ChatGPT dots are OpenAI's ongoing agents inside ChatGPT. The product is described around responsibilities that continue between conversations, using chosen apps and remembered context. A founder evaluating a dot should decide which responsibility it owns and how facts will be corrected as the work changes.
Evidence checked October 22, 2026. This guide separates official product statements from an original planning exercise. We did not run the product or verify account-specific access.
What is documented, and who can access it?
The official dots page describes a cloud computer, ChatGPT memory, connected tools and review or pause controls. The more specific getting-started help describes gradual, plan- and workspace-dependent rollout. A feature appearing on the website is not proof that it is enabled in your account.
At this source check, dots access for Pro users excludes the EEA, Switzerland and UK and requires age 18 or older. Business Premium eligibility differs, and Enterprise beta access requires an administrator to enable it. Recheck the help page for the current conditions rather than treating “available in ChatGPT” as universal access.
This article concerns ChatGPT dots, not a third-party product called Dot, a “connect the dots” metaphor or every regular ChatGPT conversation.
An ongoing responsibility needs an update rule
A one-off prompt can end when the draft is returned. An ongoing responsibility also needs a rule for what changes the answer. A founder's approved story may be stable, while a meeting time, source document or permission to disclose a metric changes quickly.
Keep durable preferences separate from task facts. “Use concise meeting briefs” can be a preference. “This investor is reviewing our deck” needs dated evidence. If those are mixed together, an old fact may return with the confidence of a standing instruction.
A context-and-correction worksheet
| Context item | How to label it | What makes it stale | Review action |
|---|---|---|---|
| Writing preference | Preference, supplied by founder | Founder changes preference | Confirm and replace the instruction |
| Current investor status | Dated fact with source | New explicit reply or meeting outcome | Reconcile the working record |
| Approved company metric | Value, period, definition and owner | New period or corrected calculation | Replace the value and preserve the correction reason |
| Relationship interpretation | Inference, with supporting observations | Contrary evidence | Ask owner before using externally |
| Sensitive detail | Restricted input, specific permitted purpose | Purpose ends or permission changes | Follow actual removal controls and verify scope |
Use a “current for this task” packet rather than relying on a remembered statement alone. Attach a version date and explicitly name which prior statement it supersedes. This is an operating recommendation, not a claim about how all memories are implemented or removed.
Fictional worked example: correct the responsibility, not just the sentence
Fictional planning exercise, not output from ChatGPT dots. Ren is preparing a conversation with Marais Ventures, a fictional fund. The original packet says a pilot is planned. A new approved note says the pilot was postponed. A third note, written before the postponement, asks for a “pilot results” slide.
A useful responsibility brief would identify the newer note as the current statement, retain the source date and ask what replaces the results slide. It should not convert a planned pilot into customer traction. The follow-up draft can describe the current product and ask whether the revised discussion is still useful.
Ren's correction receipt should say what changed, which statement is superseded and which outputs need review. That receipt can be checked in an actual authorized trial. Until such a trial occurs, neither memory reliability nor successful correction has been demonstrated.
Verify controls separately from the assignment
Check scheduled work, connected-app permissions and pause controls in the actual account. When testing a correction, inspect both the next answer and any already-prepared artifacts. A correct reply does not establish that every related document changed.
The help documentation distinguishes disconnecting an app from deleting information already obtained. Read the current reset and deletion instructions before relying on removal. Do not put sensitive customer material into a trial just to see what happens.
Use a correction-aware preparation prompt
This manual briefing prompt organizes a reviewable artifact. It does not connect accounts, schedule work or authorize external actions.
Prepare an ongoing-responsibility brief from the packet I supply. Separate durable preferences, dated facts, inferences and restricted information. Identify which newer fact supersedes which older statement. Produce a meeting brief, a list of affected drafts and questions for the owner. Preserve source dates. Do not create a schedule, contact anyone or modify records. Do not assume old memory is current evidence. Return a correction receipt for my review.Where Aurora by Finta fits the decision
For relationship work, the useful comparison is whether the working record, sources and next step stay understandable after a correction. Finta's meeting-preparation workflow offers a concrete operating recipe to evaluate. Read Aurora versus ChatGPT dots before choosing a product, and the agent landscape or always-on guide for the concepts.
Questions about the interface
Can a dot call me without my asking?
The current help page says a dot cannot initiate calls at launch. Do not confuse speaking with an agent with receiving an agent-initiated phone call.
Is texting broadly available?
The product page says texting is coming next, while the linked help describes a limited US Pro beta where enabled, excluding Business and Enterprise. Confirm the account-specific beta instead of inferring universal SMS access.
Does ongoing work remove the need for review?
No such conclusion follows from the documentation. Define the responsibility, inspect its output and use the actual controls for consequential actions. An ongoing task still needs an accountable person.
Disclosure: Finta publishes this independent guide. The product discussed is not presented as a Finta partner or integration. Product access, documentation and account controls can change.