OpenAI Dots Do Not Replace Your Marketing Automation
OpenAI launched dots on September 29, 2026, and said it had tried specialist dots internally for email marketing, among other jobs, as VentureBeat reported. The obvious question followed within the day: if an agent is always on, connects to 4,000 apps and can act on its own, why keep a marketing automation tool at all?
Because they are different kinds of thing. A dot is a worker. Marketing automation is a delivery system. This post explains the difference, what goes wrong when a worker is asked to be a delivery system, and how to split the work. For what dots are, see OpenAI dots for marketing.
A worker and a delivery system
| A dot | Marketing automation | |
|---|---|---|
| Unit of work | A task or a goal | A contact |
| Memory | Your preferences and its notes | The state of every contact, separately |
| Timing | When the dot runs | When each contact is due |
| Judgment | Strong. It reads, compares and writes | None. It follows the plan exactly |
| Repeatability | Varies from run to run | The same every time |
| Record | An activity log | Delivery, replies and results per message |
Read the table by column and the conclusion is plain. Everything a dot is good at is judgment. Everything automation is good at is bookkeeping. Marketing needs both, and they do not substitute for each other.
What goes wrong when a dot does the sending
Suppose you connect an email tool with a send action, write a rule allowing it, and ask the dot to run your welcome sequence.
Retries. Agents retry when something fails halfway. A send engine is built so that a retry does not deliver twice. An agent calling a send tool has no such guarantee unless the tool provides it.
Opt-outs. Someone unsubscribes on Tuesday. The dot’s Wednesday run has to know. In an automation engine this is a hard stop inside the send path. For a dot it is one more thing to remember to check.
Overlap. The same person is in the welcome sequence and the launch campaign. An engine can see message pressure across every program. A dot sees the task in front of it.
Per-contact timing. Day three for a contact who joined Monday is Thursday. For one who joined Friday it is the next Monday. Multiply by a few thousand contacts and the dot is rebuilding a scheduler in its notes.
Attribution. When a reply or a sale arrives, which message caused it? The engine recorded the send. The dot recorded that it completed a task.
Variation. An agent may phrase, order or time things differently on each run. That is a strength in drafting and a defect in delivery.
None of this says dots are unreliable. It says that delivery is a bookkeeping problem, and that an agent with judgment is an expensive and indirect way to do bookkeeping.
Rules do not fix this
Dots have Custom Rules: allow, require approval, prohibit. It is tempting to think “require approval before sending” solves the problem.
It solves one problem, which is an unwanted send. It does not give the dot per-contact state, and approval wears down. By the twentieth prompt of the week, people approve without reading. The same pattern with an earlier generation of agents is covered in ChatGPT agent mode’s marketing limits.
The approval setting decides how often you are asked. The tool list decides what is possible. Only the second one holds when you are tired.
The split that works
| Job | Who does it |
|---|---|
| Notice that something changed | Dot |
| Check the calendar against the plan | Dot |
| Draft the campaign and the content | Dot |
| Review drafts against the brief | Dot, then a person |
| Decide to launch | A person |
| Enroll contacts and time each step | Automation |
| Respect opt-outs | Automation |
| Deliver and record | Automation |
| Tie results to the message | Automation |
The handover point is the draft. Everything before it benefits from judgment and tolerates variation. Everything after it needs to be exact.
How Marqeable is built around that split
Marqeable’s MCP server gives an assistant the first half of the table. It can read your campaigns, content, calendar, automations and business profile. It can create drafts of campaigns, content and automations, and propose revisions. It has no tool to send, publish, launch or turn on an automation.
The second half runs inside Marqeable. A person reviews the draft and launches it. Automations then handle enrollment, each contact’s timing, delivery and attribution across text and email.
So a dot connected to Marqeable can prepare a launch overnight, and the worst possible outcome is a draft you delete in the morning. The setup is in connecting Marqeable to a dot, and five concrete jobs are in five dot workflows.
We have not yet run this on a dot ourselves. Dots had not reached our account on launch day. The connector works today in ChatGPT, Claude and Codex, and we will update these posts once we have tested it with a dot.
When a dot alone is enough
There are cases where you do not need the second half.
- You send a handful of one-to-one emails a week and write each by hand
- You have no list, and marketing is posts and conversations
- You are testing a message with ten people before building anything
Once you have a list, a sequence or a consent obligation, you have a delivery problem, and the dot should be working for the engine rather than acting as one. The broader version of this argument is in AI agents vs marketing automation.
Frequently asked questions
Can an OpenAI dot replace my marketing automation tool?
Not sensibly. A dot is one agent doing tasks. Marketing automation is a system that tracks every contact separately: who enrolled, which step each is on, who opted out, what was delivered and what it led to. A dot can operate such a system. It is a poor substitute for one.
Could a dot just send the emails itself?
If you connect a tool that lets it, yes. The problems are the ones a send engine exists to solve: duplicate sends on a retry, contacts who opted out, the same person in two programs, and no record tying a reply or a sale back to the message.
What should a dot do in marketing, then?
The work around the sends: watching what changed, checking the calendar against the plan, drafting campaigns and content, reviewing drafts against the brief, and telling you what needs a decision.
How does Marqeable work with dots?
Marqeable’s MCP server gives an assistant read access to your marketing and the ability to create drafts. A person reviews and launches in Marqeable, and Marqeable’s automations handle enrollment, timing, delivery and attribution. As of launch day we have not yet run this on a dot ourselves.
The bottom line
The useful split is simple. The dot is the colleague who prepares the work. The automation is the machinery that delivers it to each contact correctly. Teams that keep those two jobs separate get the benefit of an always-on agent without handing it the one action that cannot be taken back.
See what the drafts become: Marqeable’s agents build and run your campaigns, run behavior-triggered automations across text and email, answer and qualify every visitor with AI website chat, and tie revenue to the exact message with attribution.
Marqeable runs your campaigns, answers every visitor, text, and email in seconds, and turns them into booked jobs and meetings - even at 9pm on a Saturday. We’re in private beta with a small early cohort. Get early access
