01
What are OpenAI Dots?
OpenAI dots are always-on AI agents inside ChatGPT. Unlike a normal chat, a dot can:
- Take responsibility for ongoing work
- Continue making progress between conversations
- Work on multiple tasks
- Use its own cloud computer and browser
- Work with connected apps
- Remember useful context
- Run scheduled and recurring tasks
- Research and solve multi-step problems
- Bring completed work back for review
- Ask you when human judgment is required
Dots are currently powered by GPT-6 Astra.
02
Create your business dot
Create a dot from ChatGPT on desktop web or the desktop app. Give it a clear role such as:
- Marketing Operator
- Sales Assistant
- Research Analyst
- Operations Assistant
- Customer Insights Agent
- Content Assistant
- Founder’s Chief of Staff
The narrower and clearer its responsibility, the easier it is to evaluate its work.
03
Give it business context
Explain:
- What your company does
- Products and services
- Target customers
- Business model
- Current goals
- Brand guidelines
- Team structure
- Important processes
- KPIs
- Constraints
- Tasks it should own
- Tasks requiring your approval
Treat this like onboarding a new employee.
04
Connect the right apps
Connect only the tools required for its responsibilities.
Depending on available integrations, this can give your dot access to business information from communication, productivity, development, email, calendar and other supported tools.
Always review permissions before connecting an app.
05
Define operating rules
Tell your dot exactly how independently it can operate. Use rules such as:
- Take action without asking
- Low-risk, reversible routine work.
- Take action if pre-approved
- Work you explicitly requested.
- Ask before taking action
- External communication, purchases, publishing or important changes.
- Hand off to me
- Strategic, sensitive, financial, legal or high-impact decisions.
06
Use it for market research
Ask your dot to continuously research:
- Competitors
- Customer needs
- Industry developments
- Product trends
- Market opportunities
- Pricing changes
- Emerging technologies
- Customer complaints
- Content trends
Instead of repeatedly starting research from scratch, give it an ongoing research responsibility.
07
Use it for marketing
Your dot can help:
- Research content ideas
- Analyze competitors
- Build content calendars
- Draft campaign concepts
- Repurpose existing content
- Research keywords
- Analyze campaign information
- Prepare marketing reports
- Organize creative assets
- Track relevant market developments
Keep publishing and other consequential actions behind approval when appropriate.
08
Use it for sales
Give your dot recurring sales-support work:
- Research prospects
- Prepare account briefs
- Summarize customer conversations
- Organize prospect information
- Prepare meeting briefs
- Draft follow-ups
- Research companies before calls
- Identify potential opportunities
- Prepare pipeline summaries
- Surface accounts requiring attention
Let AI prepare the information while humans handle relationship-critical decisions.
09
Use it for customer insights
Your dot can review connected customer information and help identify:
- Repeated complaints
- Common questions
- Feature requests
- Customer pain points
- Product confusion
- Churn signals
- Documentation gaps
- Emerging customer needs
Turn scattered feedback into structured business intelligence.
10
Use it for product development
A dot can support teams by:
- Watching customer feedback
- Finding recurring requests
- Scoping smaller improvements
- Investigating bugs
- Preparing implementation work
- Building and testing changes
- Preparing work for human review
OpenAI specifically demonstrates dots working on software improvements and bringing completed pull requests back for review.
11
Use it for operations
Delegate repetitive coordination work such as:
- Daily summaries
- Weekly reports
- Project tracking
- Information gathering
- Document preparation
- Meeting preparation
- Recurring checks
- Research
- Workflow coordination
- Follow-up tracking
The goal is not simply faster tasks. It is transferring appropriate ongoing responsibilities.
12
Use scheduled tasks
Give your dot recurring jobs such as:
- “Every morning, review my calendar and prepare my priorities.”
- “Every Friday, summarize progress across active projects.”
- “Every week, analyze customer feedback for recurring issues.”
- “Every month, prepare a business performance review.”
Scheduled tasks can continue without you manually restarting the workflow each time.
13
Build proactive workflows
Do not wait until you remember to ask.
Your dot can proactively review connected information, perform scheduled work, research in the background, and suggest useful next steps.
This makes dots especially useful for recurring responsibilities rather than isolated prompts.
14
Use it as a founder’s AI chief of staff
Give one dot responsibility for helping you stay on top of:
- Priorities
- Meetings
- Projects
- Research
- Follow-ups
- Deadlines
- Business updates
- Important decisions
- Outstanding tasks
Ask it to surface what requires your attention instead of flooding you with everything.
15
Delegate complex work
Do not only give it tiny tasks.
Instead of
“Summarize these reviews.”
Try
“Continuously review customer feedback, identify recurring problems, group them by severity and frequency, investigate possible causes, and prepare a weekly product improvement brief for me.”
Give it an outcome, context, constraints and decision boundaries.
16
Build a human review layer
Never assume autonomous means infallible. Review:
- Financial information
- Customer-facing communication
- Contracts
- Strategic decisions
- Purchases
- Production changes
- Sensitive information
- Important calculations
- High-impact actions
OpenAI explicitly warns that dots can make mistakes, including when following custom rules.
17
The best dot workflow
- Goal
- Define the business outcome.
- Context
- Give the dot the information required to understand the business.
- Access
- Connect only necessary apps and systems.
- Rules
- Define what it may do independently.
- Execution
- Let it complete appropriate multi-step work.
- Schedule
- Automate recurring responsibilities.
- Review
- Inspect important outputs and actions.
- Feedback
- Correct mistakes and refine instructions.
- Expand
- Gradually delegate more responsibility after the workflow proves reliable.
18
A simple dot prompt framework
- Role
- “You are my [business role].”
- Objective
- “Your primary objective is [measurable outcome].”
- Responsibilities
- “You are responsible for [tasks].”
- Context
- “Here is how our business works: [context].”
- Tools
- “Use these connected sources when relevant: [tools].”
- Rules
- “Always follow these constraints: [rules].”
- Approvals
- “Ask me before [sensitive actions].”
- Cadence
- “Perform this work [daily/weekly/monthly].”
- Output
- “Report results using [desired format].”
- Escalation
- “Bring something to me immediately when [conditions].”
19
Best business use cases
- Marketing
- Research, campaigns, content, competitive intelligence
- Sales
- Prospect research, meeting preparation, follow-ups, pipeline support
- Customer success
- Feedback analysis, issue discovery, account preparation
- Product
- Customer insights, feature research, bug investigation, development support
- Operations
- Reporting, coordination, recurring workflows
- Finance
- Information organization, recurring analysis, reporting preparation
- HR
- Research, documentation, coordination, recurring administrative work
- Leadership
- Research, priorities, project oversight, decision preparation
20
The 3-level dot system
Level 1 · Assist
Dot researches, analyzes, drafts and recommends. Human executes.
Level 2 · Operate with approval
Dot completes workflows but asks before consequential actions. Human reviews and approves.
Level 3 · Operate autonomously
Dot independently handles predefined, low-risk workflows within strict rules. Human monitors results and exceptions.
Start with Level 1 and expand autonomy only after the workflow proves reliable.
21
What not to do
- Do not give unrestricted access unnecessarily
- Do not automate poorly understood processes
- Do not connect sensitive systems without reviewing permissions
- Do not let it make high-impact decisions without appropriate oversight
- Do not assume every output is correct
- Do not create vague responsibilities
- Do not automate a broken workflow
- Do not ignore security and privacy requirements
- Do not confuse activity with business impact
22
Start small, then expand
The biggest shift with OpenAI dots is moving from AI that waits for individual prompts to an agent that can keep working toward defined goals.
- Start with one repetitive, measurable, low-risk workflow
- Define the outcome
- Connect only the information it needs
- Set strict boundaries
- Schedule recurring work
- Review its performance
- Then gradually expand what it owns
The real opportunity
Do not ask: “What tasks can AI do for me?” Ask: “What ongoing business responsibility can I safely delegate?”
- Outcome
- Context
- Boundaries
- Schedule
- Review
- Expand
Start with one repetitive, measurable, low-risk workflow, then gradually expand what your dot owns.
