My Two-GPT Workflow for Creating Better AI Content Workflow
- Braden Barty
- Jun 16
- 6 min read
Updated: 5 days ago
Most AI content does not fail because the model is bad. It fails because the workflow is lazy.
Most people ask one AI model to be the writer, editor, strategist, creative director, and skeptical buyer. Then they wonder why the output sounds polished but dead.
The fix is simple: use one AI to create, another AI to critique, and a human to decide what ships.
That’s the workflow: Creator. Critic. Human judgment.
And it works way better than asking one model to magically do everything.
What This AI Workflow Looks Like
Here’s the workflow I’d steal if I were building better AI-assisted marketing content.
This example uses cold email, but the same idea works for LinkedIn posts, newsletters, video scripts, ads, landing pages, YouTube titles, and almost anything else where the first draft needs pressure before it goes live.
Use one AI model as the creator. In this case, that model is trained around Alex Berman-style cold email thinking: brevity, hooks, reply rate, and directness.
Then use a second AI model as the critic. In this case, that model is trained around Alex Hormozi-style offer thinking: value proposition, buyer psychology, offer strength, and whether the pitch actually feels worth caring about.
The names are not the main point. The roles are.
One model creates the asset. Another model stress-tests the weak spots. Then a human decides what survives.
For the cold email example, the first AI has one job: write a short, specific, outbound-focused email. The second AI has a different job: critique the offer. Is the value prop strong? Is the pain clear? Would the prospect care? Is this actually worth replying to?
That distinction matters.
Not rewrite it. Critique it.
Because the moment the critic starts rewriting, you are no longer running a stress test. You are blending voices.
And that is how you get AI soup.
Why This Works
Good marketing is rarely made by one brain doing one pass.
A writer brings the draft. A strategist challenges the angle. An editor cuts the weak parts. A creative director protects the voice. A buyer, usually in someone’s imagination, asks, “Why should I care?”
Most people collapse all of that into one prompt: “Write me a cold email.” Then they follow it with, “Make it better.”
That’s where AI starts smoothing instead of sharpening. The copy gets cleaner, but not more persuasive. The sentences get polished, but the point gets weaker. The final draft becomes technically better and actually worse.
Using two specialized AI roles creates useful tension. One model protects the format. The other attacks the weakness. That tension is where the better work comes from.
The Workflow
1. Start With the Creator
Use the first model to create the draft. For example, you might prompt it with: “Write a cold email using Alex Berman-style cold outreach principles. Keep it short, specific, and focused on getting a reply.”
Give it the context it needs: who the prospect is, what they do, why you’re reaching out, what problem you can help with, and any relevant company research.
The goal is not perfection. The goal is a strong first draft.
2. Send It to the Critic
Now send the draft to a different model with a different job. For example: “Review this cold email through the lens of offer strength, buyer psychology, value proposition, and clarity. Give feedback only. Do not rewrite it.”
That last line is the whole game.
Do not let the critic rewrite it.
If the critic starts rewriting, the original voice gets muddy. The Alex Berman style email slowly turns into a Alex Hormozistyle offer essay. And now you have the classic AI problem: more words, more polish, less punch.
The critic’s job is to find the weak spots. Not take over the keyboard.
3. Send the Feedback Back to the Creator
Now give the critique back to the original creator model and ask it to revise within the original constraints.
A good prompt would be: “Here is feedback on your draft. Revise the email to address these points while maintaining your original cold email style. Keep it under 100 words.”
That creates a useful constraint. Improve the idea, but protect the format. Do not let the email become a landing page.
A cold email expert usually wants brevity. An offer expert usually wants more value, proof, and urgency. Both are useful, but if you let the offer brain dominate, your cold email gets too heavy.
The Hidden Trap: The AI Echo Chamber
This workflow is powerful, but don’t loop forever.
Round one usually creates a big improvement because the critic catches the obvious problems: the hook is weak, the pain point is vague, the email is too long, or the offer is not specific enough.
Round two is often where you hit the best version. The draft gets sharper, the value prop gets clearer, and the copy still sounds natural.
Round three or four is where things can get weird.
The creator model starts writing to please the critic model instead of the human reader. That’s when you get copy that feels like it was approved by a committee of robots wearing Patagonia vests.
Technically optimized. Emotionally dead.
The email may score higher, but that does not mean it will perform better. Because the real audience is not the critic model. It’s the person opening the email.
Where Marketers Can Use This
This workflow works anywhere the first draft needs pressure before it goes live.
For LinkedIn posts, have one model draft the post, then use another to critique the first two lines, the clarity of the idea, and whether the post speaks to the right audience.
For YouTube titles, have one model generate options, then use another to critique specificity, curiosity, and whether the title matches the actual video. A clever title that attracts the wrong viewer is not a win.
For newsletter drafts, have one model write the article, then use another to critique the structure, the examples, and whether the reader would actually save or share it.
For sales pages, have one model write the copy, then use another to review objections, proof, CTA clarity, and whether the offer feels specific enough.
For video scripts, have one model write the script, then use another to flag where attention drops, where the opening drags, and where the viewer might ask, “Why am I still watching this?”
That’s the real value. Not more content. Better pressure before publishing.
Two Tips to Make It Work Better
1. Give the Critic a Real Scoring Rubric
Do not ask, “Is this good?”
That gets you polite nonsense.
Ask the critic to score specific things: clarity, audience relevance, offer strength, specificity, length, and likelihood of getting a response.
Then define what a 10 means. For example: “A 10 is an email you would confidently send to a high-value prospect. Be strict. Do not give a high score unless the draft is genuinely strong.”
AI is often too nice. Make it earn the compliment.
2. Protect the Format
Every asset has a job.
A cold email should not read like a sales page. A LinkedIn post should not read like a white paper. A video hook should not read like a podcast intro.
So when you send feedback back to the creator model, protect the format.
For cold email, say: “Keep this under 100 words. Prioritize reply rate over completeness.”
For LinkedIn, say: “Keep the opening sharp. Do not over-explain before the point.”
For video, say: “Get to the tension in the first 5 seconds.”
The critic can push for improvement, but the creator needs constraints. Otherwise, every draft slowly turns into the same overstuffed AI blob.
Braden’s Take
This is the kind of AI workflow marketers should be building. Not because it is complicated, but because it is disciplined.
Most people are still using AI like a vending machine: prompt in, content out, hope for the best.
Better AI work looks more like a production process. You assign roles, create tension, protect the goal, and know when to stop.
That last part is important because AI can keep improving something forever. But at some point, “better” just means “more acceptable to the model.” Not more useful to the reader. Not more persuasive to the buyer. Not more human.
Use AI to create. Use AI to stress-test. But do not outsource the final judgment.
That is still your job.
Annoying, I know. But also kind of the whole point.
Action Steps
Try this on one piece of content this week.
Pick something simple: a LinkedIn post, cold email, video hook, newsletter intro, ad concept, or sales page section.
CLICK HERE FOR CUSTOM Alex Berman GPT
CLICK HERE FOR CUSTOM Alex Hormozi GPT
Choose the creator role.
Tell one AI model what it is responsible for creating and what format it must protect.
Choose the critic role.
Tell another AI model what it is responsible for evaluating. Make sure it critiques only.
Give the critic a rubric.
Ask it to score clarity, specificity, audience fit, offer strength, and likelihood of action.
Send the feedback back to the creator.
Ask for a revision that addresses the feedback while preserving the original format and voice.
Stop after two rounds.
That is the system: Creator. Critic. Human judgment.
One model can give you a draft.
A better workflow gives you something worth publishing.
Comments