ChatGPT Dots Custom Rules for Marketing: How to Set Permissions Your Agent Will Actually Follow

TL;DR

• ChatGPT dots custom rules for marketing let you set four permission levels per action, from full autonomy to "hand off to me", so your always-on agent can publish, spend, email, and update CRM data within boundaries you define.
• The four permission levels are: Take action without asking, Take action when you say so, Ask before taking action, and Hand off to you. Built-in auto-review always applies and cannot be overridden.
• For marketing teams, the most useful configuration is to let your dot research and draft autonomously, require approval before publishing or spending, and block destructive CRM actions entirely.
• Custom rules live under Settings > Personalization > Custom Rules in ChatGPT desktop. You create one rule per action, describe it in plain language, and pick the autonomy level.
• Common mistakes include confusing app permissions with custom rules, writing rules that are too vague, and forgetting that auto-review can block actions even when a rule says "take action without asking."

If you are a marketer who just got access to ChatGPT dots, you have probably already asked your always-on agent to schedule a campaign, draft an email blast, or update a CRM record. And then you wondered: can it actually send that? Will it check with me first? What happens if it spends my ad budget while I am asleep?

The answer lives in ChatGPT dots custom rules for marketing, a permissions layer that decides exactly how much autonomy your dot has for each category of action. Dots ship with sensible defaults, but the real power for marketing teams comes when you tailor those rules to match your workflow, your risk tolerance, and your approval chain.

This guide covers the four permission levels every marketer needs to know, provides copy-paste prompt templates for the five most common marketing jobs, includes a read-vs-change-vs-approve matrix, and walks through the troubleshooting gotchas that the official docs gloss over.

What Are ChatGPT Dots Custom Rules and Why Marketers Need Them

OpenAI launched dots on September 29, 2026, always-on agents powered by GPT-6 Astra that live inside ChatGPT, run their own cloud computer, connect to over 4,000 apps, and keep working between conversations. For marketers, this means an agent that can draft social posts, monitor campaign performance, research competitors, and even build landing pages while you are in meetings or asleep.

But handing an autonomous agent the keys to your ad manager, email platform, and CRM without guardrails is a fast way to wake up to a budget-draining surprise. That is exactly what ChatGPT dots custom rules for marketing are designed to prevent.

According to the official ChatGPT Learn documentation, custom rules are "optional controls for specific ongoing boundaries." They supplement your conversational instructions with persistent, action-level permissions that your dot tries to follow every time, not just in the current thread. The four options are:

A Gartner survey on AI agent governance published in mid-2026 found that marketing teams using persistent agents without permission boundaries reported 3× more "unauthorized publish" incidents than teams that configured action-level controls before deployment. Well-configured ChatGPT dots custom rules for marketing are the single most effective way to capture the speed gain without the risk.

Auto-Review: The Safety Net You Cannot Turn Off

Before your dot saves a draft, spends a dollar, or sends an email, an automatic review checks the action against your instructions, your custom rules, and OpenAI's built-in safety requirements. If the review determines the action could affect your accounts or share information, it requires approval, even if your rule says "take action without asking." This is by design. You cannot disable auto-review, and you cannot use custom rules to bypass safety constraints like using saved login credentials without confirmation.

How to Set Up ChatGPT Dots Custom Rules for Marketing

Setting up custom rules is straightforward once you know where to look. Follow these steps in the ChatGPT desktop app or on desktop web:

1. Open your dot's profile. Select your dot's name or avatar in the ChatGPT sidebar.
2. Navigate to Custom Rules. Go to Settings > Personalization and select Custom Rules under the Permissions section.
3. Add a rule. Click Add, then describe the action in plain language, for example, "Schedule a LinkedIn post" or "Increase ad campaign spend above $50."
4. Choose the permission level. Pick one of the four behaviors from the table above.
5. Save. The rule takes effect immediately for your dot across all channels, ChatGPT, Slack, and Teams.

A few important constraints to remember:

• Custom rules apply to your dot within the same account. They do not automatically grant app access; you must connect and authorize each app separately in the Plugins settings.
• Rules are instructions your dot tries to follow, but it can still make mistakes. Review consequential work.
• If your workspace admin has disabled custom rules (common in Enterprise deployments), you cannot edit saved rules and they do not apply.

For a deeper look at how Metaflow helps teams manage approval rules for AI agents across multiple platforms, check out how to build agent approval rules for marketing.

ChatGPT Dots Custom Rules for Marketing: Rules by Job

Here is where this guide moves beyond documentation and into practical workflow. Each of the following sections addresses a common marketing job-to-be-done and gives you a ready-to-use prompt block, plus the custom rule configuration you need. Across all five jobs, ChatGPT dots custom rules for marketing follow the same pattern: specify the action precisely, pick the permission level, and let the dot execute within that box.

1. Content Publishing and Scheduling: ChatGPT Dots Custom Rules for Marketing Automation

Your dot can draft blog posts, social updates, and email newsletters. Without a custom rule, it will draft but will ask before publishing anything. That is the safe default. If you want your dot to autonomously schedule low-risk social content (e.g., curated links, evergreen posts) but require approval for homepage copy, create two rules that cover both scenarios:

• Rule 1: "Schedule social media posts in Buffer for evergreen content" → Take action without asking
• Rule 2: "Publish or update content on the company blog" → Ask before taking actionPrompt template:

[Your dot's name], monitor our content calendar in [tool name]. When a scheduled post is due in the next 24 hours and the copy is already approved in the draft folder, publish it without asking. If the copy has not been reviewed yet, flag it to me and wait.

2. Ad Spend and Budget Management

This is the highest-risk category for most marketing teams. A runaway agent could burn through a monthly budget in hours. OpenAI's auto-review catches some of this, but adding a custom rule is cheap insurance.

• Rule 1: "Adjust ad campaign budgets on Google Ads and Meta Ads" → Ask before taking action
• Rule 2: "Pause underperforming ad sets when ROAS drops below 2.0" → Take action without asking
• Rule 3: "Increase daily ad budget above $100" → Hand off to youPrompt template:

Monitor our [platform name] campaigns daily. Flag any campaign where ROAS drops below [threshold] and automatically pause it. For budget increases, prepare the recommendation and show me the projected impact before making any changes.

3. Email and Social Sending

Sending email campaigns to your subscriber list is another action that should never happen without human review, even if your dot wrote perfect copy. The consequences of a tone-deaf send or broken link are too steep.

• Rule 1: "Write and prepare email campaigns in Mailchimp" → Take action when you say so
• Rule 2: "Send email campaigns to more than 100 recipients" → Ask before taking action
• Rule 3: "Reply to customer support tickets" → Hand off to youPrompt template:

Draft the weekly newsletter in [platform name] each [day of week] morning. Show me the full preview. I will give you a thumbs-up or request edits before you send. Never send to more than [number] recipients without a separate approval.

4. CRM Data Operations

Your CRM (HubSpot, Salesforce, or similar) contains your most sensitive marketing data. A mistaken merge or deletion can take days to undo.

• Rule 1: "Create new contact and lead records in CRM" → Take action without asking
• Rule 2: "Merge or delete CRM records" → Ask before taking action
• Rule 3: "Update deal stage for existing opportunities" → Take action when you say soPrompt template:

Import leads from our [source, e.g., webinar landing page] into [CRM name]. Tag each lead with the campaign name and source. If you find a duplicate, flag it to me rather than merging automatically. Never delete records.

5. Analytics and Reporting

Low risk, high value. Let your dot run wild with data analysis and report generation.

• Rule 1: "Pull and analyze data from Google Analytics, Meta Ads, and Google Ads" → Take action without asking
• Rule 2: "Generate weekly performance reports in Google Slides" → Take action without asking
• Rule 3: "Share reports with the team in Slack" → Take action when you say soPrompt template:

Every Monday morning at 9 AM, pull last week's metrics from [sources]. Build a one-page report in [tool name] covering sessions, conversions, CPA, and ROAS. Share the report link in our [#team-channel] Slack channel, but do not post commentary, just the link and a one-line summary.

The Read vs. Change vs. Approve Matrix

One mental model that makes ChatGPT dots custom rules for marketing easier to configure is to classify every action across two dimensions: what the action touches (read, change, or delete) and who must approve it (agent, human, or both). The entire set of custom rules you write for your dot should be guided by this logic.

This matrix is a starting point. Your actual configuration depends on team size, campaign risk, and compliance requirements. For a deeper framework on how to design approval workflows for AI agents across your marketing stack, read our guide on building no-code AI agents for marketing.

Common Mistakes and Troubleshooting with ChatGPT Dots Custom Rules

Even with well-written rules, things can go wrong. Here are the most common issues marketers run into, and how to fix them.

Rule is configured but the dot still asks for approval. Auto-review is likely blocking the action. OpenAI's safety documentation explains that certain actions always trigger review regardless of your custom rule. Check whether the action involves sharing information outside your account, using a saved password, or modifying account-level settings. These are non-negotiable guardrails.

The dot ignores the rule entirely. You may have written the rule too vaguely. OpenAI's docs state that "custom rules are instructions your dot tries to follow, and it can make mistakes." A rule like "Be careful with money" is too vague. Replace it with "Ask before increasing any ad campaign budget above $50", specific, measurable, and actionable. When you are testing ChatGPT dots custom rules for marketing for the first time, start with stricter permissions and loosen them as you build trust with your dot. This is covered in more detail in the complete guide to top AI agents for marketing.

The dot still makes changes inside a connected app even though the rule says ask first. Remember that custom rules are separate from app permissions. Granting a dot access to HubSpot or Google Ads through the Plugins panel gives it the ability to make changes at the API level. The custom rule tells the dot when to act; the app connection tells the system whether it can act. Revoking an app's write permissions at the plugin level is a hard block. Using a custom rule alone is a soft instruction that the dot can misinterpret.

Confusion between Custom Instructions and Custom Rules. Custom instructions (found under your ChatGPT profile photo > Settings > Personalization > Customization) set the dot's general behavior, tone, and preferences. Custom Rules (under Permissions) set permissions boundaries. If you want your dot to write in your brand voice, use custom instructions. If you want it to not send something without your say-so, use custom rules. They complement each other but are not interchangeable.

Deleting your dot deletes everything. Per the official controls page, "save any results you need before confirming; the action cannot be undone." Deleting your dot removes its conversations, memories, and scheduled tasks. It does not undo changes already made in connected apps, though, so if your dot published a campaign before you deleted it, that campaign stays live.

FAQ

How many ChatGPT dots custom rules can I set?

There is no documented hard limit, but practice suggests keeping your rule set under 15, 20 rules. Each rule should cover one action type at one permission level. If you find yourself writing more than 20, group related actions under broader descriptions, for example, "Manage social media publishing" instead of separate rules for LinkedIn, Twitter, and Instagram.

Can I use ChatGPT dots custom rules for marketing on mobile?

You can review rules on mobile, but you must create and edit them on desktop. Open your dot's profile on the ChatGPT desktop app or desktop web, navigate to Settings > Personalization, and select Custom Rules under Permissions. Mobile support for rule creation has not been announced.

What happens when my workspace disables custom rules?

If your organization's admin has disabled custom rules in the workspace settings, saved rules stop applying and you cannot edit them. The dot reverts to its built-in behavior, which already includes safety defaults and auto-review. This setting is common in Enterprise and Healthcare deployments where compliance teams want uniform agent governance.

Can custom instructions override custom rules?

No. Custom instructions shape how your dot works, tone, format, preferences. Custom rules shape when your dot takes action. The two systems run independently. If a custom instruction tells the dot to "always schedule posts immediately" but a custom rule says "ask before publishing," the rule wins because it sits in the permission layer that auto-review checks.

Should I set custom rules before or after connecting apps?

Before. Connect apps first, then configure rules before assigning any real tasks. This way the dot immediately knows which actions require approval and which it can take independently. Your first round of rules should be intentionally strict, you can always loosen them later.

Your Next Step

Dots are new, OpenAI launched them less than a week ago. The marketing teams that get the most value will be the ones that invest time early in configuring ChatGPT dots custom rules for marketing with specific, action-level permissions rather than relying on vague high-level instructions.

Start with the strictest settings. Let your dot research, draft, and analyze freely. Require approval before it publishes, spends, or deletes anything. Audit its work weekly. Loosen rules only when you have confidence that your dot understands your standards.

For teams that want a centralized way to manage agent permissions across multiple AI platforms, not just ChatGPT but Claude agents, custom-built agents, and API-driven workflows, Metaflow provides a unified approval-rules layer. See how it works in the AI agent approval rules for marketing walkthrough.