Notebook / Guides

Your stories, not a prompt: a content system for small-business owners

Apply

The interview is the product. The AI is the editor. Here is the system, who it comes from, and the templates you can copy.

By Elie Mendlow / Last updated October 26, 2026

Most small-business owners know they should market themselves. Most also know the feeling of opening a blank page on a Sunday night and writing nothing. AI was supposed to fix that. Instead, a lot of what it produces reads like everyone else: smooth, vague, and forgettable.

The people who get real results from AI-assisted content are not typing "write me a post." They run a system. We studied how several well-known creators and operators describe their process, found the pattern they share, and turned it into the MCAFax marketing skill you can add to your own AI. This article explains the system, credits where each idea came from, and is honest about what we do not know yet.

Where this started: Alex Lieberman's Content Machine

Alex Lieberman co-founded Morning Brew and now co-runs Tenex. In a July 20, 2026 episode of the How I AI podcast with Claire Vo (episode page), he walked through what he calls the Content Machine. He also summarized it in a LinkedIn post.

His core principle, in his words: "It's not a 'write me a post' prompt. The raw material must come from you. The machine never invents your voice and never fabricates your insight."

The steps he describes, condensed:

1. Load the creator's profile, style guide, and a running list of lessons.
2. An "Oracle" scans what he wrote in the last seven days for "spikes," moments where he naturally said something worth expanding, plus a reader that scans outside sources. Ideas are scored 0-10 and saved to an idea bank.
3. An optional research brief with linked facts and contrarian angles.
4. An "Interview Panel" of AI personas asks one question at a time and, per his post, "won't advance without 2-3 specific stories."
5. The interview becomes raw notes with his exact words preserved.
6. Only then does it draft, using his style guide and lessons.
7. A "Writer's Council" of reviewer personas, including a "Slop Detector," scores the draft. Anything under 9/10 is revised. Missing information goes back to the interview and is "never invented."
8. One approved anchor piece is repurposed into many platform-native pieces.
9. A learning loop compares his first draft with his final edit and saves confirmed lessons.

According to the episode write-up, he answers interview questions out loud with a voice-to-text tool, and his voice guide includes the rule "write like you're texting a friend." He also argued that bland output is usually an input problem: "The only time the content machine actually produces slop is more of an indictment of the person not sharing good enough ideas during the interview step." He also said he mapped his old manual process before automating anything.

You do not need his setup to use his insight. The interview is the product. The AI is the editor.

Six others who run a similar playbook

Ruben Hassid. In a February 2026 LinkedIn post, Hassid describes talking for 60 seconds into a voice-to-text app, pasting the transcript into an AI with strict rules ("No emojis. No jargon. No fluff."), then editing: "AI gives you 70%, your taste gives another 30%." In a June 2026 post and newsletter, he goes further: export your own past posts, have the AI find the hooks and formats that worked, and turn that into a personal skill that asks you questions before writing. His warning: "Feed it only your best."

Justin Welsh. In his essay on his LinkedIn writing process, Welsh lists three steps: start with the information you are teaching, then write the "trailer" (the hook and the lines that pull the reader down), then end with a short TL;DR summary and a "call-to-conversation" question.

Ship 30 for 30 (Dickie Bush and Nicolas Cole). Their Lean Writing method says to test an idea small and in public, then "Expand 'What Works'" into longer formats: a short post becomes an essay, then a thread, a guide, an email course, and eventually a product.

Gary Vaynerchuk. His content model is a "reverse pyramid": one piece of "pillar content," such as a video or podcast, cut into "dozens of smaller pieces of content, contextual to the platforms."

Tiago Forte. His Building a Second Brain guide lays out CODE: Capture, Organize, Distill, Express. His line "our brains are for having ideas, not storing them" is the case for an idea bank. With your notes collected, he says, you "never need to sit down to an empty page."

David Perell. In his Ultimate Guide to Writing Online, Perell writes that "speaking is the first draft of thought" and describes a Content Triangle: test ideas in conversation with friends, then with readers, then with editors. In an interview with Every, he says he watches for reactions he sums up as CRIBS: confusing, repeated, interesting, boring, surprising.

The common pattern

Different tools, same bones:

• Capture ideas continuously (Forte, Lieberman's Oracle). You never start from nothing.
• Talk before you write (Lieberman, Hassid, Perell). Speaking pulls out real stories faster than typing.
• Write down your voice (Lieberman's voice guide, Hassid's skill built from your own posts).
• AI drafts, a human edits (all of them). The owner makes the final call.
• Substance first, hook second (Welsh).
• Test small, expand what works (Ship 30, Perell).
• One anchor, many native pieces (Vaynerchuk, Lieberman).
• Learn from every edit (Lieberman's lessons loop, Hassid's quarterly refresh).

The MCAFax content system

Here is how we put it together for a small-business owner with an hour a week.

1. Keep an idea bank. Every time a customer asks a question, pushes back, or thanks you, write one line in a note. Those are your topics. Score each idea on three things: is there a real story, a clear opinion, and something useful for a buyer?

2. Get interviewed. Pick one idea. Let the AI ask you one question at a time, and answer out loud. Do not let it draft until you have two or three specific examples in your own words. This is the step Lieberman says separates good content from slop.

3. Draft from your notes, in your voice. The AI writes only from what you said, using your voice guide.

4. Write five hooks, pick one. Keep a hook bank: a running list of opening lines and how they did. Five patterns are useful starters: a pain point, a strong opinion, a real number with its source, a story opener, and a myth versus a fact.

5. Review before you publish. Score the draft against a checklist (below). Revise anything weak. If a fact is missing, you supply it. The AI never fills the gap.

6. Repurpose up a ladder. One anchor piece becomes three to five short posts, one email, one FAQ answer, and one short video script, each written for its platform. Post the small versions first, and expand the one that gets replies.

7. Save the lesson. When you change the draft the AI wrote, note why. Next time it starts from that lesson.

Voice interview questions

Ask these one at a time, out loud if you can. Do not let the draft start until two or three answers are specific.

1. What happened in the business this week that a customer would recognize?

   Push back: If the answer is that it was busy, ask what happened on one specific day.

2. What question did a customer ask you more than once?

   Push back: Ask what they asked, as close to their words as you remember.

3. What objection did you hear, and what did you say back?

   Push back: A vague we handle that is not a story. Ask for the sentence you said.

4. Tell one moment: who was there, what they said, and what you did.

   Push back: Do not invent a composite person. If a name is private, use a role and a real detail.

5. What did you have to explain twice?

   Push back: Ask what was confusing the first time.

6. What mistake did you make, and what changed after it?

   Push back: A lesson with no scene is a slogan. Ask where you were when it went wrong.

7. What opinion about your work would you say out loud to a customer you like?

   Push back: Quality matters is not an opinion. Ask what you disagree with.

8. What should the reader do, ask, or stop doing next?

   Push back: One next step, not a list of ten.

9. Which numbers are you allowed to cite? For each one: the number, the source, and the date.

   Push back: If there is no source or no date, the number stays out.

10. What is off limits: private names, a confidential deal, or anything you would not say in public?

    Push back: If you are unsure, leave it out.

Ask one question at a time. Push back on vague answers. Do not draft until you have 2-3 specific stories or examples in the words of the owner. Suggest talking instead of typing (voice-to-text).

Voice guide template

• Who I serve and what I sell
• Three to five topics I own
• How I sound (how I would text a customer I like)
• Words I use; words I never use
• Openings and structures from my best past pieces
• Proof I am allowed to cite (real numbers, with source and date)
• Lessons from past edits

Review checklist

• Every claim traces back to my notes or a linked source.
• Every number has a source and a date. No invented statistics.
• No guaranteed outcomes: no promised approvals, savings, income, or rankings.
• Finance-safe wording: use "may" and "can," not "will." No rate, fee, or approval promises. Point readers to a professional for legal, tax, or lending specifics.
• Plain language, short sentences, no jargon or AI filler words.
• It sounds like me.
• At least one real story or example.
• The first line earns the second.
• One clear point, one clear next step or question.
• Right length and format for the platform, with nothing private.

Copy the templates

These match the read-only file the marketing skill loads: https://mcafax.com/api/skills/marketing.json. Updated 2026-10-26. This skill is new. There is no results data yet: no impression counts, no customer counts, and nothing to report. Do not invent any.

# Voice guide

Save this in your own notes as voice-guide.md. Fill every line from your words.

- Who I serve:
- What I sell:
- What I never say:
- Topics I own (3-5):
- Tone (how I would text a customer I like):
- Words I use:
- Words I ban:
- Hook patterns from my best pieces:
- Structures I use (story then lesson, list, myth vs fact):
- Proof I am allowed to cite (real numbers, with source and date):
- Lessons (fixes from past edits; these override the guide):

Ask one question at a time. Push back on vague answers. Do not draft until you have 2-3 specific stories or examples in the words of the owner. Suggest talking instead of typing (voice-to-text).

1. What happened in the business this week that a customer would recognize?
   Push back: If the answer is that it was busy, ask what happened on one specific day.
2. What question did a customer ask you more than once?
   Push back: Ask what they asked, as close to their words as you remember.
3. What objection did you hear, and what did you say back?
   Push back: A vague we handle that is not a story. Ask for the sentence you said.
4. Tell one moment: who was there, what they said, and what you did.
   Push back: Do not invent a composite person. If a name is private, use a role and a real detail.
5. What did you have to explain twice?
   Push back: Ask what was confusing the first time.
6. What mistake did you make, and what changed after it?
   Push back: A lesson with no scene is a slogan. Ask where you were when it went wrong.
7. What opinion about your work would you say out loud to a customer you like?
   Push back: Quality matters is not an opinion. Ask what you disagree with.
8. What should the reader do, ask, or stop doing next?
   Push back: One next step, not a list of ten.
9. Which numbers are you allowed to cite? For each one: the number, the source, and the date.
   Push back: If there is no source or no date, the number stays out.
10. What is off limits: private names, a confidential deal, or anything you would not say in public?
   Push back: If you are unsure, leave it out.

Offer 5 hook options, one per pattern, drawn only from the raw notes. The owner picks one. Keep winning hooks in the bank with what happened. Never invent a number to make a hook stronger. Leave the result blank until the owner knows it.

Patterns:
- Pain point: Name a problem the buyer already feels, in the words of the owner.
- Strong opinion: A view the owner actually holds. Do not borrow a hotter take.
- Specific number: Only a real number from the notes of the owner or a linked source, with the date. If you do not have one, skip this pattern.
- Story opener: A concrete moment from the raw notes: a place, a person, or a sentence someone said.
- Myth vs fact: A thing people repeat, then what the owner has actually seen.

Record for each hook you keep:
hook | pattern | piece | date | what happened (only what the owner observed; blank until they know)

1. Anchor piece: The full draft. One clear point, written from the notes of the owner.
2. Short posts: 3-5 posts. One idea each, written for that platform, not a chopped copy of the anchor.
3. Email: The same point, written as a note to one customer.
4. FAQ answer: The question a buyer would type, answered in the words of the owner.
5. Short video script: Spoken, under a minute, from the same notes. No new facts.

Write each piece native to its platform. Test the small versions first. Expand the one that earns replies. Do not invent engagement numbers.

Score each item 0, 1, or 2. Add them up (20 points). Divide by 2 for a score out of 10. Under 8 out of 10, revise. If a fact is missing, ask the owner. Never fill the gap.
Output the score out of 10, the top 3 fixes, and rewritten lines. Never add a fact the owner did not give.

1. Accuracy: Every claim traces back to my notes or a linked source.
2. Numbers: Every number has a source and a date. No invented statistics.
3. No guaranteed outcomes: No guaranteed outcomes: no promised approvals, savings, income, or rankings.
4. Finance-safe wording: Finance-safe wording: use "may" and "can," not "will." No rate, fee, or approval promises. Point readers to a professional for legal, tax, or lending specifics.
5. Plain language: Plain language, short sentences, no jargon or AI filler words.
6. Voice: It sounds like me.
7. Specificity: At least one real story or example.
8. Hook: The first line earns the second.
9. One point: One clear point, one clear next step or question.
10. Platform fit: Right length and format for the platform, with nothing private.

The MCAFax marketing skill

MCAFax ships skills, not prompts. A skill is a set of instructions you add to your own AI assistant once. After that, you ask it in plain words and it follows the system.

The marketing skill has two modes:

• Create: idea to interview to draft to hooks to repurposed pieces.
• Review: paste any draft, yours or the AI's, and get a score out of 10, the top three fixes, and rewritten lines, checked against the list above.

On first run, it helps you build your voice guide from a few pieces you are proud of, or from a short interview. It links back to the Notebook, the guides, the prompts library, and the toolkit. It writes drafts only. You decide what gets published.

Copy the skill, or download the file, on the marketing skill page.

An honest note

This skill is new. We have no results data yet: no impression counts, no customer numbers, nothing to brag about. The numbers the creators above share about their own reach are their own claims. We have linked to where they made them, and they are not a promise of what you will see. We will report what we learn as owners use it.

Sources: Lieberman (podcast, LinkedIn, write-up); Hassid (1, 2); Welsh; Ship 30; Vaynerchuk; Forte; Perell.