Meet dot from OpenAI: a personal assistant that keeps work moving between conversations, within the permissions you give it. Start with a clear goal, better instructions, and thoughtful control.
Less complexity.
A clearer starting point.
Give dot a goal, context, and boundaries, instead of starting from scratch each time. It can keep working and return with results or questions, asking for approval when needed.
OpenAI says dot is powered by GPT-6 Astra and has a cloud computer and browser. You can reach it through ChatGPT, voice, and Slack, depending on availability and setup.
Read the sourceYou open a conversation and ask what to do next.
It may share a useful update within a clearly defined responsibility.
The focus is on a reply in the current session.
It can continue a task between conversations and check back with you.
Many of the next actions remain with you.
It uses approved tools, with controls for approval.
An editorial comparison to illustrate a way of working, not a performance test or a complete account of other chat products’ capabilities.
Five suggested steps to turn an experiment into a workflow that fits you.
Official getting-started guideStart in the desktop or web app supported for your account. Availability depends on your plan, region, age, organization settings, and the rollout.
Check access requirementsBegin with one recurring task, such as developing content ideas or reviewing priorities. Explain the outcome you want and what would make it useful to you.
Choose the apps and sources the task requires. More data access is not a goal in itself. Start with the minimum and expand when needed.
Review the controls and Custom rules in your account. Make it clear when your assistant should prepare work and when it should ask you before sending or taking action.
Review accuracy, style, and timing. Share an example of the result you want, and refine the scope before adding more responsibilities.
Copy a prompt, add your context, and try it on a specific task. Every example starts with preparation and review.
Turn one topic into five angles worth developing.
Act as my content assistant. First, ask about my topic, audience, and platform, then develop 5 distinct ideas. For each idea, include an opening hook, a content outline, and a suggested caption. Flag anything that needs source verification, and do not invent figures or results. Prepare drafts for review only; do not publish or send anything.
Clear priorities, with replies ready to review.
Review the messages I give you access to and identify what needs action from me, why it matters, and any deadline. Rank them by actual importance and link each point to its original message. Prepare short draft replies and flag commitments or sensitive information. Do not send a reply, delete a message, or edit one without my approval.
Client context, better questions, and a clear next step.
Help me prepare for a client meeting. Ask about the client and the meeting goal, then use approved sources to summarize the context, needs, and open questions. Prepare a proposed presentation outline and a follow-up draft, separating facts from assumptions. Flag any price, promise, or commitment that needs my decision. Do not send a proposal or confirm a commitment on my behalf.
A short brief to help you choose your next step.
Help me organize my day using only the sources I approve. Choose the 3 most important things that need my attention, explain why they matter now, and give the next step and any confirmed deadline. Ask how much time I have if you do not know. Do not assume every message is urgent, change appointments, or contact others without my approval.
Two simple tools to prepare your instructions and estimate your workload. No account connection or sign-in required.
Choose a role, add your goal, and get a clearer starting point.
This is an instruction prompt. Your actual approval controls are set in your account.
Act as my content creation assistant. Suggest angles that fit the audience, then prepare drafts with a hook, an outline, and a caption. Flag any claim that needs a source. My goal: Ask me what outcome I want before getting started. Use only the sources I have approved, and separate verified information from assumptions. Ask me for any missing essential information. Prepare the results here as drafts for review only. Do not send, publish, delete, buy, change an appointment, or share my data with another party without my approval. Check with me before making any commitment or taking a sensitive action.
The result is generated on this page. This tool does not send your input to dot or an external service.
A simple planning estimate. Adjust the numbers to match your experience.
A possible time difference calculated from your inputs, not a guaranteed result.
(Manual time − your time with dot) × tasks ÷ 60
The default numbers are illustrative only. This is not a performance study or a measure of actual savings, and it does not account for cost or output quality.
Good delegation defines both the outcome and the boundaries. Review your account settings before connecting sources or expanding your assistant’s responsibilities.
Research, summaries, comparisons, and organized drafts from approved sources.
Check accuracy, context, dates, names, and promises before accepting a result.
App access, the scope of work, and what needs your approval or direct involvement.
Writing “ask me before sending” makes your preference clear, but does not automatically create a rule in your settings. Some sensitive or high-impact actions may require additional approval or your direct involvement, regardless of the prompt.
Read about controlsNot necessarily. Availability varies by plan, region, age, organization settings, and rollout. Check the official access information for current requirements.
Start with only the apps the task needs. Review the permissions you grant, and use the computers and apps guide to understand connection options. Do not grant access just because it is available.
No. Copying only places the text on your clipboard. Paste it into your conversation with dot, adjust the context, and review permissions before starting. The prompt builder on this page does not connect to your account.
No. It is the result of a formula based on your numbers. The result may be zero or negative. Measure the actual time spent prompting, reviewing, and revising, and assess quality before deciding how useful it is.
This is an independently written guide, not an automatically updated bulletin. Features and access requirements may change. The official documentation is the most up-to-date reference.