How to Use
ChatGPT and Claude
for MarketingPrompts Are Just Level 1...
Everyone is asking AI to write their marketing. Nobody is asking it to DO their marketing.
In this article
You typed "chatgpt for marketing" into Google and got back a thousand articles of prompts.
Let's answer the question you were really asking...
AI can do three different jobs for you, and which one you get depends on how you're running it, not which subscription you pay for.
Level 1 is the Claude/GPT chat web browser
You ask questions and get answers.
Level 2 is the Desktop app
It starts building things (ads, emails, etc) for you.
Level 3 is the Desktop app + automation
Now you're cooking. It builds ads for you, deploys them AND monitors them (and updates/pauses them based on performance if you want).
Most people use Level 1. They treat AI like a google search because that's what they're used too.
How to use ChatGPT and Claude for marketing
Try these three things, in order:
- Work out which of the three levels you're on today.
- Pick one marketing task you already do well by hand.
- Give it your material (ie existing ads, emails, content, etc) instead of adjectives. Five examples of your own good work beat any instruction to "make it punchier".
Below are three prompts, one per level. Only the first runs today with no setup. The other two are there so you know what's possible.
Runs in any Claude or ChatGPT web chat: Click the block to select it all
Here are five things I've published that did well: [paste them]. Here's who my customers are and what they buy: [two or three sentences]. Read those first. Then, without asking me anything yet: 1. Tell me what those five have in common that I probably can't see. 2. Draft the next one, start to finish, in the same register. 3. List what you'd need access to in order to do this without me next time.
That third instruction is the useful one. It hands you your own route to Level 2, in the order that fits your business instead of mine.
Once your files and tools are connected via Desktop app: Click the block to select it all
Read the folder I've connected and the results sheet in it. Find the five best performing ads of the last year and tell me what they share. Then build the next one as a finished ad I can open and look at. Use my own words from the folder, not general best practice.
Once your accounts are reachable via AI and it runs on a schedule: Click the block to select it all
Once a week, pull last week's numbers from my ad accounts and my email platform. Rank what's working by [the outcome I care about], not by clicks. Write it into one file, shortest version first. Flag anything that changed by more than [my threshold], and tell me what you'd turn off. Don't turn anything off without asking.
Notice what happens to your sentences as you climb. At Level 1 you describe what you want. At Level 3 you describe what you want to be true, and where the answer should end up.
Three AI Levels: chatting, working, running without you
Allan Dib's Lean Marketing splits leverage into three kinds: tools, assets, and processes. That maps onto the levels almost exactly.
Level 1 is a tool.
ChatGPT or Claude in a browser tab or on your phone. You ask, it answers, you copy what it wrote and paste it where it needs to go.
It's genuinely useful, and it has one permanent property: you're the transport. Nothing arrives anywhere unless you carry it.
For what it's worth, when I pulled 1,050 AI-related keywords from Google Keyword Planner in August 2026 (US, English, Google Search), prompt searches came to 0.1% of the volume in the marketing group. And by 'pull' I mean I told Claude 'give me top ai related keyword searches in us'. I didn't manually click anything inside the Google Ads UI.
Level 2 is a tool plus your assets.
Both companies let you connect their Desktop apps to your files.
Claude's own description of connectors is that they let it "access your apps and services", "retrieve your data", and "take actions within connected services", and on the desktop it reaches your local files and folders through a desktop extension. ChatGPT connects to Google Drive, SharePoint, Dropbox, Box, Gmail, Outlook, HubSpot and others, and which of them you get depends on your plan.
Once it can reach your files / accounts, it stops describing things and starts building them.
Level 3 is a process.
Same desktop app, plus automation: work that runs on a schedule or on a trigger instead of when you sit down.
This mode runs through tools like Claude Code and Codex.
Anthropic describes Claude Code as a tool that "reads your codebase, edits files, runs commands, and integrates with your development tools". Codex works the same way.
Both reach outside services through MCP connectors, described in the same docs as an open standard for connecting AI tools to external data sources, ie - Facebook ads, Google Ads, Mailchimp, Hubspot, etc.
That's the part that matters here. It's getting your AI to reach an ad platform or an email platform instead of just your documents.
Level 2 is a few clicks. Level 3 is a real one-time piece of setup, and no article can do that part for you.
Content marketing: your next asset isn't an article
The most useful thing these tools have done for me isn't writing. It's building things my customers can use.
The Spark's Conversation Starters is the one I'd point at.
It's a connections questions tool I build for my weekly email for couples, and it earns its place because it gives readers something instead of telling them something.
I'm thinking content the way we've known it may be on its way out.
The most valuable thing you publish this year probably isn't another article, it's something your customers can use.
That's the shift I'd bet on, and it starts at Level 2, where you stop describing the thing and start building it.
Level 1, the chat window.
You type: "Here are three posts that did well for us. Give me ten topic ideas in the same lane, then outline the best three."
You get topics and outlines in the window. Same for headlines, rewrites, and meta descriptions.
It's all text you then have to carry somewhere yourself, which is why content work with AI usually stalls here.
Level 2, the desktop app.
Connect all your posts and it works from those instead of your description of it.
ChatGPT/Claude can pull from Google Drive, SharePoint, Dropbox or Box, and its research mode produces a cited report drawing on the sources you allow it rather than a chat reply. Plus local files and folders on the desktop, or your cloud services anywhere (ie your Wordpress site, etc).
Then it builds. ChatGPT's canvas gives you a document you edit side by side with it, and Claude's artifacts turn a description into a working, interactive page with no code written by you.
At Level 1 you get a description of the page. At Level 2 you get the page.
Level 3, the desktop app plus automation.
Claude Code and Codex work in the real files of your site: write the page in, run the build, check it, commit the change, open a pull request.
This is also where the thing you've been meaning to give away finally gets made.
Dib's point about giving the result in advance is that the asset worth building is the one that helps before anyone buys. For most businesses that's a calculator, a checklist/audit, or a template.
At Level 1 you get it outlined. At Level 3 it gets built and put on your site.
One honest limit. These tools are better at improving writing than at replacing a writer, and the clearer you write, the more you get out of them.
Email marketing: next week's email is already written
I use Keap / Infusionsoft for my CRM.
Now with Level 3 AI, I have the option (if I want it) to build my own CRM and save $14k a year.
I'm planning on doing this within the next 30 days.
That's a big build. So if you're not at Level 3, don't feel like you need to do that now. I just want you to see what's possible.
Level 1, the chat window.
You type: "Write ten subject lines for this email, five plain-benefit and five curiosity."
You get ten, you pick one, you paste it into Keap (or Mailchimp, or whatever you use).
Give it the rules up front: one topic, one thing to click, ask for a reply. That's the difference between ten usable subject lines and ten clever ones.
Level 2, the desktop app.
Now it can reach your emails (ie last year's export, your best sends, your open and click data).
It writes from what your list responded to instead of from general email advice.
ChatGPT's connector list includes Gmail, Outlook and HubSpot, depending on your plan. Claude's connectors can retrieve your data and take actions in the services you've connected, with published examples like sending a message in Slack or creating an issue in Linear.
Is your email platform on either list? That's the first thing to check, because it decides whether the work ends up inside the platform or comes back to you as text to paste.
You're still approving everything at this level.
Level 3, the desktop app plus automation.
With your email platform reachable from Claude or ChatGPT, one instruction covers the whole job:
Pull the last 12 months of sends from my email platform. Rank them by [revenue / replies / whatever I care about], not opens. Write next week's email in the style of the top 3. Create it as a draft in the platform. Don't send it.
You want to email weekly. Most people don't, because writing it is the thing that gets pushed to tomorrow. At Level 3 the draft is sitting there waiting on your yes.
Then the audit almost nobody does by hand: which sequences still earn their place, and which ones have been sending to nobody for a year?
And you keep the send button. The job can end at your approval and still be Level 3, as long as everything before your approval happened without you.
Marketing automation: it watches your ads so you don't
I have standing rules watching ad performance that pause what stops working. I'm not in a dashboard when it happens, and I don't get asked.
The part I didn't expect was how much of the job was never really judgment. It was noticing, on time.
Dib makes a point that stung a little when I first read it: the automations inside the systems you already pay for are the most underused feature you own. The abandoned-cart email is the famous example.
Level 1, the chat window.
You type: "Map a five-email welcome sequence for this offer, with the trigger for each one."
You get a sensible map. Then you go build all of it by hand in your automation tool.
Level 2, the desktop app.
Some of the building happens instead of being described. Claude can take actions inside connected services, and ChatGPT's agent mode works a browser directly: it sees the page, clicks, fills forms, and navigates, asking you to confirm anything high-impact.
It's supervised work. It's also a real change from getting instructions you then follow.
While you're in there: tag more than feels necessary. Irrelevant sends are the fastest way to get ignored, and the tags are what make the next level's rules possible.
Level 3, the desktop app plus automation.
Your automation stops being something you rent inside somebody's flowchart.
Your AI writes the rule, runs it, and fixes it when it breaks. Claude Code has scheduled runs both in the cloud and on your own machine, and both Claude Code and Codex work from a command line, so they drop into whatever already runs your jobs.
One piece of advice that'll save you a month: automate something you've run by hand first. If you've never done the job yourself, you can't tell a good result from a confident wrong one, and that's the failure mode of automation in general, not of these tools.
Paid ads: ask, don't read
No more looking at reports.
Instead I get to just ask Claude (or gpt): 'What ads are below cpa right now and how much have they spent so far? Hmm, if it's over $100 turn those ones off'
And boom it's done.
Most of what I want to know about an ad account was never really a report anyway. It's a question. Which hooks/headlines are working? What ads have best cost per checkout?
Level 1, the chat window.
You type: "Here are three ads that worked for us. Write ten variants of the opening line, five short and five long."
You get ten. You pick two, and you paste them into Google Ads or Meta yourself.
Then it ends. Nothing comes back the other way, because what got clicked never reaches the window. Next month you're describing your own results from memory.
(That's the real ceiling at Level 1, and it's easy to miss, because the copy it writes is decent.)
Level 2, the desktop app.
Export your performance data and put it where the app can reach it.
Ask what your winners have in common before you ask it for anything new. Then have it write the next set from that answer.
Your best ads are the brief (ie input) to give to Claude/GPT.
Level 3, the desktop app plus automation.
With your ad accounts reachable from the Claude or ChatGPT, which is the setup part, you really unlock some power.
An audit across every campaign rather than the three you had time to open and export to CSV.
OR launching and pausing ads/campaigns just by asking for it, instead of by clicking 1000 things through Facebooks stupid UI dashboard.
Think of how much time that would save you.
Agency running your accounts? Get faster answers.
Plenty of people reading this own the logins while an agency does the driving, and the agency is doing good work. None of this is about firing anybody.
What changes at Level 3 is access.
A question occurs to you on a Tuesday, the answer is sitting in accounts you already own, and today you wait for the monthly report to see it. With your own accounts reachable from Claude/ChatGPT, you can ask them yourself that morning, get the answer within five minutes, and move on to the rest of your day.
Where to start, based on your level
You can name your level now.
If you're on Level 1, don't jump to automation. Get your own material in front of it first: your best posts, your best emails, your customer language. Most of the disappointing output at Level 1 comes from asking for work while giving it nothing of yours to work from.
If you're on Level 2, connect one more thing than you've connected today. Files first, then the one service where you do the most repetitive work. Then hand over a whole job instead of a piece of one.
If you're headed to Level 3, start with something read-only. An audit of what's live, or a ranking of what worked across a year. Get it right while it's only reading, then let it write.
That's the shift at every level. Stop prompting. Start shipping.
The test isn't whether the output was good, it's whether the work left your desk.
I run my own marketing for SHO at Level 3 today. My ad accounts and my email platform are operated from Claude Code, and I don't log into the dashboards to do the work. I'm not describing where this is heading, I'm describing a Tuesday.
If you read all this and decide the chat window is plenty for what you're running, that's a real answer. Level 1 does useful work, and plenty of businesses would be fine forever at Level 2.
If you want help getting your own accounts there
The distance between the levels isn't knowledge, it's setup: which accounts you can reach, what to hand over first, and what stays under your hand.
That's a conversation rather than a tutorial, which is why this article stops where it does.
If you'd like to work through your own setup with me, I offer a paid consulting session.
Bring one marketing job you already do well by hand, and we'll work out which level it belongs at.