Original Insight
“I’m having it build an outline first, and then I edit the outline, and then I have it build a draft. And then I have it come out in 3 short post formats: one which is conservative, one is middle of the road, the other is really way out there and edgy, you know, intended to stop the scroll.” — Don Back
“Once I’ve got the first draft out of ChatGPT, I bring it out, and it’s 100% me after that. Print it out, pick up a pencil or a pen, and I go through and edit and rewrite it in my language.” — Don Back
“Train the model. Say, I want you to write like me. I’m going to give you examples of what I don’t sound like and what I should sound like, and give it 3, 4, 5 examples. And then say, write me a profile that explains how to get any piece of content into my style.” — Lou
Expanded Synthesis
Don Back’s content production workflow — refined through weeks of experimentation — is one of the most complete and practically implementable AI content systems described in the July sessions. It deserves to be treated as a framework, not just a personal anecdote.
What makes it valuable is not the AI component specifically. It is the human judgment layer that sits at each decision point. Don has figured out exactly where AI creates leverage (overcoming the blank page, generating variations, structuring arguments) and exactly where human judgment is irreplaceable (voice, editing, publishing decision). The result is a workflow that is simultaneously faster and more distinctively human than either pure AI generation or pure human writing.
The workflow in full:
Step 1 — Psychographic anchor Don built a detailed ideal client profile using a structured multi-session Chat process: assembled a panel of experts, developed 11 psychographic dimensions, created a comprehensive avatar document. This is not done for every piece of content — it is built once and referenced repeatedly. The output: a single source of truth about who he is writing for.
Step 2 — Content pillar extraction From the avatar profile: “Give me 10 content pillars.” Each pillar is a domain of topics that maps to the ideal client’s core challenges, fears, aspirations, and identity.
Step 3 — Topic calendar From each pillar: “Give me 6 months of topics.” This eliminates the perpetual question of what to write about. The editorial calendar exists before any writing begins.
Step 4 — Outline first For each piece, generate an outline. Edit the outline before drafting. This is the highest-leverage human intervention in the process — shaping the structure of the argument before a single word of the draft exists.
Step 5 — Draft generation with format variations Generate the draft in three formats simultaneously: conservative, middle of the road, and edgy/provocative. This is a powerful mechanism because it prevents you from defaulting to the safe version by habit — you see what’s possible at each end of the spectrum.
Step 6 — Print, pencil, and voice Remove the draft from the AI environment entirely. Print it. Edit with a pen. This physical separation is deliberate: it changes the cognitive mode from “AI user” to “writer.” By the time Don finishes editing, it is his writing.
Step 7 — Voice profile training Lou’s extension of this is the logical next step: take 10-12 of these before/after pairs (AI draft vs. your edited version) and train a voice profile. Give the AI examples of what your unedited writing looks like and your edited version. Ask it to derive a style guide: “Write me a profile that explains how to get any piece of content into my style.” Now the editing step becomes significantly shorter because the first draft is already closer to your voice.
Why this beats the standard AI content workflow:
Most people use AI for content in one of two broken ways. The first is pure generation: prompt the AI, receive text, publish it. This produces generic, AI-sounding content that doesn’t build a real audience. The second is incremental revision: prompt, receive text, ask for improvements through ten rounds of back-and-forth, eventually give up and rewrite most of it yourself. This is slow and frustrating.
Don’s system is neither. It uses AI for the structural scaffold (outline, draft, variations) and human judgment for everything that actually determines whether the content builds relationships and trust. It’s a genuine partnership rather than either replacement or iteration.
The engagement signal: Don reported that content produced through this system is getting more engagement on LinkedIn than his previous approach. The specific observation: “It resonates with people when I stick it up on LinkedIn.” This matters because it suggests the human editing step is successfully embedding the authenticity that AI alone cannot produce.
Practical Application for PowerUp Clients
The Content System Setup Protocol:
Week 1: Build the foundation
- Create your ideal client psychographic profile (use Don’s multi-stage approach with a panel of experts prompt)
- Extract 5-10 content pillars
- Generate 3 months of topics per pillar
Week 2: Establish your voice profile
- Find 5 pieces of writing that represent your ideal voice
- Find 5 pieces of AI draft that represent what you don’t want
- Ask AI: “Analyze the difference between these two sets. Write a style guide that would transform the second set into the first set.”
- Test the style guide on new drafts before committing
Week 3: Build the production workflow
- Set a consistent format for your content request (outline first, then draft, then 3 variations)
- Establish your editing ritual (print + pencil, or equivalent analog intervention)
- Define your publication cadence and protect that time in your calendar
The ADHD-friendly format tip (from Don):
- Heavy bullet structure
- Single points per bullet
- Highly scannable
- Clear hierarchy This is not just personal preference — it matches how LinkedIn’s algorithm rewards content that people actually read rather than scroll past.
Journal prompts:
- Where does my current content creation process create the most friction? Is it ideation, structuring, writing, or editing?
- What would my ideal voice sound like if AI could produce it without my editing? What’s the gap?
- What 10 pieces of content have I written that best represent who I actually am and how I think?
Additional Resources
- Insight - Codify Your Judgment Into Skills, Not Just Prompts
- Insight - Persistent AI Memory via MCP — Building a Cross-Session Intelligence Layer
- Hook-Story-Offer framework (Tony Robbins / Dean Graziosi) — referenced as the narrative structure to embed in content pillars
Evolution Across Sessions
This insight builds directly on July 17’s discussion of Kimi’s writing quality and represents a full workflow implementation. It connects to the July 24 voice authentication work Kasimir is doing via MCP — these are two different implementation paths toward the same goal: AI-assisted content that sounds genuinely like you. Together they form a comprehensive picture of what the near future of personal content production looks like.
Next Actions
- For me (Lou): Build and share a one-page Content System Setup Protocol as a mastermind resource; demo the voice profile training technique in a future session
- For clients: Complete Week 1 of the Content System Setup Protocol; focus first on the psychographic anchor — everything else is easier once this exists