Guide · ChatGPT Dots
ChatGPT Dots
for Marketing
A ChatGPT Dot is a personal agent that runs a standing job inside ChatGPT: it reads from the apps you connect, drafts the work, and acts only where a Custom Rule lets it. For marketers that is a real step up from a chat window, because the same weekly read of HubSpot, Shopify, or Klaviyo can run every Monday without anyone re-typing the prompt. This guide covers which marketing jobs belong to a Dot, how the plugin and the rule decide what it may touch, why Google and Meta ads need a different answer, and the handful of things — sending, posting, publishing — that stay with a person.
When should a marketing job become a Dot?
Assign a job to a Dot when it recurs, the data lives in a plugin you can connect, and you have already decided what the Dot may do on its own. If any of those three is missing, the chat window is still the better tool.
Most marketers meet Dots the wrong way round. They open one, ask it the same question they would have asked ChatGPT, and conclude it is a slightly more elaborate chatbot. It is not. A Dot is a standing job: a plugin that supplies the data, a Custom Rule that sets the boundary, and a schedule or trigger that runs it without you. A question you will never ask again gains nothing from any of that and should stay in a normal chat.
The useful test is whether you can name next Monday's version of the job. "How did the Shopify launch do?" is a chat question. "Every Monday, read Shopify and HubSpot, tell me which products moved and which deals stalled, and draft the week's discount code for me to approve" is a Dot. The second one has a plugin, a recurring shape, and an obvious place for a rule.
| You need | Chat or Dot | Why |
|---|---|---|
| One answer, no follow-up next week | Chat | Nothing a Dot persists is worth setting up for a question you will not repeat. |
| A weekly read from HubSpot, Canva, Shopify, Klaviyo, Slack, or Teams | Dot, once you have written one rule | The plugin supplies the data every run. The rule decides what the Dot may do with it. |
| A page, email, or ad to go live on its own | Neither — keep a person on the publish | Drafting can be allowed. Publishing, sending, and spending are approve or block. |
| A shared team Dot with its own IT credentials | Not yet | Specialist dots are an enterprise pilot. A personal Dot on Pro or Business Premium is what you can staff today. |
| Live changes in Google Ads or Meta Ads | Dot plus a person, or an ads agent | There is no first-party ads plugin. Section 04 covers what that means. |
Setup itself takes about ten minutes once you know the job. How to set up a Dot for marketing walks through it step by step; the two decisions that matter most are which plugin to connect first and which Custom Rules to attach before the first run.
What can a Dot read, draft, and change?
Two things decide what a Dot may do: the plugin, which determines what it can see and touch, and the Custom Rule, which determines whether it is allowed to. Get the pairing right and the Dot is safe; get it wrong and you have an agent with write access and no brakes.
A Dot reads whatever its connected plugins expose. The first-party plugins worth planning around today are HubSpot, Canva, Shopify, Klaviyo, Slack, and Microsoft Teams. Search Console and GA4 are reachable too, but through a connector rather than a native plugin, and the setup posts for each explain the difference. From those reads a Dot can draft almost anything a marketer drafts: social posts, CRM notes, discount codes, content briefs, ad recommendations, email flow copy.
Whether the Dot may then do anything with the draft is a separate question, and it is answered by Custom Rules. ChatGPT presents the options with four labels, but they collapse to three positions. Allow means the Dot takes the action without asking — the label reads "take action without asking." Approve means the Dot finishes the draft and waits for you; both "ask before" and "when you say so" land here. Block means the Dot does not do it at all, and hands the task to you — the "hand off to you" label.
| Plugin | What the Dot reads | What it may change on allow | Rule for anything a customer sees or that spends |
|---|---|---|---|
| HubSpot | Contacts, deals, activities | Create and update records | Approve |
| Canva | Designs, brand kit | Create a design from a template | Approve before it becomes the asset you ship |
| Shopify | Orders, inventory, products | Discount codes, inventory flags | Approve |
| Klaviyo | Campaign and flow performance | Drafts and templates only | Block on send |
| Slack, Teams | Channel history | Post a summary | Approve — the ping is not the sign-off |
| Google Ads, Meta Ads | Not a first-party plugin | Nothing native | See section 04 |
The two rows people misread are Klaviyo and ads. Klaviyo exposes performance data and lets the Dot draft flows and templates, but it does not let the Dot send, and you should block send explicitly rather than rely on that. Google and Meta are not on the list at all, which is a bigger topic than one row can hold.
A good rhythm for a new Dot is one job, one plugin, one rule, and two weeks of drafts on approve before you move anything to allow. The plugin list covers each integration in detail, and the Shopify connection guide is the best worked example of a plugin paired with a rule.
Which recurring jobs should stay in draft?
SEO, AEO, website growth, content, email, and the weekly growth scoreboard are all jobs a Dot can prepare completely and should never finish on its own. The Dot does the reading and the first draft; a person publishes.
These are the jobs where Dots earn their keep, precisely because they are mostly reading and drafting. A Monday growth scoreboard that pulls GA4 and HubSpot, names what moved, and proposes one test is a Dot job from end to end — right up until someone has to decide what to run. The same is true of an SEO refresh queue built from Search Console, an AEO pass that checks whether AI answers cite you, or a Klaviyo flow rewrite. In every case the Dot can get you to the decision. It should not make it.
| Job | What it reads | What it drafts | What it changes | Who ships |
|---|---|---|---|---|
| Weekly funnel, experiment backlog, channel scoreboard | GA4 or HubSpot — not the ad account | Monday diagnosis, one proposed test, what moved | Nothing | You, in the growth meeting |
| SEO | Search Console, once connected | Refresh brief, new title | Not the URL | You |
| AEO | The same, plus whether AI answers cite you | The passage you would want cited | Not the page | You |
| Website growth | Search Console and GA4 | Decay queue, CRO notes | Nothing | You |
| Content | Notion, Drive, Canva | The post and the asset | A Canva file, if allowed | You publish |
| Klaviyo performance | Flow copy, templates | No send | You send |
The temptation, once a Dot has produced good drafts for a month, is to widen the allow-rule so it can publish too. Resist it. The publish step is not a permission problem; it is a pipeline problem — review, scheduling, distribution, and learning from what shipped — and that pipeline is what content-led growth is built to run.
Why can't a Dot run Google or Meta ads on its own?
There is no first-party Google Ads or Meta Ads plugin. Add the Metaflow plugin in ChatGPT. The Dot asks. Metaflow reads the connected account and drafts the change. Spend waits for you.
This is the question most people are really asking when they ask about Dots for marketing, so it deserves a straight answer. ChatGPT does not ship a plugin for Google Ads or for Meta Ads. If you want a Dot to see your campaigns, add the Metaflow plugin in ChatGPT. The Dot inherits it. The Dot asks. Metaflow reads campaigns, search terms, budgets, and performance, and it drafts the change — pause this, add that negative, move this budget.
What it cannot do is make those changes go live without you, and you should not try to make it. Spend is never an allow-rule. Metaflow queues the change, you approve or block it, and that is the whole operating model. It works well for a single account reviewed weekly. It does not scale into a system that manages budgets on a cadence across a portfolio unless that work lives in Metaflow, because every change still runs through one person’s approval queue.
| Path | What you get | What you do not get |
|---|---|---|
| A native Dot plugin | Nothing — it is not in the plugin list | A settings screen, an API key, or a write |
| The Metaflow plugin | Reads of campaigns, terms, and budgets, plus queued changes | A change that goes live without you |
| ChatGPT Ads | A place to buy ads inside ChatGPT | Any access to your Google or Meta account |
| A dedicated ads agent | Changes applied inside the account under a policy you set | A Dot, or a Dot allow-rule |
One more source of confusion is worth clearing up: ChatGPT Ads is OpenAI's advertising product for showing ads inside ChatGPT. It has nothing to do with your Google or Meta accounts and is not a pipe into either. Dots and ChatGPT Ads covers the distinction.
When the ad account is the job rather than something a Dot reads about, use an agent built for it. Google leaves to AI agents for Google Ads, Meta leaves to AI agents for Meta Ads, and a paid book across both channels leaves to performance marketing agents. The Dot still does not run the ad account. The Metaflow plugin runs that agent from ChatGPT.
What does a person still have to send?
A Dot can research a prospect, write the sequence, and brief you on the account. It cannot send the email, and it cannot operate a browser on LinkedIn. Outbound is a drafting job for a Dot and a sending job for you.
Outbound is where the read-and-draft model is most obviously the right one. A Dot connected to HubSpot and a research source can scan for ICP matches each weekday morning, assemble a company-intelligence brief, draft a cold email sequence, and even summarise your LinkedIn Ads performance. Every one of those outputs lands in front of a person. None of them go out on their own, and the LinkedIn case is stricter still: the Dot does not get a browser session on linkedin.com, so posting, messaging, and connection requests are off the table entirely.
| Job | The Dot does | Blocked |
|---|---|---|
| Cold email | Research, sequence draft | Send |
| Outbound system | Prospect brief | Enrollment and send |
| Weekday ICP scan, post drafts, LinkedIn Ads summary, company brief | Any browser action on linkedin.com — posting, messages, connection requests |
The four LinkedIn jobs in that table are the four in the LinkedIn post, and they are the complete list; there is no fifth one where the Dot gets to click. If what you need is the send itself — enrolment, sequencing, replies — that is outbound automation, not a wider Dot rule.
What can a Dot not do for your marketing?
Most of these limits are deliberate. If a row in this guide says leave, the right move is a different tool, not a broader allow-rule that papers over the gap.
Dots are new, and some of what follows will change as the plugin list grows and specialist dots move out of pilot. But the pattern underneath is stable: a Dot is a personal agent with a rule, and it is designed to stop short of the actions that reach a customer or spend money. Plan around that rather than against it.
Checklist
- No first-party Google Ads or Meta Ads plugin. Add the Metaflow plugin. The Dot asks. Metaflow reads and queues the change.
- Metaflow is the ChatGPT plugin for this job. Connect the ad account in Metaflow. The Dot does not write the campaign itself.
- Specialist dots are a pilot. A standing org identity with IT-scoped credentials is an enterprise trial, not something a team can roll out. A personal Dot on Pro or Business Premium is what you can staff now.
- Klaviyo does not send, LinkedIn does not get a browser, and pages are published by a person. These are the three places people most often try to widen a rule and should not.
- ChatGPT Ads is a different product. Buying ads inside ChatGPT has nothing to do with your Google or Meta accounts.
- One client's allow-rule does not cover the next client. Agencies copy the rules and split the Dots per client. They do not share one Dot across accounts.
- Prompts are the job description, not the gate. A prompt library tells the Dot what to do. Only a Custom Rule decides whether it may.
Agencies have the sharpest version of every limit above, because every client is a separate set of plugins and rules. Dots for marketing agencies covers the per-client architecture; Dots prompts for marketing is the copy-paste library, with the reminder that none of it replaces the rule.
FAQ
A Dot is a personal agent inside ChatGPT that runs a standing job. You connect a plugin such as HubSpot, Shopify, or Klaviyo, write the job, and attach Custom Rules that say what the Dot may do on its own. It then runs the job on a trigger or schedule without you re-prompting it. The difference from a normal chat is persistence: the data, the rule, and the job survive between sessions.
No. There is no first-party Google or Meta ads plugin. Add the Metaflow plugin. The Dot asks. Metaflow reads the connected account. You approve or block the write. Spend should never sit under a Dot allow-rule. If you need bids to move on a policy while you sleep, that job lives in Metaflow, not in the Dot.
They are the three positions a Custom Rule can take. Allow means the Dot acts without asking. Approve means it finishes the draft and waits for you. Block means it does not perform the action and hands the task to you. ChatGPT shows four labels in the UI, but "ask before" and "when you say so" both map to approve.
Not yet. Specialist dots — a shared Dot with its own role and IT-scoped credentials — are an enterprise pilot. What a team can staff today is a personal Dot per person on Pro or Business Premium, with Custom Rules copied across. Treat the specialist version as something to watch, not something to plan around.
No. Slack and Teams deliver the ping that tells you a Dot has something waiting. The approval itself is the Custom Rule — allow, approve, or block. A message in a channel is not a sign-off, and you should not treat a posted summary as permission for whatever comes next.
No to both. A Dot can research prospects, draft sequences, and brief you on an account, but send stays blocked. On LinkedIn the restriction is stricter: the Dot does not get a browser session, so posting, messaging, and connection requests are not possible at all. If sending is the job, use outbound automation rather than a Dot.
Key takeaways
- A Dot is a standing job with a plugin and a rule. A one-off question belongs in the chat window.
- Allow drafts. Approve anything that publishes, sends, or spends. Block the rest, and give new Dots two weeks on approve before you widen anything.
- The plugin decides what a Dot can see; the Custom Rule decides what it may do. You need both set before the first run.
- Google and Meta ads go through the Metaflow plugin. The Dot asks. You approve. The Dot does not write the account.
- Specialist dots are a pilot. Plan with personal Dots and copied rules.
More reading
ChatGPT Dots