OpenAI Dots brings always-on AI agents into ChatGPT, with a focus on ongoing responsibilities rather than one-off prompts. A dot can keep context between conversations, use its own cloud computer, work across connected apps, run recurring tasks and bring results back for review.
OpenAI announced dots on September 29, 2026. The company describes them as persistent agents powered by GPT-6 Astra. For marketing teams, the interesting part is not another AI writing tool. It is the possibility of keeping context moving between research, campaign monitoring, CRM data, reporting and team decisions.
That makes the real question less about what a dot can generate and more about whether it can reliably keep a recurring operating loop moving:
research → campaign monitoring → CRM → Slack or Teams → reporting → recurring work
Most marketing teams already have tools for every stage. The problem is that the context rarely moves cleanly between them.
What OpenAI Dots is and what it actually changes
A dot is an always-on AI agent in ChatGPT designed to stay attached to ongoing responsibilities. OpenAI says each dot has its own cloud computer, can work across connected apps, remember context, schedule recurring work and continue making progress between conversations.
That is different from a normal chatbot workflow. A chatbot can analyse a campaign report when someone uploads it. A dot can be given an ongoing responsibility, such as monitoring an approved set of campaign and CRM signals, then returning with findings when something changes.
OpenAI also says dots can proactively review connected information and form memories from it. Scheduled work can run without a new prompt, while the user can pause the dot, review its activity and define rules for actions that require approval. The official product name is simply “dots,” although searches for ChatGPT Dots refer to the same experience.
The product direction builds on ideas OneMetrik has already covered in the OpenAI Agents API. Useful agents increasingly depend on persistent context, tools, permissions and clear handoffs, not only model intelligence.
Why OpenAI Dots matters for marketing workflows
Marketing work rarely happens inside one application. A competitor changes pricing, campaign CPL rises, sales notices weaker lead quality, someone flags the issue in Slack and the weekly report gets updated days later. The useful opportunity with OpenAI Dots is reducing those disconnected handoffs.
| Workflow stage | What a dot could help with | What should stay human-led |
| Research | Monitor approved sources and collect material changes | Decide what deserves action |
| Campaign monitoring | Detect anomalies and investigate context | Approve budget and targeting changes |
| CRM | Compare campaign performance with pipeline quality | Control sensitive record changes |
| Slack or Teams | Bring evidence into the right discussion | Make strategic decisions |
| Reporting | Prepare summaries with supporting context | Validate attribution and conclusions |
| Recurring work | Repeat the process as inputs change | Review exceptions and performance |
This is where the broader shift toward AI marketing automation tools becomes more practical. The value is no longer just producing something faster. It is reducing the manual work required to move information from one decision to the next.
Research becomes a trigger instead of a tab graveyard
Marketing teams usually do not have an information shortage. They have an information routing problem. Competitor announcements sit in browser tabs, customer objections live in call transcripts, product updates appear in Slack and closed-lost reasons sit inside the CRM.
An OpenAI Dots workflow could maintain a running research brief across approved sources and surface a change when it becomes relevant to a live campaign or sales conversation. The useful output is not a list of 20 links. It is context such as: a competitor changed enterprise pricing, recent sales calls mentioned the same objection and closed-lost notes show the issue is becoming more frequent.
Now the research has somewhere to go. Paid media can review messaging, product marketing can revisit positioning and sales can prepare for the objection. That is much closer to an operating workflow than asking an AI tool to summarise the news.
Campaign monitoring becomes more useful when it can see CRM outcomes
Campaign platforms generate endless alerts, but alerts rarely include enough business context. Google Ads might show that CPA increased 18%. LinkedIn might show CPL rising 22%. Spend might be pacing ahead of plan. Those numbers deserve investigation, but they do not automatically deserve action.
A dot with approved access could identify the change, compare recent periods and inspect relevant CRM signals before escalating it. “CPL increased 22%” could become “CPL increased 22%, but SQL rate improved and cost per opportunity declined.” That is a more useful business signal.
Our AI PPC reporting guide makes the same distinction. Collection and anomaly detection are useful, but reporting becomes valuable when the team connects the metric to an outcome.
This is also why B2B marketing attribution matters before recurring analysis is handed to an agent. Ad platforms know clicks, spend and conversion events. The CRM knows whether those conversions became qualified leads, opportunities and revenue. Faster AI analysis of incomplete data is still incomplete analysis.
A dot cannot repair poor measurement. Clean source tracking, lifecycle stages and opportunity data still need to exist. Our guide to connecting ad spend to pipeline and revenue covers that foundation in more detail.
Slack and Teams can become the decision layer
A surprising amount of marketing context already lives in team conversations. Sales says lead quality is slipping, paid media identifies where it started, product marketing explains a messaging change and someone remembers that targeting was expanded five days earlier. The problem is that those observations often stay inside separate threads.
OpenAI documents support for adding dots to Slack and, for eligible workspaces, Microsoft Teams. That creates a more useful path for bringing an investigation to the people who need to make a decision, rather than sending another isolated dashboard alert.
A useful update might say: “Paid search CPL increased 16% this week. The increase is concentrated in Campaign A. Opportunity rate remains stable. Sales flagged two poor-fit accounts after targeting expanded Monday. No budget change recommended yet.” The point is not that the agent makes the decision. The point is that it carries the evidence into the decision.
OneMetrik’s coverage of the Slackbot AI workspace reflects the same direction: collaboration tools are increasingly becoming places where AI can bring together shared context, scheduled work and team decisions.
Reporting becomes the memory of the workflow
Most automated reporting tools automate presentation. They do not preserve the reasoning behind a decision. A dashboard knows spend increased, but it may not know that the increase followed a targeting change, that CRM quality improved or that competitor research influenced new positioning.
Because dots are designed to keep context across ongoing work, the same agent following the workflow could prepare a weekly report that captures what changed, what was investigated, which CRM signals supported the finding, what the team decided and what still needs monitoring.
That turns reporting into operational memory. It also fits the broader goal of AI performance marketing, where platform metrics matter only when they are connected to lead quality, CAC, pipeline and revenue.
Recurring work is the real test for always-on AI agents
The promise behind always-on AI agents sounds impressive, but the harder test is whether the same workflow still works on the fourth Monday. Consider a weekly paid media and pipeline review. The agent checks approved campaign data, compares results with CRM outcomes, investigates meaningful changes, looks for relevant research or internal context, prepares a concise report and surfaces unresolved issues to the team. Then it repeats the process the following week without rebuilding the brief from scratch.
That is where persistent context starts having operational value. Teams testing a dot should measure at least four things:
- Analyst time saved.
- Factual corrections required.
- Important issues the agent missed.
- Findings that actually influenced a decision.
If an agent saves 45 minutes but creates an hour of correction work, the automation failed. That is a better standard for AI agents for business than simply counting completed tasks.
Controls determine how far OpenAI Dots should go
Persistent agents need tighter controls than one-off chat sessions because they keep encountering new information over time. OpenAI lets users define rules for actions a dot can take without asking, actions that must be pre-approved, actions that require confirmation and actions that should be handed back to the user.
For marketing teams, the safest progression is straightforward: start with research, then monitoring, then reporting and draft preparation. Actions involving budgets, CRM changes, publishing or customer communication should sit behind stronger approval rules until the workflow has proved reliable.
OpenAI also warns that dots can make mistakes, including when following custom rules. That makes review quality part of the workflow design, not an optional final step.
According to OpenAI’s current setup guidance, dots are rolling out gradually. They are available to eligible Pro users and Business Premium users, while Enterprise access is available as a beta when enabled by a workspace administrator. Creation currently starts on desktop web or the desktop app, while mobile interaction depends on rollout availability.
OpenAI says dots usage will not count toward eligible Pro, Business and Enterprise plan allowances during the initial launch period, with longer-term usage terms to follow. That makes early pilots easier, but teams should still measure the cost of reviewed, accurate work rather than the number of tasks completed.
How to test OpenAI Dots with one recurring marketing workflow
Teams do not need to rebuild marketing operations around agents immediately. Pick one process with predictable inputs, clear thresholds and a known reviewer. For many B2B SaaS teams, a weekly paid media and pipeline brief is a practical starting point.
- Give the dot the minimum read access it needs.
- Define the campaign metrics and CRM stages that matter.
- Set thresholds for what deserves investigation.
- Require supporting evidence for every conclusion.
- Keep consequential actions behind approval.
- Run the workflow long enough to see whether it stays useful as conditions change.
Then ask one question: did the process become easier without making the decisions worse? If yes, add another responsibility. If not, more autonomy will probably make the problem larger.
What marketing teams should watch next
The next phase of OpenAI Dots will be less about headline capabilities and more about whether integrations, governance and economics hold up in recurring business use. Integration depth matters because marketers need the right data from the specific CRM, analytics and advertising systems they already use. Governance matters because persistent agents need clear permissions, review rules and activity history.
Cost is the third variable. Temporary launch allowances can encourage experimentation, but high-volume recurring work still has to justify itself against analyst time and workflow quality. The useful metric is not tasks completed. It is cost per accurate, reviewed workflow.
OneMetrik Takeaway
OpenAI Dots matters because the unit of AI adoption is shifting from individual tasks toward ongoing responsibilities. For marketers, the strongest opportunity is not automated copywriting. It is reducing the handoff cost between research, campaign monitoring, CRM, team conversations and reporting.
Research should inform what the campaign team watches. Campaign changes should be checked against downstream CRM outcomes. The relevant evidence should reach the people making decisions. Reporting should preserve why those decisions happened. Then the process should repeat without rebuilding the context next Monday.
Start narrow, keep the agent read-heavy while the workflow proves itself, measure correction time as carefully as time saved and give it more authority only after the process is consistently reliable. If dots can keep that loop moving, they become much more useful than another chatbot waiting for the next prompt.