This free email course gives you the five lessons that make the difference, one short email a day. They come from my own three publications and learnings along the way.

What’s inside:

• Day 1: The question that decides if your Substack ever makes money, and an AI interview prompt to answer it.

• Day 2: The 90-day window where subscribers actually convert, and what to do inside it.

• Day 3: Why one pillar post beats fifty good ones. Plus a gift: a simplified version of the Notes prompt that took me from 7 to 129 new subscribers a day.

• Day 4: The fear that keeps most Substacks unprofitable, and the test that dissolves it.

• Day 5: The system that runs it all on autopilot, and an offer reserved for course graduates.

Every lesson takes under five minutes to read and comes with a task you can do the same day.

The course is free. Subscribe and day one lands in your inbox tomorrow.

Subscribe now

Already subscribed, or want the whole course right now instead of waiting five days? The complete edition lives below for paid members: all five lessons plus every prompt, in one place you can come back to anytime.

Read more

The issue wasn’t a lack of options. It was the opposite. Without a framework to filter through them, every palette looks fine and nothing feels right.

Professional brand designers solve this with a discovery session. They sit you down, ask specific questions about your business, your personality, your audience, and they narrow the entire universe of design choices down to a handful that actually fit. The session costs a few thousand dollars. The actual filtering process takes about 30 minutes.

I turned that filtering process into a prompt.

What this prompt does

You paste it into Claude or ChatGPT, and it runs you through a guided brand design session. Step by step, one question at a time. It covers:

• Your brand personality and audience

• Color direction (with real examples to react to)

• Design style and aesthetic

• Typography feel

• Layout and visual tone

At the end, it compiles everything into a brand style guide you can actually hand to a designer, use in Canva, or reference every time you create content.

The whole thing takes 15-20 minutes.

The prompt

Copy everything inside the code block below. Paste it into a new conversation with Claude or ChatGPT. Then just answer the questions.

You are a senior brand designer running a discovery session with a new client. Your job is to help them define their complete visual brand identity through conversation.

RULES:
- Ask ONE question at a time. Wait for their answer before moving on.
- Give concrete examples with every question. Don't ask abstract questions like "what vibe do you want?" — show them options and let them react.
- Be warm and conversational, like a designer who's done this 500 times and genuinely enjoys it.
- When showing options, use groups of 3-4. More than that gets overwhelming.
- After each answer, briefly reflect back what you heard before asking the next question. This builds confidence that you're tracking their preferences.
- If they seem unsure, help them eliminate options instead of picking favorites. "What do you definitely NOT want?" is often easier to answer.

Run through these phases in order:

---

PHASE 1: WHO ARE YOU? (3-4 questions)

Get to know their business and brand personality. Ask about:
- What they do and who they serve (in plain language, not marketing speak)
- 3 words their ideal client would use to describe them after working together
- A brand they admire the FEEL of (not necessarily in their industry) — and what specifically they like about it
- Whether their brand voice is more "trusted expert," "encouraging friend," "bold provocateur," or "calm guide" — or some mix

---

PHASE 2: COLOR DIRECTION (4-5 questions)

Guide them toward a color palette. For each question, show specific examples.

Before suggesting palettes, incorporate current color trends so your suggestions feel fresh and relevant — not recycled from 2020. Here's what's trending in 2026:
- Pantone Color of the Year: Cloud Dancer (a serene, billowy white — great for backgrounds and breathing room)
- Industry picks: Transformative Teal, Patina Blue (smoky, complex, versatile)
- Pinterest's trending shades: Persimmon (warm orange-red), Cool Blue, Plum Noir (moody purple), Wasabi (electric chartreuse)
- The big shift: away from cold grays and stark minimalism, toward earthy neutrals (caramel, umber, sienna, clay), smoky jewel tones, and emotionally expressive accent colors
- Hot accents: coral red, deep cobalt blue, electric sage/chartreuse
- The direction: color contrast and emotional expressiveness over safe neutrals. Brands are using variable, dynamic palettes that feel alive rather than corporate.

Weave these trends naturally into your suggestions. Don't list them as trends — just make sure the palettes you propose feel current, not dated.

Start with emotional direction:
- "Colors set a mood before anyone reads a word. Which of these moods fits your brand best?"
  - Option A: Warm and approachable (think: Mailchimp, Headspace — soft corals, caramel tones, warm neutrals)
  - Option B: Bold and energetic (think: Notion's red, Figma's gradient — persimmon, deep cobalt, high contrast)
  - Option C: Clean and premium (think: Apple, Stripe — Cloud Dancer whites, one expressive accent color)
  - Option D: Rich and grounded (think: Patagonia, Aesop — transformative teal, earth tones, smoky jewel tones)

Then narrow with specifics:
- Show 3 specific palette options based on their mood choice. For each palette, list 5 hex codes and describe where each would go (primary, accent, background, text, highlight).
- IMPORTANT: Always show colors visually, not just as hex codes. Generate a simple HTML file or use an artifact/canvas to display actual colored squares for each palette so the client can see the colors side by side. Hex codes alone are meaningless to most people — they need to see the actual colors to react.
- Make sure at least one palette option incorporates a trending 2026 color so the client's brand feels current from day one.
- Ask which one they gravitate toward — or if they'd want to mix elements from two.
- Ask about their "never" colors: "Any colors that feel completely wrong for your brand? Some people have strong reactions to orange, or purple, or pastels."
- Test the palette: "Imagine your website header in [their primary color] with [their accent] buttons and [their background] behind your text. Does that feel right or does something need adjusting?"

---

PHASE 3: DESIGN STYLE (3-4 questions)

Help them define their visual aesthetic:

- "Which of these design styles feels most like home for your brand?"
  - Option A: Minimal and modern — lots of white space, clean lines, simple icons (example: Linear, Notion, Cal.com)
  - Option B: Warm and editorial — feels like a well-designed magazine, photography-forward, elegant typography (example: The Cut, Monocle)
  - Option C: Bold and graphic — strong shapes, chunky fonts, high contrast, not afraid of being loud (example: Gumroad, Cash App)
  - Option D: Hand-crafted and organic — illustrations, textures, natural imperfections, feels human-made (example: Basecamp, Mailchimp's old brand)

- "How do you feel about illustrations vs. photography vs. abstract graphics for your brand imagery?"

- Now show them real visual references they can actually look at and react to. This is critical — people can't describe what they want, but they know it when they see it. Based on ALL their previous answers (mood, colors, style, imagery preference), suggest 5-6 specific real websites or brands. For each one, give the URL and a one-line note on what to look at specifically (e.g., "look at their homepage hero section" or "notice how they use illustrations between text blocks"). Pick examples that are:
  - Relevant to their industry or audience where possible
  - A mix of close matches and slight stretches so they can triangulate
  - Currently live and actually visitable (no dead links)
  - Diverse enough that at least one should click
  Ask them to open 3-4 of these and tell you: "Which one made you think 'yes, that's the vibe'? And was there anything specific — a layout, a color moment, a section — that caught your eye?"
  If none resonate, ask: "What specifically felt off? Too polished? Too playful? Too busy? Too plain?" Use that to suggest 3 more targeted examples.

- "On a scale from 'playful and casual' to 'polished and professional' — where does your brand sit? Think 1 is a handwritten note and 10 is a law firm letterhead."

---

PHASE 4: TYPOGRAPHY FEEL (2-3 questions)

Guide them on font direction (not specific fonts yet — the feel):

- "Typography carries a surprising amount of personality. Which of these feels right for your headlines?"
  - Option A: Strong and geometric — clean, modern, confident (think: headings on Stripe or Apple)
  - Option B: Rounded and friendly — approachable, warm, casual (think: headings on Slack or Mailchimp)
  - Option C: Serif and editorial — sophisticated, timeless, magazine-quality (think: headings on Substack or Medium)
  - Option D: Handwritten or script — personal, creative, organic (think: highlight accents, not full headings)

- "For your body text, do you prefer something that feels more like reading a book (serif) or reading a clean app interface (sans-serif)?"

- Based on their choices, suggest 2-3 specific Google Fonts pairings they can preview for free.
- IMPORTANT: Always show font pairings visually, not just as names. Generate a simple HTML file or use an artifact/canvas that loads the actual Google Fonts and displays each pairing with a sample headline, subheadline, and body paragraph — ideally using content relevant to the client's brand. Seeing "Poppins + Source Serif Pro" in action is completely different from reading the name. Use their chosen brand colors in the preview so it feels real.

---

PHASE 5: COMPILE THE STYLE GUIDE

After all questions are answered, generate their complete brand style guide in this format:

# [Brand Name] — Visual Style Guide

## Brand Personality
[2-3 sentence summary of who they are, how they want to feel, drawn from Phase 1 answers]

## Color Palette
| Role | Color Name | Hex Code | Where to Use |
|------|-----------|----------|-------------|
| Primary | [name] | [#hex] | Main headers, key UI elements |
| Secondary | [name] | [#hex] | Supporting elements, sections |
| Accent | [name] | [#hex] | CTAs, buttons, highlights |
| Background | [name] | [#hex] | Page backgrounds, cards |
| Text | [name] | [#hex] | Body text, paragraphs |

[Show a simple text representation of the palette — five squares or blocks in a row with their hex codes]

## Typography
- Headlines: [Font name] — [why it fits]
- Body text: [Font name] — [why it fits]
- Accent/quotes: [Font name or style] — [when to use it]
- Where to get them: [link to Google Fonts or note if it's a system font]

## Design Style
- Overall aesthetic: [their style in 2-3 descriptors]
- Imagery approach: [illustration / photography / abstract / mixed]
- White space: [minimal / moderate / generous]
- Playful ←→ Professional scale: [their number]/10

## Visual References
[List the 3-5 brands or websites discussed that best represent their direction]

## Quick Rules
- Always do: [3 specific visual rules based on their choices]
- Never do: [3 specific things to avoid based on what they rejected]

## Design Tool Setup
Here are the exact settings to plug into Canva, Figma, or any design tool:
- Brand colors: [list all hex codes in a copy-paste-friendly format]
- Font pairing: [headline font] + [body font]
- Recommended Canva templates that match this style: [suggest 2-3 template search terms]

---

PHASE 6: BUILD YOUR FIRST THREE ASSETS

After delivering the style guide, tell the client: "Your style guide is done. Now let's put it to work. I'm going to generate three things for you right now so you walk away with real assets, not just a document."

Then generate these three things using the style guide you just created:

### 1. Substack Wordmark

Generate a wordmark image for their publication. Specs:
- Exactly 1344 × 256 pixels (21:4 aspect ratio)
- Their publication name in their headline font direction
- Use their primary color for the text, their background color for the background
- Keep it clean. Just the name. No icons, no taglines, no decorations.
- Generate a version with the background color AND a version with a transparent background
- Tell them: "Upload this in Substack under Settings > Publication details > Wordmark."

### 2. Publication Logo

Generate a square publication logo. Specs:
- Square format (1:1)
- Use the first letter or first two letters of their publication name
- Set in their headline font direction, in their accent color, on their background color
- Alternatively, create a single simple icon that represents their brand
- Must be readable when shrunk to 80 × 80 pixels, so keep it simple
- Tell them: "This shows up in Substack recommendations, subscriber inboxes, and the app."

### 3. Header Image Prompt Template

Write them a reusable image generation prompt template they can use for every article header. The template should:
- Bake in their illustration style, color palette, and layout preferences from the session
- Have one clear placeholder where they swap in the article's concept
- Produce headers that all feel like they belong to the same publication
- Include their exact hex codes in the prompt
- Tell them: "Use this template every time you publish. Just change the concept line. Your headers will look consistent without you having to think about design each time."

Ask if they want to adjust any of the three assets before wrapping up.

---

Now begin. Greet the client warmly and ask your first question.

Pro tip

Run this prompt twice. First time for your primary brand. Second time for a sub-brand or specific product line, if you have one. Most solopreneurs need one core palette and one variation. The prompt handles both, and the second run takes half the time because you already know what you like and what you want to avoid.

You can also feed it your existing website URL and ask it to audit your current design against what the session uncovered. The gaps are usually obvious once you have a clear guide to compare against.

If you’re reading this after 2026, add one line to the top of the prompt before you paste it: “Before suggesting palettes, search for the current year’s color trends from Pantone, WGSN, and Pinterest so your suggestions feel current.” The AI will update itself.

If you know someone who’s been “meaning to figure out their brand” for months, send them this. One prompt, one session, done.

Share

Pro tip

Run this prompt twice. First time for your primary brand. Second time for a sub-brand or specific product line, if you have one. Most solopreneurs need one core palette and one variation. The prompt handles both, and the second run takes half the time because you already know what you like and what you want to avoid.

You can also feed it your existing website URL and ask it to audit your current design against what the session uncovered. The gaps are usually obvious once you have a clear guide to compare against.

If you’re reading this after 2026, add one line to the top of the prompt before you paste it: “Before suggesting palettes, search for the current year’s color trends from Pantone, WGSN, and Pinterest so your suggestions feel current.” The AI will update itself.

Want to see what AI can do for you?

If you like you can work with me on that and run the AI Berg Diagnostic on your company.

I’ve done this with 20+ businesses now and the patterns are remarkably consistent. Reply to this email and tell me your company size and industry. I’ll tell you honestly whether I can help. Or book a free call here.

Book a free call with me

🚀 Work with me

AIBergStudio is my AI consultancy. I work with business owners, offices and companies in to figure out where AI fits and build the first system that actually runs.

We start with a free call. I look at your business, find the highest-impact workflow, and tell you exactly what I’d do. If it makes sense to work together, we go from there.

Book a free call with me

A little about me

I live for helping people figure out AI without the hype. When I’m not building workflows, I’m adventuring with my rescue dog Mambo, who has zero interest in AI but very strong opinions about what we eat.

P.s. I publish every Wednesday. Subscribe and it lands in your inbox. Free.

Claudia Faith holds a Master of Science in AI. She's a VC-backed founder who's worked with Fortune Global 500 companies to deploy AI where it actually matters. As a Fractional Chief AI Officer, she handles strategy and implementation, and offers 1:1 coaching for business owners who don't want to be the last in their industry to figure this out. If you want help thinking through your own AI strategy, start here.

I kept putting it off. Writing newsletters feels natural to me, but sales emails are a different thing entirely. You’re asking someone to spend money, and if the sequence is bad, you’ve wasted the launch window.

Most of the advice I found was either too vague (”just be authentic!”) or too complex (47-email funnel maps that assume you have a team of three). Nothing I could actually use in the next 48 hours.

So I tried something I hadn’t seen anyone talk about. Instead of writing the emails myself or following one guru’s framework, I turned Claude into a sales email research assistant. I connected it to my Gmail, pointed it at the best marketers I knew, and let it study their actual sequences.

What came out of that was a mega prompt that writes complete, ready-to-send email sequences based on 6 proven blueprints. I used it for my own launch. Now I’m walking you through how I built it.

Step 1: Connecting Claude to my inbox

Claude can connect directly to your Gmail account. On claude.ai, you go to your integrations settings and authorize Gmail access. In Claude Code (the CLI version I use), it works through something called MCP servers, which are basically plugins that let Claude talk to external services. Either way, once it’s connected, Claude can search your inbox, pull full email threads, and read the actual content.

I’d been subscribed to Ramit Sethi, Amy Porterfield, and Liz Wilcox for a while. That means I had real sales email sequences from some of the best in the business sitting in my inbox. Not blog posts about email strategy. The actual emails, in the order they were sent, with the real subject lines and CTAs.

I told Claude: “Search my Gmail for emails from Ramit Sethi that mention ‘Earnable’ or ‘enrollment.’” Then the same for Amy Porterfield’s Digital Course Academy launch, Liz Wilcox’s email marketing membership, and a few others. Claude pulled the full threads, organized them by date, and laid them out so I could see each sender’s complete sequence from first email to final close.

Then I asked it to analyze all of them side by side. For each sender, it tracked:

• How many emails were in the sequence

• The timing between sends

• What percentage of each email was story vs. pitch

• Where the CTA appeared (first 25%, middle, or P.S.)

• How urgency was built (and whether the deadlines were real)

• The emotional arc across the full sequence

That comparison is where things got interesting. Ramit Sethi starts with 100% value for the first two emails before mentioning any offer. Russell Brunson does the opposite, weaving the product into a story from email one. Both work, but for completely different audiences and price points.

If you don’t have these kinds of emails in your inbox, subscribe to a few marketers whose style you like and wait for their next launch. Or ask colleagues to forward sequences they’ve received. The raw material matters more than the tool you use to analyze it.

Step 2: Researching the frameworks online

My inbox gave me the raw material. But I also wanted the theory behind it. I’d seen what these marketers actually sent. Now I wanted to understand why they structured it that way.

I had Claude do a deep web search on each marketer’s email methodology. Not blog roundups, but their published frameworks, course materials, podcast interviews where they explain their process, and case studies from people who’ve used their systems. For each one, I asked Claude to find the underlying structure: how many emails, what each one is supposed to accomplish, and what triggers the transition from value to pitch.

Six frameworks kept coming up:

1. Frank Kern’s 4-Day Cash Machine. Built for fast sales under $50. Four emails, four days, done.

2. Ramit Sethi’s 5-Day Sequence. Earns trust before asking for anything. Starts with pure value, builds to the pitch gradually.

3. Russell Brunson’s Soap Opera Sequence. The product emerges naturally from a story arc. Each email ends on a cliffhanger.

4. Amy Porterfield’s Webinar Launch Sequence. Designed around a live or recorded event. Multiple touchpoints building to one moment.

5. Liz Wilcox’s Launch Sequence. Personality-driven, casual tone. High frequency, low pressure, works well for smaller lists.

6. Ryan Deiss’s Profit-Logic-Fear Framework. An evergreen structure where urgency builds through emotional stages rather than calendar deadlines.

Each one fits specific conditions. Kern’s is for a flash sale to a warm list. Porterfield’s needs a webinar funnel behind it. Sethi’s works when your audience doesn’t know you well yet. Brunson’s works best when you have a compelling origin story.

Most people pick a framework because their favorite marketer uses it, not because it fits their situation.

Step 3: Turning research into a reusable system

At this point I had two things: a pile of real email sequences from my inbox, and a breakdown of the six frameworks behind them. Good research, but still just research. I needed something I could use again without redoing the whole process.

I told Claude: “Take everything you’ve learned from the inbox analysis and the framework research. Build me a structured mega prompt that can write sales email sequences for any offer, any price point, any list size. Not just mine.”

By “mega prompt” I mean a single, self-contained prompt I could paste into any Claude conversation and have it walk someone through the full process. No context from the research session needed. Just the prompt, and it handles the rest.

Here’s what it does:

It starts with discovery. Asks you 10 questions across three rounds: what you’re selling, the price point, list size, timeline. Then it goes deeper into your relationship with your audience, whether you have testimonials, and what you’ve sold before.

Based on your answers, it recommends one of the six blueprints. A $47 product with a 3-day window gets Kern’s Cash Machine. A $997 course with a webinar gets Porterfield’s sequence. It tells you why it picked that one, so you can push back if it feels wrong.

Then it collects the specifics: product name, core promise, main objections, your personal story, bonuses, guarantees. Everything the emails will need.

After that, it writes the actual emails. Complete messages with subject lines (including an A/B test option), preview text, full body copy, and notes on timing and segmentation. Each email follows the blueprint’s structure but uses your details and voice.

Last, it gives you a delivery plan. Sending schedule with times, segmentation instructions, benchmarks so you know what good open and click rates look like, and a plan for people who didn’t buy.

What actually changed the output

I didn’t expect the output to be this different from what you get when you ask Claude to “write me some sales emails.” But it was.

A bare prompt gives you bare emails. They sound like every other launch sequence. The research changed the output because Claude wasn’t guessing. It had studied real sequences from people who’ve made millions from email, and it knew why specific structures work for specific situations.

The emails it wrote for my AI Business Build-Out sounded like me. They used my actual story, tied urgency to a real deadline, and placed CTAs where the research showed they’d perform best.

Three things made the biggest difference:

• CTA placement. High-performing sequences put a CTA in the first 25% of the email, again in the middle, and one more in the P.S. Most people save the CTA for the end, which means anyone who stops reading halfway never sees it.

• The “you” ratio. Across every sequence Claude analyzed, “you” appeared 3 to 5 times more often than “I.” Sounds obvious. But sit down and write a sales email and watch how fast it becomes about your product instead of the reader’s problem.

• Deadline authenticity. Every framework that used urgency did it with real reasons: a live cohort start date, a genuine bonus expiration, a price increase that was actually happening. The prompt asks for real deadlines and real reasons. It won’t invent scarcity for you.

You can build this for anything

The email prompt took about 48 hours to build, research included. The process is what I want you to take away, maybe more than the prompt itself.

Connect AI to your real data. Your inbox, your analytics, your existing content. Have it research proven frameworks from experts who’ve done this at scale. Let it find patterns across multiple approaches. Then turn those patterns into a structured system you can reuse.

I’ve been doing this for other parts of my business too. I ran the same process on Substack Notes that went viral and pulled out the structural patterns. I did it with landing pages. I’m working on one for article formats right now.

AI doesn’t replace the judgment calls. You still pick the framework, decide which details matter, and edit the emails until they feel right. But the research and pattern-finding that would take weeks to do by hand takes an afternoon this way.

Want the prompt and start selling? Get it here

Get the email sales skill here

Want to see what AI can do for you?

If you like you can work with me on that and run the AI Berg Diagnostic on your company.

I’ve done this with 20+ businesses now and the patterns are remarkably consistent. Reply to this email and tell me your company size and industry. I’ll tell you honestly whether I can help. Or book a free call here.

Book a free call with me

🚀 Work with me

AIBergStudio is my AI consultancy. I work with business owners, offices and companies in to figure out where AI fits and build the first system that actually runs.

We start with a free call. I look at your business, find the highest-impact workflow, and tell you exactly what I’d do. If it makes sense to work together, we go from there.

Book a free call with me

A little about me

I live for helping people figure out AI without the hype. When I’m not building workflows, I’m adventuring with my rescue dog Mambo, who has zero interest in AI but very strong opinions about what we eat.

P.s. I publish every Wednesday. Subscribe and it lands in your inbox. Free.

Claudia Faith holds a Master of Science in AI. She's a VC-backed founder who's worked with Fortune Global 500 companies to deploy AI where it actually matters. As a Fractional Chief AI Officer, she handles strategy and implementation, and offers 1:1 coaching for business owners who don't want to be the last in their industry to figure this out. If you want help thinking through your own AI strategy, start here.

So let me be specific about what mine actually is: an automated workflow. It pulls data, reads files, runs scripts, and writes output to a folder. Four steps chained together so I don’t have to run them manually every Sunday.

Three months ago I shared the Notes prompt that changed how I show up on Substack. It went viral. 132 likes, hundreds of saves, and a flood of DMs from writers who started using it the same week.

That prompt still works. But I stopped using it the way I shared it.

Somewhere between February and now, I turned that prompt into something bigger. I gave it memory. I connected it to my analytics. I let it learn from what actually performs instead of guessing from a formula sheet. And I wrapped it in this agent (read: automated workflow 🙃) that handles everything from pulling last week’s data to scheduling next week’s Notes in about 20 minutes on a Sunday morning.

Why the Old Prompt Stopped Being Enough

My Notes prompt from February did one thing well: it followed viral formulas and wrote in my voice. But every week I was still doing four manual steps around it.

• Step 1: I’d scroll through my Substack dashboard, mentally noting which Notes performed and which didn’t.

• Step 2: I’d figure out what topics to cover next week based on gut feel and whatever was trending in my niche.

• Step 3: I’d paste the prompt into Claude, wait, then edit 21 Notes by hand.

• Step 4: I’d copy each Note into my scheduler one at a time.

That was still 90 minutes. Some weeks closer to two hours if I second-guessed the topic mix. For someone running two newsletters, a consulting business, and a coaching program, two hours of Notes production every Sunday felt like a tax I kept paying.

The prompt was doing 30% of the work. I was still doing the other 70% manually.

What Changed: the Agent Layer

An agent is a prompt that can do things. It reads files, runs scripts, pulls data from a browser, writes output to specific folders, and chains those steps together without you orchestrating each one. You describe the outcome. The agent figures out the execution.

My Notes agent has four steps, and it runs all of them in sequence from a single command.

Step 1: Pull Performance Data

The agent opens my Substack dashboard through Puppeteer (a browser automation tool connected to Claude Code via MCP). It reads the last 7 days of Notes: likes, comments, restacks, and subscriber conversions for each one.

Then it summarizes the patterns. Which formats pulled the most engagement. Which topics converted subscribers. Which Notes flopped.

This is the part most creators skip entirely. They publish, move on, and base next week’s strategy on memory. Memory is unreliable. The data isn’t.

One insight from a recent pull: a Note with 34 likes converted more subscribers than one with 266 likes. The short, specific, actionable Note outperformed the one that went semi-viral. I would never have caught that by scrolling casually. The agent caught it in 8 seconds.

Step 2: Generate the Topic Mix

Based on the performance summary, the agent proposes 21 topics for the coming week. Three Notes per day, seven days. It pulls from my content pillars (AI tools, prompting, creator workflows, building with AI) and weights the mix toward whatever performed best last week.

It also cross-references my recent articles. If I published a piece about creator AI stacks on Tuesday, the agent pulls the most surprising number, the most actionable tactic, and the core contrarian insight from that article and turns each into a Note. Repurposing happens automatically.

I review the topic list, swap maybe 2 or 3, and approve.

Step 3: Write 21 Notes in My Voice

This is where the original viral prompt lives, but upgraded. The agent reads my voice profile (sentence rhythm, transition style, banned phrases, 18 AI writing fingerprints to avoid) and my Notes formula library (7 proven formats ranked by engagement). It writes all 21 Notes in one pass.

The mix is intentional. At least 2 micro-Notes under 100 characters. At least 3 in the 100 to 250 range. A few longer ones in the 400 to 600 range. No two consecutive Notes use the same format. This variety pattern came directly from the performance data: accounts that alternate between ultra-short and medium-length Notes get 40% more profile visits than those posting the same format every time.

Every Note hits the checklist before output: zero rhetorical questions, zero generic motivation, all numbers specific and exact, no hashtags, no emojis in the text. The agent enforces rules I used to enforce manually by re-reading everything twice.

Step 4: Export and Schedule

The agent writes all 21 Notes to a CSV file with columns for content, date, and time. I upload that CSV to my batch scheduler (a tool I built that lets you schedule Substack Notes in bulk, since Substack only supports scheduling one at a time through the interface).

Sunday morning to Sunday upload: 20 minutes. Sometimes 25 if I rewrite a few Notes by hand. The rest of the week, Notes post automatically at the times the data says my audience is most active.

The Prompt (Copy-Paste Ready)

Here’s the core prompt I feed the agent. This is the instruction set for Step 3, the actual Note generation. Steps 1, 2, and 4 are handled by the agent’s tool chain (Puppeteer for data, scripts for CSV export).

You are writing Substack Notes for [YOUR NEWSLETTER NAME] by [YOUR NAME].

ABOUT THIS NEWSLETTER:
[1-2 sentences describing your newsletter's focus and who reads it. Example: "This newsletter helps creators and solopreneurs implement AI tools and build online businesses. Readers are AI-curious but action-oriented, they want workflows not theory."]

VOICE RULES:
- Write like you're texting a smart friend who runs a business
- Use "you" and "I" freely
- Mix sentence lengths. Short punchy lines between longer explanations.
- Use specific numbers. 884 not "almost 900." 23 not "over 20."
- [Add 3-5 phrases you use naturally. Mine: "Here's exactly what I do," "The difference is everything," "This is where most people stop."]

Sentence Structure (no AI language):
No "It's not X, it's Y" / "Not because X, but because Y" negative parallelism
No "Not X. Not Y. Just Z." dramatic countdowns
No fake "From X to Y" ranges where X and Y aren't on a real scale
No self-posed rhetorical questions answered immediately ("The result? Devastating.")
No filler transitions ("It's worth noting" / "Importantly" / "Notably")
No tricolon abuse (three parallel constructions back to back)
No anaphora abuse (repeating the same sentence opening multiple times)
No tacking "-ing" phrases for shallow analysis ("highlighting its importance", "reflecting broader trends")

Tone:
No grandiose stakes inflation (everything is civilization-changing)
No invented compound labels ("the supervision paradox", "the acceleration trap")
No "Here's the kicker" false suspense transitions
No "Think of it as..." patronizing analogies
No "The truth is simple" asserting obviousness instead of proving it
No "Experts argue..." vague attributions without named sources

Word Choice & Formatting:
No ornate words where simple ones work ("tapestry", "landscape", "paradigm", "synergy", "ecosystem")
No excessive short punchy fragments as standalone paragraphs
No listicles disguised as prose ("The first step is... The second is...")
No unicode arrows or smart quotes

NOTE FORMATS (use at least 5 of these, never repeat the same format twice in a row):

1. Imperative Command: Open with a direct command. Specific steps. Action close.
   Example: "Grab a notebook. Write exactly what you want your business to look like in 12 months..."

2. Bold Claim + Reality Check: Start with a contrarian statement. Expand on what really happens.
   Example: "You don't become consistent by planning harder. You become consistent by removing the decisions that drain you before you start..."

3. Definitional Framework: Define 2-3 levels of the same concept. Reveal the deeper truth.
   Example: "A tool is something you try once. A workflow is something you run weekly. A system is something that runs without you..."

4. The Confession: What went wrong. The messy middle with real numbers. The unexpected outcome.
   Example: "I spent 4 hours last Tuesday writing 6 Notes. Three of them got under 10 likes. The one I almost deleted got 266..."

5. Quick Tip: "Quick tip:" opener, one specific action, done.
   Example: "Quick tip: schedule your Notes at 7:30 AM, 12:15 PM, and 6:45 PM. Those three windows get 2x the impressions of random posting times on my account."

6. Ultra-Short Truth: 1-2 sentences max. A universal observation that hits immediately.
   Example: "The creator posting 3x a day with a system will always outperform the one posting once a week with 'inspiration.'"

7. Data Truth: Lead with a surprising number. Explain why it matters. Challenge an assumption.
   Example: "34 likes, 12 new subscribers. 266 likes, 3 new subscribers. Virality and conversion are not the same metric."

STRUCTURE:
- Generate exactly 21 Notes for 7 days (3 per day)
- No two consecutive Notes use the same format
- Length distribution:
  - 2+ micro Notes (under 100 characters)
  - 3+ short Notes (100-250 characters)
  - 10+ medium Notes (250-450 characters)
  - Up to 4 full Notes (450-600 characters)

TOPIC MIX:
- Cover these content pillars across the week: [LIST YOUR 3-5 MAIN TOPICS. Mine: AI tools, prompting/strategy, creator workflows, building with AI]
- Weight toward whatever you performed best in last week. Here's my data: [PASTE YOUR LAST WEEK'S TOP 3 NOTES AND THEIR LIKE/RESTACK COUNTS]
- Include 3-5 Notes that repurpose insights from this article: [PASTE YOUR MOST RECENT ARTICLE OR KEY PARAGRAPHS]

OUTPUT:
For each Note, include: the Note text, which format you used, and the character count. Group them by day (Day 1 through Day 7, 3 per day).

How to make this yours in 15 minutes:

Fill in the brackets. The voice rules and banned list matter most. Read 10 of your published posts and pull out phrases you actually use, and phrases you’d never write. That contrast is what makes AI output sound like you instead of like everyone else. The Note formats above work across niches. I’ve tested all 7 on my account. Use them as-is until you have enough data to know which ones your audience responds to, then cut the weak ones and double down.

The performance data section is optional for your first batch. Skip it, generate 21 Notes, post them for a week, then come back and paste your top performers into the prompt before generating the next batch. That’s how the feedback loop starts.

The System Diagram

Here’s how all four steps connect:

Puppeteer → Performance Summary → Topic Mix → Notes Generation (Voice Profile + Formula Library) → CSV → Batch Scheduler → Published Notes

Each arrow is automated. The only human touchpoints are reviewing the topic list (Step 2) and reviewing the final Notes (Step 3). Everything else runs.

The performance data from this week feeds back into next week’s generation. The agent gets better every cycle because it’s learning from real engagement numbers, not static formulas.

What This Actually Means for Your Business

I share the technical workflow because the request in my DMs is always “show me how.” But the workflow is the smallest part of what’s happening here.

The bigger shift: I freed up 90 minutes every Sunday and redirected that time toward the work that actually grows revenue. Coaching calls. Building offers. Closing clients.

Last week I spent 20 minutes on Notes production and 3 hours on a coaching session that paid $1,200. The week before, I used the freed-up time to restructure my AI Business Builder program from a group format to fully 1:1, which made it dramatically more valuable for the people who sign up.

That restructuring happened because I had the space to think about it. When your content production eats your entire Sunday, you never get around to the strategic work that compounds.

Every creator I work with in my coaching program hits this wall. They’re spending 60% of their time producing content and 40% on the business that content is supposed to support. The ratio should be flipped. AI agents are how you flip it.

What You Get in the AI Business Builder

I sold a spot two days ago. One spot left for June.

Here’s what happens in four weeks of private 1:1 calls:

You walk out with a packaged offer you can describe in one sentence and price at $10K, because we stress-test your positioning live until it holds up. A sales page that’s published and ready to send to prospects the day we finish. 2 to 3 AI automations handling your onboarding, delivery, or follow-up, so taking on a second client doesn’t mean doubling your hours. And Drift, my AI writing system ($1,399 value), generating content in your voice whenever you need to promote your offer.

My client, Julia M. landed a $10K client within a month of finishing the program with me. One new client of her pays for the program four times over.

Start Building This Week

You don’t need my exact stack to get the core benefit. Here’s the minimum version:

• If you have Claude Code: Create a CLAUDE.md file with your voice rules, your banned phrases, and your Notes format preferences. Ask Claude to generate 7 Notes for the week. Review, edit, post. That’s your v1. Add the performance data loop when you’re ready.

• If you don’t have Claude Code: Take the prompt above, strip out the file references, paste your voice rules directly into the prompt, and use it in any Claude conversation. You lose the automation layer but keep the voice matching and format variety.

• If you want the full system built for you: That’s the AI Business Builder. I sold a spot on Sunday. One left for June. We build your automations together, on your business, producing real output before week four. → Save your spot

🚀 Work with me

AIBergStudio is my AI consultancy. I work with business owners, offices and companies in to figure out where AI fits and build the first system that actually runs.

We start with a free call. I look at your business, find the highest-impact workflow, and tell you exactly what I’d do. If it makes sense to work together, we go from there.

Book a free call with me

A little about me

I live for helping people figure out AI without the hype. When I’m not building workflows, I’m adventuring with my rescue dog Mambo, who has zero interest in AI but very strong opinions about what we eat.

P.s. I publish every Wednesday. Subscribe and it lands in your inbox. Free.

Claudia Faith holds a Master of Science in AI. She's a VC-backed founder who's worked with Fortune Global 500 companies to deploy AI where it actually matters. As a Fractional Chief AI Officer, she handles strategy and implementation, and offers 1:1 coaching for business owners who don't want to be the last in their industry to figure this out. If you want help thinking through your own AI strategy, start here.