How to build an AI marketing team with ChatGPT Work and Dot

ChatGPT has quietly turned into a powerful platform for building AI agents that can actually run parts of your marketing. With ChatGPT Work, Codex, and the new 24/7 Dot assistant, you can set up a small "AI marketing team" that plans campaigns, designs landing pages, creates content, and even talks to you in Slack—without writing any code.

ChatGPT Work, Codex, and Dot: what each one does

Before you start building agents, it helps to understand the three main pieces you’ll be using: Work, Codex, and Dot. They’re powered by similar underlying agent tech, but they’re designed for different kinds of jobs.

ChatGPT Work: for business outcomes

ChatGPT Work is where you send broader, multi-step business tasks. Think of it as your workspace for:

• Research and analysis
• Reports and presentations
• Spreadsheets and planning docs
• Marketing assets and websites tied to a specific project

You use Work when the deliverable is a business outcome: a campaign plan, a deck, a content calendar, or a landing page.

ChatGPT Codex: for working with code and files

ChatGPT Codex is a developer-focused environment. It’s built for:

• Inspecting and editing files across a repository or project folder
• Generating and modifying code
• Managing project structure behind the scenes

In this marketing setup, you don’t have to touch Codex directly. Work and Dot will route tasks to Codex when they need to create or update project files, skills, or websites.

Dot (your 24/7 assistant): for ongoing responsibilities

Dot is a named assistant that acts like an executive assistant with its own computer. It can:

• Use ChatGPT Work, Codex, and your custom tools
• Access your local files (with permission)
• Join your Slack workspace
• Get its own email address
• Talk to you via voice using the GPT voice model

In this setup, Dot is not tied to a single project. It sits above your projects and agents, takes tasks from you wherever you are, and coordinates work with the rest of your AI marketing team.

Designing your AI marketing team structure

Instead of one giant "do everything" agent, it’s more effective to build a small team of specialized agents, each responsible for a clear area of marketing.

A simple structure looks like this:

• Marketing team lead – your main point of contact inside a project
• Growth strategist – turns goals into campaigns and strategy decks
• Content creator – writes posts, articles, and campaign content
• Creative designer – designs visuals and ad creatives
• Website designer – builds and refines pages and landing sites
• Dot (e.g., "Emily") – your executive assistant that coordinates everything across tools and projects

Each specialist agent owns one domain (strategy, content, design, web, etc.). Dot sits outside the hierarchy and can jump into any project to delegate work and keep things moving.

Setting up your local project workspace

To keep your AI team organized, start with a local project folder for your brand. For example, imagine you’re working on a brand called Lexora, an online language learning platform for working professionals.

Your folder might include:

• A context file with brand information: what you do, target customers, tone of voice, visual style, key offers, and positioning
• Subfolders for marketing outputs: ads, social content, landing pages, lead magnets, reports
• An agents.md file: a simple handbook explaining how AI agents should work inside this project (roles, expectations, routing rules)

This structure becomes the shared memory and workspace for your AI agents. You then create a new ChatGPT Work project and connect this folder as a source so the AI understands your brand and has a place to save outputs.

Why you need reusable agent skills

Just because ChatGPT knows your brand doesn’t mean it knows how to work like your team. That’s where agent skills come in.

Skills are reusable playbooks that define how to do specific tasks. Instead of rewriting long prompts every time, you:

• Define a clear process once (e.g., how to build an ad set, or a landing page)
• Save it as a skill inside the project
• Let your agents call that skill whenever they need it

If you want a deeper primer on what skills are and how they work across different tools, see this guide to AI agent skills.

Creating your first marketing skill: ad creatives

A great first skill to build is an ad creative skill that turns a campaign brief into on-brand ads.

The process looks like this inside ChatGPT Work:

1. Describe the skill: for example, "Take a campaign brief, generate ad concepts, and produce final ad images with brand logo, colors, and copy."
2. Let ChatGPT generate the skill: it will create a new skill file under a hidden .agent folder in your project (no coding required).
3. Reuse the skill: whenever you need new ads, attach or reference the campaign brief and ask the project to generate ads using the ad creative skill.

The result: the AI will call your ad creative skill, propose visual concepts for a platform like Instagram, design layouts with consistent branding, and save everything to your project folder—concepts, copy, and image assets.

Adding a website builder skill for landing pages

Next, you can create a website builder skill that generates high-impact websites and landing pages based on specific marketing goals.

When defining this skill, you might ask it to:

• Create one-page offer sites or campaign landing pages
• Use your brand context and goals (e.g., free trial signups)
• Generate a design reference file listing great example sites for inspiration
• Deploy to ChatGPT Sites for quick hosting and testing

Once the skill is created, you don’t need to stay inside the project to use it. You can simply message Dot (e.g., "Emily") and say something like: "Find the Lexora project and build a one-page free trial offer website. Deploy it with ChatGPT Sites." Dot will route the task to Codex and Work, use the website builder skill, and return a draft site you can iterate on.

What the AI-built website looks like in practice

Using the website builder skill, ChatGPT can generate a complete landing page in under 30 minutes, including:

• A clean, professional layout with clear sections and hierarchy
• On-brand colors, typography, and imagery
• Interactive elements you can click and drag within zones
• Images created with the latest GPT image model

You can then refine copy, swap in proprietary images or videos, and keep using the same skill to generate new pages for future campaigns.

Skills vs agents: when to use which

As your library of marketing skills grows (ad creative, website builder, content scheduling, analytics summaries, etc.), you’ll eventually want to add specialized agents on top.

A simple rule of thumb:

• Use skills when a task is repeatable, with a clear, consistent process and output format (e.g., "Build a landing page," "Generate six Instagram posts," "Export a PDF deck").
• Use agents when a task requires judgment: interpreting goals, prioritizing, choosing which skills to use, and owning an area of work (e.g., "Design a Q4 growth strategy," "Plan a launch campaign").

Skills are your playbooks. Agents are your specialists who decide which playbook to run and when.

Building a growth strategist agent

One of the most valuable specialist agents you can create is a growth strategist.

You don’t need a huge prompt. Keep it simple and define:

• Persona: a data-informed growth marketer who understands your audience and product
• Scope: responsible for turning business goals into growth strategies and campaigns
• Role: uses research, analytics, and your skills library to create strategy docs and decks

In ChatGPT Work, you create this agent inside the project. ChatGPT generates the underlying configuration files in the background (stored under a doc_codex folder), but you never have to edit them manually.

Now, when you tag this agent in the project and give it a goal—"Propose Q4 campaign ideas for our free trial"—it will:

• Interpret the goal in the context of your brand
• Select relevant skills (e.g., research, ad creative, presentation building)
• Produce a structured campaign strategy and a slide deck

The decks are typically clean and professional, with each slide built around a core idea. You can also ask it to add images or diagrams as needed.

Creating a content creator agent

Next, you can add a content creator agent that owns all campaign content production.

Define it with:

• Persona: a content marketer who deeply understands your audience
• Scope: social posts, blog articles, email sequences, and supporting campaign copy
• Role: turns briefs and strategies into publish-ready content in your brand voice

Once created, this agent can work hand-in-hand with the growth strategist: the strategist proposes the plan, and the content creator fills it with posts, captions, and long-form content.

Letting Dot coordinate your agents through Slack

One of the biggest productivity wins comes from connecting Dot to Slack and letting it coordinate your agent team for you.

After you add Dot (e.g., "Emily") to a Slack channel, you can:

• Message it like a teammate
• Reference existing briefs and campaigns
• Ask it to coordinate with your project agents and skills

For example, once a campaign brief is approved, you might say in Slack:

"We need two weeks of social content for the upcoming campaign launch. Use the approved brief and coordinate with the team."

Dot will then:

• Locate the correct brand project
• Call the growth strategist and content creator agents as needed
• Use relevant skills (e.g., content generation, scheduling, asset packaging)
• Keep you updated in Slack as work progresses

After a short while, Dot can return with a compiled PDF overview of all posts (visuals and copy) plus a ZIP file containing:

• The campaign brief
• A posting schedule
• Individual folders for each post with images and text

From your perspective, it feels like delegating to a real teammate and getting a neatly organized delivery back.

Auto-scheduling content to your tools

If you connect Dot to tools like Notion, it can go one step further and schedule content for you.

For example, you can ask Dot to:

"Take the approved two-week content plan and schedule all posts in our Notion content calendar."

Dot will then use a content scheduling skill, create or update entries in your Notion database, and confirm when everything is ready for the team to review.

Updating your project instructions for better routing

As you add more skills and agents, it’s important to update your agents.md (or similar) project instruction file.

You can ask ChatGPT to:

• Review all agents and skills in the project
• Add routing guidance explaining when to call each specialist agent vs using a skill directly
• Clarify responsibilities (e.g., "Growth strategist owns campaign design," "Content creator owns social and blog content," "Website designer owns landing pages")

This helps the system make smarter decisions about delegation and reduces the need for step-by-step instructions from you.

Running a full multi-agent marketing task

Once your system is in place, you can give it a complex, multi-step marketing task and let the team handle it.

For example, inside your brand project you might say:

"Create a new lead magnet campaign to support our Q4 launch. Use past campaign performance data to inform the strategy."

The AI will typically:

• Share a plan and list which skills and agents it will use
• Have an analyst or strategist agent review historical performance
• Use creative and website skills to build the lead magnet and landing page
• Coordinate subagents to produce all assets and flows

You can even watch this in real time under a "subagents" view, where you see your multi-agent team working together.

What the lead magnet workflow can produce

A well-configured multi-agent setup can automatically generate:

• A polished lead magnet (e.g., a team meeting guidebook) with on-brand design, practical tips, worksheets, and a clear CTA
• A campaign landing page hosted locally first for testing, using the same visual identity as the guidebook
• An embedded lead capture form connected to your tool of choice or to ChatGPT Sites’ built-in database

You can then test the full flow: fill out the form, receive the downloadable guidebook PDF, and verify that leads are stored correctly. Once you’re happy, you can ask ChatGPT to deploy the page via ChatGPT Sites so it’s live and ready to share.

Using Dot for multi-agent tasks on the go

You don’t have to be in the desktop app to trigger complex workflows. Because Dot can be reached by Slack, email, or even voice calls, you can:

• Email it a new idea or brief
• Call it to talk through a campaign concept
• Ask it to coordinate with your agents and tools while you’re away from your desk

As long as Dot has access to the relevant project and integrations, it can spin up multi-agent tasks, route work, and send you a summary when everything is ready.

Scaling up your AI marketing system

Once you’ve built a basic structure—brand context, a few core skills, and 1–2 specialist agents—you can keep expanding:

• Add more skills for analytics, email sequences, A/B test planning, or reporting
• Introduce more specialist agents (e.g., a performance analyst, a CRO specialist, or a dedicated email marketer)
• Refine your project instructions and routing rules over time

If you’re interested in similar setups on other platforms, there are also guides on how to build an AI marketing team with Claude Code or other agent frameworks. The core ideas—clear context, reusable skills, and focused agents—stay the same.

The real power comes when your system knows your brand well enough that you no longer have to re-explain it or micromanage tasks. You simply state the outcome you want, and your AI marketing team figures out the steps.