matters.
How to Schedule Work With a Dot
You can ask a dot to set reminders or run recurring checks.
For example:
Every weekday morning, review my calendar.
or:
Check this product price once per day.
Scheduled and proactive updates appear in your conversation.
OpenAI says scheduled tasks can be reviewed from Scheduled in the dot’s profile, where you can manage active, paused, and completed tasks and change their timing. OpenAI Help Center
This is why I include the schedule directly in many of the prompts above.
“Track competitors” is vague.
“Every Friday, review these three competitors and report new articles from the previous seven days” gives the task a rhythm.
How to Pause a ChatGPT Dot
If I am travelling, taking a break, or simply do not want ongoing work running for a while, I do not necessarily need to delete the dot.
Open the dot’s profile, select the ••• menu, then choose Pause.
You can resume it later. OpenAI Help Center
This is useful for recurring work you want to keep but temporarily stop.
How to Reset a ChatGPT Dot
Reset is different from Pause.
According to OpenAI, resetting deletes the dot, including its:
- Conversations
- Saved memories
- Scheduled tasks
After resetting, you return to a new ChatGPT conversation and can create another dot later. OpenAI Help Center
So I would not use Reset when I simply want a quiet week.
For that, Pause is the better option.
Can a Dot Use Your Computer?
Yes, but local computer access is optional.
A dot already has its own cloud computer.
OpenAI says access to your local computer starts turned off. If you decide to connect a computer through the ChatGPT desktop app, the dot can access files and work on that connected computer after you allow access. OpenAI Help Center
For my own use, I would follow a simple rule:
If the task works without local computer access, I would leave it off.
I would enable additional access because the task requires it, not because the option exists.
Can You Text a ChatGPT Dot?
OpenAI currently offers texting as a limited beta where available.
At the time of this update, the texting beta is limited to Pro users in the United States, is not available in Business or Enterprise workspaces, and may not be available to every eligible Pro user. OpenAI Help Center
This is another detail I expect could change as dots develop.
If you are reading this article months after publication, check OpenAI’s current documentation for availability rather than assuming the launch restrictions are still the same.
FAQ
Are ChatGPT dots free?
Dots are not currently described by OpenAI as a general Free-plan feature. They are rolling out to eligible Pro and Business Premium users, while Enterprise users can access the beta when enabled by their workspace administrator.
Because plans and usage terms can change, check OpenAI’s current documentation before upgrading specifically for this feature.
Can I use ChatGPT dots in India?
OpenAI currently says Pro access is rolling out in eligible markets excluding the European Economic Area, Switzerland, and the UK. India is not listed among those launch exclusions. Access is gradual, so an eligible account may still need to wait for the rollout.
Can I create a ChatGPT dot on my phone?
Not currently.
OpenAI says dots are created using ChatGPT on desktop web or the ChatGPT desktop app.
After creation, you can talk to your dot using the mobile app when mobile access is available. Dots are currently not supported on mobile web.
Can a dot keep working when my computer is off?
A dot has its own cloud computer and is designed for ongoing work between conversations. It can also run recurring tasks you schedule.
You therefore do not need to keep your personal computer running for work being done on the dot’s cloud computer.
Does a dot need access to my computer?
No.
Local computer access is optional and starts turned off. Your dot already has its own cloud computer.
Only enable local access when the work actually requires it.
Can a ChatGPT dot send messages or take actions?
Dots can work with connected tools, and OpenAI provides confirmation and custom-rule controls for certain actions.
Depending on the action, you can use rules such as asking before taking action or handing the action back to you.
I would use stricter approval rules for actions such as sending messages, publishing information, purchasing something, or changing important data.
How do I delete everything my dot knows?
OpenAI says Reset deletes the dot, including its conversations, saved memories, and scheduled tasks.
Simply disconnecting an app does not delete information the dot already obtained from that app.
How many apps can dots connect to?
OpenAI says its plugin ecosystem gives dots access to more than 4,000 apps. The apps available to you can depend on your account and environment.
Which ChatGPT Dot Prompt Should You Try First?
I would not start by giving my first dot control over something important.
I would begin with a research-focused or read-only responsibility.
For example:
Weekly Competitor Blog Tracker
or:
Four-Week Learning Companion
These let you see how the dot works without immediately giving it responsibility for sending messages, publishing content, or buying something.
After a few runs, I would look at its actual behavior.
Did it send too much information?
Did it misunderstand what counts as important?
Did it repeat topics?
Did it need another source?
Did it contact me too frequently?
Was one part of my instruction unclear?
Then I would change the prompt based on those problems.
That is one of the most useful things I have learned from testing AI prompts in general.
The best version of a prompt is often not the first version you write. It is the version you reach after seeing how the AI interpreted your first instruction.
With dots, that matters even more.
You are not simply asking ChatGPT for one answer.
You are defining a responsibility that may continue after you close the conversation.
So before giving a dot a long prompt, I would make sure five things are clear:
What is the goal? What information does it need? What sources should it use? What can it do without me? When should it report back or stop?
Get those right first.
The wording can be improved later.
Sources and Further Reading
I used OpenAI’s current documentation to verify the product details in this guide. Since dots are new and still rolling out, I recommend checking these official pages if you are reading this article later and need the latest availability, plan, privacy, or setup information:
