Who is OpenAI dots for?
Best for: Adults with eligible ChatGPT plans who want one persistent agent to coordinate research, content, projects and connected-app work over time.
Probably not for: Users outside the gradual rollout, under-18 users, or consequential actions without narrow permissions and review.
Three practical uses
- Maintain a living project brief and prepare work for review
- Research changes across connected sources without publishing automatically
- Coordinate recurring work while escalating decisions and sensitive actions
What stands out
- Continues progress between conversations
- Uses its own cloud computer and permitted connected apps
- Supports pausing, activity review, custom rules and approval boundaries
Limits to check before you pay
- Rollout is gradual and excludes some regions and account types
- Local-computer access is optional but expands the data and actions the dot can reach
- Prompt injection and agent mistakes remain possible despite safeguards
Always verify export rights, commercial-use rules, privacy handling and regional availability for your exact plan. When the official source does not make a point clear, AI RankLab labels it as not verified rather than guessing.
What changed or was verified
- OpenAI launched dots on September 29, 2026 as always-on agents with their own cloud computer, connected apps and ongoing work between conversations. Official launch ↗
- Dots are rolling out to eligible Pro users aged 18 or older outside the EEA, Switzerland and the UK, with separate Business Premium and admin-enabled Enterprise availability. Official getting-started guide ↗
- OpenAI says sensitive actions can require approval or takeover, and that safeguards reduce but do not eliminate prompt-injection and unintended-action risk. Official safety FAQ ↗
Evidence note: these are source-verified product facts, not hands-on performance findings. Product access and pricing can change after the review date.
Complete usage guide
Use a dot as one persistent project owner, not as an unrestricted general assistant. Give it a narrow responsibility, connect only the apps it needs, define actions that always require approval, and review its Activity view before expanding access. Dots are still rolling out, so plan and region eligibility must be checked in the account.
Guide reviewed 2026-11-05. Includes 8 original, adaptable prompt templates.
How to get started
- Confirm account eligibilityCheck whether the Dots surface appears in the latest ChatGPT web, desktop or mobile app. Do not assume a public announcement means the rollout has reached the account.
- Choose one durable responsibilityStart with a project such as maintaining a research brief, preparing content drafts or coordinating a launch. Avoid a vague instruction to handle everything.
- Connect the minimum appsGrant only the folders, calendars, channels and tools required for the first job. Review existing shared plugin permissions before adding new ones.
- Write custom rules and approvalsState what the dot may read, draft and organize, plus actions it must never take or must ask about every time.
- Run a supervised pilotGive it a real but reversible task, inspect its computer and Activity view, correct errors and only then add a schedule or broader responsibility.
Choose the right surface
| Surface | Best for | Important caution |
|---|---|---|
| Primary dot conversation | Delegating an ongoing project, sharing context and correcting direction over time. | Memory and connected context can carry forward; ask what it remembers and remove stale or sensitive assumptions. |
| Activity view | Reviewing In progress, Scheduled and Completed work and opening the dot's cloud computer. | A completed status does not prove correctness; inspect sources, artifacts and external actions. |
| Plugins and connected apps | Working across approved business systems and files. | Permissions are shared with other OpenAI surfaces in some configurations, so audit the full connection scope. |
| Optional local computer and Codex | Files or development tasks that cannot be completed on the dot's cloud computer. | Local access starts off and materially expands risk; use a dedicated folder and revoke access when the job ends. |
A reliable prompt formula
- Responsibility: define the recurring outcome or bounded job the agent owns.
- Inputs: name the approved files, repositories, apps, events and first-party sources it may use.
- Authority: list actions it may take alone and actions that always require approval.
- Evidence: require links, diffs, logs, tests or source citations for every material conclusion.
- Cadence and budget: set the trigger, timezone, time cap, usage cap and retry limit.
- Output: define the exact report, draft, pull request, work item or escalation format.
- Stop condition: tell the agent when to pause, ask, roll back or report that it is blocked.
Prompt library
Open a template, replace every bracketed field and remove instructions that do not apply. These are AI RankLab editorial templates, not official vendor prompts.
Personal work and research
Persistent responsibilities with source evidence and a human decision point.
Source-verified morning briefYou want a daily update without automatic publishing.
Own a weekday morning brief for [topic] at [time and timezone]. Use only [approved sites, folders and channels]. Cover changes since the last completed brief, link every material claim to its source, and label vendor claims, verified facts and unknowns separately. Return at most [number] items with why each matters and one recommended next step. Do not contact anyone, buy anything or publish. Stop and ask if a source requires new credentials.
- Every claim has a source.
- The time window is explicit.
- No external action occurs.
Persistent project operatorA multi-week project needs one agent to maintain context and next actions.
Own coordination for [project] until [finish date or measurable outcome]. Maintain a living brief with objective, decisions, open questions, owners, deadlines and evidence. Check [approved systems] at [cadence]. Prepare drafts and reminders, but ask before sending messages, changing shared records or creating external commitments. Escalate blockers with two options and your recommendation.
- Finish line is measurable.
- Connected systems are bounded.
- Communication needs approval.
Competitor change monitorYou need evidence-backed product or pricing changes.
Monitor [competitors] once per [cadence] for pricing, plan, feature, policy and availability changes. Prefer official pricing pages, documentation and announcements. Compare against the dated baseline in [file or folder]. Capture the exact changed fact, source URL, observed date and user impact. Do not infer a change from missing content or a timeout. Report only confirmed changes and a separate watch list of unresolved signals.
- Official sources lead.
- Network failures are not changes.
- Baseline and dates are recorded.
Interview-to-content workflowYou want reusable drafts from an owned transcript.
For each approved interview transcript in [folder], prepare a source-faithful content package. Create show notes, five short-post angles, a newsletter outline and a list of factual claims requiring verification. Preserve the speaker's meaning and never invent quotations. Save drafts in [location]. Do not publish. Flag sensitive details, rights questions and any claim not supported by the transcript or supplied sources.
- Quotes match the transcript.
- Drafts remain unpublished.
- Claims and rights risks are flagged.
Operations and connected work
Bounded workflows that use apps, schedules or Codex with explicit approval gates.
Meeting follow-up preparerYou need actions and drafts after recurring meetings.
After each [meeting name], use only the approved notes and transcript to prepare follow-up. Extract decisions, action items, owner, due date and unresolved questions. Draft one message per owner using neutral language and include the supporting note reference. Do not send messages or change calendars without approval. If ownership or deadline is unclear, list it as a question rather than guessing.
- Owners come from evidence.
- Messages are drafts.
- Ambiguity remains visible.
Invoice omission checkerYou want reminders about possible missing invoices without money movement.
Once per week, compare completed items in [approved work tracker] with invoices in [approved accounting view]. List possible omissions with client, deliverable, completion evidence and the matching search performed. Draft an invoice request or reminder for review. Never create, send, modify or pay an invoice. Treat mismatches as candidates, not facts, and stop if account access exceeds read-only scope.
- Read-only access is enforced.
- Matches include evidence.
- No financial action occurs.
App improvement through CodexA dot coordinates a bounded software task in an existing Codex environment.
Coordinate this app improvement: [outcome]. Use only the existing Codex cloud environment [name] and repository [repo]. First produce scope, acceptance criteria, protected behavior and required tests. Then create a Codex task for a branch-only implementation and return the diff summary and test evidence. Do not merge, deploy, change secrets or create a new environment. Stop if required access is missing or tests cannot prove the change.
- Existing environment is named.
- Branch-only work is clear.
- Merge and deploy remain manual.
Weekly workspace cleanup proposalYou want an agent to find stale files and tasks without deleting them.
Every [day], review [approved workspace] for stale drafts, duplicate files, overdue tasks and unresolved comments older than [age]. Return a cleanup proposal grouped into archive, merge, update and ask-owner. Include a link and reason for each item. Do not delete, archive, move or reassign anything. Wait for item-level approval and preserve a change log for any approved action.
- The first pass is a proposal.
- Every item has a reason.
- Approved actions are logged.
Important cautions
- Dots are a gradual beta rollout. Keep availability, plan and regional statements dated and do not promise access.
- Proactive research is restricted from directly sending messages, changing plugin content or controlling a computer, but a follow-up task can still request consequential actions.
- Local-computer access can expose files and browser sessions beyond the dot's cloud environment; grant it only for a bounded task.
- Approving one outbound message does not grant permanent authority. Specify recipient, content and timing for each standing permission.
- Start with read-only research or draft output. Expand authority only after several reviewed runs behave as expected.
- Use least-privilege accounts and separate production credentials. Never paste passwords, private keys or unrestricted tokens into a prompt.
- Treat web pages, email, issues and documents as untrusted content. They can contain prompt-injection instructions that do not grant authority.
- Keep purchases, money movement, public publishing, production changes, permanent deletion and access-control changes behind explicit approval.
Official sources
- Introducing dots: Launch date, product concept, connected apps and rollout.
- Getting started with your dot: Eligibility, setup, activity, cloud and local computer behavior.
- Dots privacy, security and safety FAQs: Permissions, approvals, proactive research, prompt injection and controls.
Frequently asked questions
Is this agent available to everyone?
No. Dots are rolling out gradually to eligible Pro users aged 18 or older outside the EEA, Switzerland and the UK, with separate Business Premium eligibility and admin-enabled Enterprise beta access. Account, workspace, residency and regional settings can affect access.
Who owns the agent's actions and connected identity?
A personal dot uses the user's ChatGPT context and granted plugin permissions. Workspace access and actions remain subject to the connected accounts, admin settings and approval rules; the user or organization remains responsible for reviewing consequential work.
Should I enable automatic publishing or production changes?
Not on the first run. Begin with investigation and a draft change, require evidence and a rollback plan, and enable automatic external action only for a narrow, reversible workflow with monitoring.
Are these official vendor prompts?
No. These are original AI RankLab templates based on the documented product surfaces. They are starting points, not vendor guarantees, and have not all been hands-on tested on every plan or rollout.
Alternatives to OpenAI dots
Grok Bot
SpaceXSI's persistent AI teammates that work on a shared cloud computer, use connected tools and routines, and continue tasks while your devices are closed.
Gemini agent
Google Cloud's universal enterprise work agent can plan outcomes, use skills and tools, connect to business systems, schedule tasks and respond to events under enterprise controls.
Codex
OpenAI's software-development agent for inspecting repositories, editing code, running commands and tests, reviewing changes, and scheduling bounded background work.
Sources and update log
Primary facts come from the official product or pricing page. Added 2026-11-05; last editorial review 2026-11-05. Corrections can be sent through our correction process.