S3: Take control of your voice

English edition. S1–S6 are available in English; S7–S9 are published in Japanese.

The right to be heard was yours from the start.

On this page

• What you get from this skill
• Why it matters
• Copy-paste path
• The core: take your voice back
• Choose the media structure
• Step 1: make the action log fill itself
• Step 2: pick what to show, and run the cycle
• Step 3: build landing pages at scale
• Step 4: landing pages that actually sell
• Are you done?
• Practice: pitch a new project end to end

What you get from this skill

Not tricks for a million views. A content system for the business you actually have: your daily work becomes material, and a loop of design → publish → measure → improve keeps running in content/. Plus a pipeline of landing pages that turns social attention into inquiries.

S1 gave you your own map. S2 gave you a target. S3 is where both become a voice that reaches people.

Why it matters

The algorithm quietly takes your voice

You’re free to say what you know. But on social platforms, however good your work is, if the algorithm decides not to show it, nobody sees it. Bend to what the algorithm likes, and you drift further from what you actually do. Anyone who’s hired a social media consultant knows the feeling: post more, chase trends, find something viral — until you forget why you were posting.

The more you pander, the less of you is left

Copying the format that’s “working,” chasing whatever’s trending, making easy content outside your field — the audience you collect is an audience for that content, not customers for your business. That’s why follower counts rise while sales don’t.

---

Copy-paste path: the shortest route through S3

Paste these into your AI agent one at a time, in order. Replace everything in 【brackets】 first, and read what comes back before moving on. The reasoning is in Steps 1–4 below.

① Add logging rules (Step 1)

Add these rules to the rules file (AGENTS.md):
- When a task I asked for is finished, add one line to today's section of actionlog/YYYY-MM.md
- Record: time / what I asked / what you made or changed (file paths) / result and next step
- One task = one line. Never rewrite earlier lines
- Always check the date from the system clock
- When you save a journal entry or voice memo into inbox/, extract only "what I did" and add it to the action log as well

② Set a publishing policy (Step 2)

Create ontology/concepts/publishing-policy.md.
Ask me one question at a time to decide whether each of these is Public / Limited / Internal / Secret:
my face / business name / numbers like revenue and costs / staff / customers / suppliers / screens I'm working on

③ Weekly: pick what to show (Step 2 · every week)

Read this week's actionlog/ and the journal entries in inbox/, and pick 10 pieces of work or decisions the business should show publicly.
Read ontology/concepts/publishing-policy.md first. Drop anything that isn't Public, or say how it could become Public (anonymize, aggregate, get permission).
For each: what to show / the impression it gives / who it would reach / the angle / risks if published.

④ Design one piece (Step 2)

I choose 【number】 from the list.
Create content/【YYYY-MM-DD_title】/brief.md and design it with four parts: Desire, Means, Process, Result.
The subject is always the person on screen (me).
Then check it against the four principles: a start a five-year-old gets / complete in one piece / no clowning / the desire belongs to the person on screen.

⑤ Bring the results back (Step 2 · one week after publishing)

This screenshot shows the insights for 【post name】 one week after publishing.
Read the numbers and add them to "Results after publishing" in content/【folder】/brief.md.
Write "not available" for anything you can't read.

⑥ Build the offer list (Step 3)

Read research/analysis/entry-strategy.md and action-plan.md.
Break what we offer into units that can each be pitched on a single landing page, and create content/offer-list.md.
For each: who the page is for / what it offers / the one action we want visitors to take

⑦ Write a landing-page brief with the eight elements (Step 4)

Write a landing-page brief for "【offer】" from content/offer-list.md.
Eight elements: above the fold (3 headline options, a photo that proves we're real, the call to action, a proof point) / the pain / the solution / the benefits / why us (comparison table from research/competitors/) / customer voices and results / the offer (3 bonus ideas we can make with AI) / FAQ and how it works
Use only real results and real testimonials — never invent any. The main image is a real photo of our work. Design for phones first.

⑧ Build and publish (Step 3)

Build the landing page from this brief. Don't add anything that isn't in the brief.
Every page gets meta tags (title, description, Open Graph) and the GA4 tracking tag.

Review it, write what you changed and why back into the brief, then say “publish.”

---

The core: take your voice back

The goal of S3 is to use the algorithm only as much as you need, and get your voice to the people you actually want to reach. You can’t ignore the algorithm — it’s part of every platform. You don’t have to serve it either.

The formula

Material (an ordinary scene from inside your field)
  ×
Framing (narration, camera angle, the way you show your point of view)
  =
Content where the point of view is interesting — not the stunt

The most important rule: don’t clown around. Shock and gimmicks attract people who want shock and gimmicks. The people you want — those looking for someone with a real point of view — want perspective, not stimulation. Your ordinary work, decisions, and scenes from the field are the content.

So the three things that matter most:

1. Reach the most carefully chosen audience you can
2. Show the most important thing that came out of your actual work
3. Show it in a way that leads to an action

[No]
Copy a trending format and post "something that'll do numbers" outside your field
→ you collect people who want stimulation → followers up, customers flat

[Yes]
Show today's real decision: "I reviewed food cost and cut one dish from the menu"
→ it reaches people who want expertise → "I'd like to talk to this person"

”Show internal stuff?” — selectively. Build in Public works.

Showing how your business really runs — progress, decisions, even failures — is called Build in Public. Behind it is the process economy: people value not just the finished product but the process of making it. When products and prices look alike, being able to see how the people inside think becomes a reason to trust. Saying you reviewed your food cost — without the exact numbers or supplier names — shows expertise and honesty.

But showing everything is not Build in Public. Customer data, contracts, open negotiations, personal information, credentials, and your real competitive edge put people and the business at risk. Sort information into four levels and record the rules in your publishing policy:

Level	Who sees it	Examples	Rule
Public	anyone	published work, how you do things, the reasoning behind decisions, lessons from mistakes	respects others’ rights and permissions; shows real expertise
Limited	chosen people: customers, members, partners	detailed case studies, course material, partner updates	decide who, why, and whether it may be reshared
Internal	you and your team	detailed costs, undecided matters, HR, live negotiations	needed for work; no good reason to publish
Secret	the minimum named people	passwords, credentials, personal data, trade secrets, legally protected info	never on social media; never casually into outside AI tools

---

Choose the media structure

You’re not building one social account. You’re building three layers:

Social posts or ads (how people find you)
     ↓
A promotional web page (how you think, why you're worth it, how to buy)
     ↓
An inquiry (book, ask, buy)

Social media is only the entrance. What people look at next is a page that shows how your company thinks and works — the place where your Build-in-Public story lives in full, not dependent on any one platform. Keep your copy, images, and settings in your own files and backups, and you can move hosts anytime.

And you can run several of these at once. The brief-first building you learned in S1 lets you generate landing pages whenever the offer, audience, or region genuinely differs. Work that used to go to an agency now happens on your own desk.

Tip: get one pipeline working end to end before multiplying. Run the content loop (Steps 1–2), get one path from post → page → inquiry working, then build more pages (Step 3). Pages nobody visits are just maintenance.

---

Step 1: make the action log fill itself

Your best weapon on business social media is data about what you actually do every day. Not stories, not ideals — what happened at work today. You started actionlog/ in S1. But logging “on purpose” doesn’t last. Step 1 makes the log fill itself.

① Add logging to the rules file

In S1 you put a rules file (AGENTS.md, or CLAUDE.md for Claude Code) in your workspace, which the agent reads at the start of every task. Now add logging rules (prompt ①). From then on, every time you have the AI draft a document, research something, fix a web page, or edit text, a line lands in your log automatically. Working with AI becomes your activity record.

Tip: rules are where repeated instructions get promoted. Don’t try to write the perfect rules file. Whenever you notice “I keep telling it this,” promote it to one line.

Tip: short and few. Dozens of rules get ignored — by agents as by new hires. Keep only what must always apply; move procedures into separate files and say “for X, read X.md.”

② Use AI as your daily assistant

“Organize today’s priorities.” “What do you think of this decision?” “Draft a reply to this email.” “Summarize what I did this week.” The more you use AI for ordinary work, the more the rule from ① turns each exchange into a log entry — without you trying.

③ Capture what happens outside the AI

Rules only see work done with the agent. Your real work — making, selling, serving customers, meetings, site visits — happens elsewhere. Your calendar fills that gap: record what you actually did, not what you planned.

Read my Google Calendar for 【yesterday / this week】 and save what I actually did to inbox/【YYYY-MM-DD】_worklog.md.
Leave out anything that was only planned and didn't happen.
Mark anything unclear as "unverified."

As of October 2026: many AI agents can connect to Google Calendar, Gmail, and similar services directly through connectors (MCP and similar integrations). If yours can’t, you can export the calendar to Markdown with Google Apps Script or an automation tool like Zapier or Make — ask the AI how. And if setting up a connection is where you’d stall, start with a weekly screenshot of your calendar: “turn this week into a work log and save it in inbox/.” Automate once the loop is running.

The other door is your evening voice journal from S1. With the last rule in ①, talking through your day before bed saves the journal to inbox/ and adds the day’s actions to the log. Calendar, journal, AI work — all three now flow into actionlog/ without effort.

For reference: the log format the author actually uses. With several AI agents in parallel, two columns are added — who did it, and roughly how many tokens it cost:

## 2026-10-12 (Mon)

| Time | Agent | Type | Business | What | object_id | Tokens | Result / note |
|------|-------|------|----------|------|-----------|--------|---------------|
| 09:00-10:00 | (me) | Meeting | Consulting | Monthly strategy call with a client | `client-id` | — | Agreed two initiatives; next call 11/9 |
| Morning | Claude | Documents | Framework | Revised the S3 page | `framework` | ~30k | Added tips and a lesson plan; not deployed |
| Afternoon | Codex | Admin | All | Sorted the inbox | `automation` | ~12k | 2 action items to the to-do list, 10 promos archived |

The rules that keep a shared log trustworthy once several agents write to it: one task = one line, append-only (never rewrite earlier lines or other agents’ lines), and no future dates (always check the system clock). Start with the S1 format and add columns as you need them.

---

Step 2: pick what to show, and run the cycle

① Once a week, let AI pick what’s worth showing

Don’t hunt for ideas on instinct. Once a week, have the AI read the activity that piled up in Step 1 and pick the work the business should show (prompt ③). An outside reader sees what you’ve stopped noticing.

It’s not only “what’s interesting.” Ask what impression of your business each piece would leave. A restaurant proud of its knife work might show not just the finished plate but the daily knife care — which says we look after our tools, we’re precise, we care about the prep nobody sees.

Tip: decide up front what’s off-limits. Before you start, place your face, business name, numbers, staff, and customers into Public / Limited / Internal / Secret, and save it as publishing-policy.md (prompt ②). Have the AI read it every week, and things you can’t show never reach the shortlist. Widening the policy later is easy; something posted can’t be taken back.

Tip: “I have nothing to show” means you haven’t looked. Ask: “In my log, what would surprise someone outside my industry?”

Tip: put the weekly pick on the calendar. Same day, same time. Once it works, save the request as a file — or as a skill (a saved procedure your agent can run by name), so “do the weekly pick” is all you need to say.

The same work writes your internal report. If you work inside a company, “read the week and report what matters” is exactly a weekly report to your manager — only the reader changes:

Read this week's action log and journal and write a weekly report for my manager:
- What I finished, and results
- What's in progress, and next week's plan
- Decisions or help I need
- Any results I can report in numbers
Save as YYYY-MM-DD_weekly-report.md.

Combine both into one weekly routine and content stops being extra work — it falls out of reporting you’d do anyway.

Let AI check the risks too. Add to the weekly request: “For each candidate, check whether it identifies a customer or supplier, leaks anything, could be read badly by a critical reader, or breaks the publishing policy.” If you work for a company, convert your social media guidelines and confidentiality policy to Markdown and have the AI read them too. The final call is always yours — once posted, it’s out.

② Design it, publish it, bring the results back

Choose one candidate and design it before you film or write. Without design, the viewer ends up wondering what they’re watching. Use four parts — written the day before at the latest — and keep the subject fixed on the person on screen:

Part	Check	Example
Desire	Is it clear what the person on screen wants? Once the viewer knows, they relax and watch	”Today I want to decide prep quantities with AI, not gut feel”
Means	Is it clear how they’ll do it?	”Give the AI bookings, weather, weekday, and past sales”
Process	Where’s the interesting part, and where’s the struggle?	”The AI disagrees with the chef’s instinct”
Result	Where does it land? (Hint at it at the start if you can)	“Less waste and nothing sells out”

Brief (template)

Created:
Publish on:
Platform: □ Reels  □ Stories  □ Carousel  □ Short video  □ Article  □ Other

[Desire]  By the end of this, I ______.
[Means]   (2–3, everyday tools only — this isn't a product demo)
[Process] 1. (right after the start)  2. (middle)  3. (climax)
[Result]  One sentence + what proves it on screen: □ number □ before/after □ caption □ the real thing

Keep each piece in its own folder: content/2026-10-20_ai-prep-quantities/ with brief.md, caption.md, and img/ — the S1 rule “one project, one folder,” applied to content.

Tip: choose platforms from where your target already is. Not from what’s trending. Picture a few real people like your S2 target: what app do they open every evening?

The 45–60 second short-video shape

Seconds	Content
0–3	State the desire (“Today I’m going to decide…“)
3–40	Process ①②③ (the means just appear on screen)
40–60	The result (a number or one line)

Check four principles before you shoot

	Principle	Meaning
A	A start a five-year-old gets	In three seconds: who wants what. No jargon, no preamble
B	Complete in one piece	Makes sense without the last episode or a follow
C	No clowning — everyday life, new angle	Your real work. No acting, stunts, or unrelated jokes
D	The desire belongs to the person on screen	Not a lecture to the viewer — what you want to achieve

Tip: when in doubt, go back to your normal day. “Would I be doing this tomorrow anyway?” If not, it’s acting.

Bring the results back. About a week after publishing, add views, saves, comments, profile visits, inquiries, and real-world reactions to the same brief. Platform insights are hard to export automatically — screenshot the insights and give the image to the agent (prompt ⑤). Screenshots also beat explanations whenever you’re stuck in a settings screen: “This is my screen. What do I click next?”

Tip: views aren’t the number that matters. For business content, watch saves, profile visits, and inquiries/visits. Views react to stimulation; those three react to trust and action. Low views with an inquiry = success. High views with nothing = failure.

Tip: put the result check on the calendar the day you publish. A brief without results is a quiz without an answer key.

What to ask once briefs pile up

- "Read the last 10 briefs. Where did the result drift from the desire?"
- "What's the difference between the pieces with an interesting point of view and the ones without?"
- "What do posts with high views but no inquiries have in common?"
- "From the most-saved posts, what should I make next?"

Finally, content must lead to an action. For someone to visit or buy, two things must come across: the appeal, and that they can actually use it — where, who it’s for, when, how to book. Appeal without access doesn’t convert; access without appeal gives no reason to move.

---

Step 3: build landing pages at scale

Now build the receiver: lots of landing pages, and a pipeline from posts to pages to inquiries, using GitHub, Cloudflare Pages, and your rules file.

① Decide what to pitch first

From your S2 entry strategy, break your offers into units that fit one landing page — one page, one offer. A page that pitches several things is for nobody (prompt ⑥). You don’t invent offers; they’re already in the business:

Business	One page per…
Real estate	property — every listing gets its own page with photos, floor plan, neighborhood, and the agent’s take, instead of a line on a portal
Education	class or course — what’s covered, outcomes, the teacher’s approach — far more than a brochure line
Restaurant	signature dish or seasonal event — the prep and the story behind it
Events	audience — fans of a performer, first-timers, corporate sponsors each get their own page

When building a page cost a lot, all this was crammed into one website or a platform listing. With the cost near zero, you can pitch at a granularity that was never possible.

② Polish the brief, build, then trigger publishing yourself

The S1 brief-first pattern, but building is never automatic:

1. Polish the brief until nothing bothers you. Page quality is decided here.
2. “Build the landing page from this brief. Don’t add anything that isn’t in the brief.”
3. Review, write fixes back into the brief, then say “publish.” A push to GitHub, and Cloudflare Pages updates the live site.

GitHub stores your code; Cloudflare Pages serves it as a website (both have free tiers). Link them once, with the agent guiding you through screenshots, and every push publishes automatically. Then add to the rules file:

## Building and publishing landing pages
- Build a page only when I say "build it from the brief" (a finished brief is not a request to build)
- Never add anything that isn't in the brief
- Every page gets meta tags (title, description, Open Graph) and the GA4 tag
- When I say "publish," push to GitHub (Cloudflare Pages publishes automatically)

Automatic: only the plumbing (publishing after a push) and the tags. You always pull the trigger for building and publishing. Adding meta tags and analytics by rule matters — retrofitting dozens of pages by hand never happens.

Tip: developer vocabulary just works. Deploy (publish to production), push (send changes to GitHub), kick off (start a process) — agents understand them, and your instructions get shorter.

Tip: never mix secrets into a public folder. Each landing page is its own project outside myproject, with its own repository. No financial data, customer lists, or Ontology files.

Tip: keep the repository private. Only the pages Cloudflare serves are public.

③ Point posts at pages

Work one page at a time: have the agent read a landing-page brief, pick the best-fitting material from the action log, and design a post for it.

Read content/【folder】/brief.md (a landing-page brief).
From the recent action log, pick the material that best fits this page's offer, and design a post for it:
- the material and why it fits
- Desire / Means / Process / Result
- the line in the caption and profile that sends people to this page

Every post now knows which page it feeds. GA4 on every page shows which source sent people where, and where they stopped.

Tip: add UTM parameters to every link. Analytics records only “came from Instagram,” not which post. Add ?utm_source=instagram&utm_medium=social&utm_campaign=folder-name — using the content folder name as the campaign makes it easy to match results to briefs. Add a rule so the agent always includes them.

---

Step 4: landing pages that actually sell

The first rule: prove you’re real

Visitors have seen thousands of AI-generated pages. Pretty pages with nothing behind them are closed in seconds. Above the fold, show that a real business stands behind this: real photos of your place and your work, you and your team, a real address and hours. Now that anyone can generate a plausible page, being demonstrably real is the differentiator — and the material is already in your action log and content folders. Use your own photos, not AI images, for the main visual.

Don’t obsess over design

Unless design is your business, the substance is your responsibility, not the polish. Put your effort into the brief — who, what, and what you want them to do — and only fix design that’s genuinely off. Heavy design presets and add-on “skills” go stale fast as agents’ default design keeps improving; ride the rising baseline.

The eight elements

1. Above the fold — a headline that says whose problem you solve and what future they get; a real image; a clear call to action; one proof point
2. The pain — the frustrations your target lives with, specifically (“that’s me”)
3. The solution — what you offer, plainly
4. The benefits — the better future, not the spec sheet
5. Why us — concrete reasons; a comparison table straight from your S2 research/competitors/
6. Customer voices and results — real reviews, before/after, real numbers
7. The offer — first-time price, guarantee, free consultation, a bonus. With AI, bonuses (checklists, diagnostic sheets, short guides) cost almost nothing to make, so every offer can have its own
8. FAQ and how it works — remove the last doubts; three or four steps from inquiry to start

Tip: use only real testimonials and results. Asked for “customer voices,” AI will happily invent them. Fake reviews and endorsements are illegal in many places (for example, under the FTC’s rules on fake reviews in the US and consumer-protection law in the UK and EU), and they contradict the whole point of this step. No results yet? Use real facts from your log — “ten years in business,” “200 loaves every morning” — or your Build-in-Public posts.

Tip: design for phones from the brief onward. Most visits come from phones. Say “phone first” in the brief and the build request, and check the live page on your own phone: does the proof of being real and the call to action fit on the first screen? Can you tell what the page is in three seconds?

The inquiry receiver, measured to the end

Receiver	When
A simple form (Google Forms, etc.)	free and fastest; let the AI write the questions from the brief
Messaging (WhatsApp Business, Instagram DMs)	casual businesses — cafés, studios, salons
Booking tool or email you already use	use what you have; don’t add tools

Two conditions: one tap from the call to action, and a clear “what happens after you send this” (reply time, next step).

Then measure the last step: set the inquiry as a key event in GA4 (a button click or the thank-you page). Now all three layers connect in numbers — which post (utm_campaign), through which page, produced how many inquiries.

I want to set clicks on this page's inquiry button as a key event in GA4.
Guide me one step at a time through the GA4 admin screens. I'll send screenshots as I go.

---

Are you done?

Step 1

• Logging rules are in the rules file, and the action log fills itself
• Work outside the AI (calendar, journal) also flows into actionlog/

Step 2

• A publishing policy sorts face, name, numbers, staff, and customers into the four levels
• A weekly routine picks 10+ things worth showing
• At least one piece designed with the four-part brief, in its own folder
• Platforms chosen from where your target is
• Three or more pieces published without clowning
• One-week results written back into the briefs, and at least one improvement found from them

Step 3

• content/offer-list.md breaks offers into one-page units
• GitHub + Cloudflare Pages publish on push
• Rules require brief-first building and add meta tags and GA4 automatically
• At least one post → page path is live

Step 4

• A live page with real proof above the fold
• The eight elements are your brief template
• An AI-made bonus attached to an offer
• No invented testimonials or results
• Checked on your phone, fixes written back
• An inquiry receiver connected, and a GA4 key event set

---

Practice: pitch a new project end to end

Use everything from Steps 1–4 to take one new project from idea to pitch.

1. Pick the project. A new product, service, event, or campaign — already inside your business. Or ask:

Read ontology/, research/, and the recent action log, and propose 3 projects worth pitching now (product, service, or event).
One line each: who, what, and why now.

2. Give it a folder and a brief.

content/
  ├── offer-list.md                ← add one line for this project
  └── 2026-10-25_new-winter-course/
        ├── lp-brief.md            ← offer plan first, then the eight elements
        ├── post-brief.md          ← four-part design (step 5)
        └── img/                   ← real photos

3. Polish the landing-page brief with the eight elements — real proof above the fold, only real voices, one AI-made bonus.

4. Build and publish — “build from the brief,” review, write fixes back, “publish,” then check on your phone.

5. Connect a post. Short video works best, but editing on the spot is hard, so start with a photo-and-text post or article. A few real photos plus the brief and log are enough. Link to the page with UTM. Put the one-week check on your calendar.

Tip: start small. One seasonal dish, a one-day event, one listing, one course. The point is to run idea → offer → page → post → numbers once. A loop that works small works big.

---

Teaching S3: three 2-hour sessions

Session	Theme	By the end
1	The content system	Media structure + Steps 1–2: logging rules, the weekly pick, content/, and the publishing policy. The first piece designed and checked, with a shooting plan
2	Landing pages at scale	Steps 3–4: offer list, GitHub + Cloudflare Pages, the eight-element brief, rules for brief-first building and tags; the first page with real proof published
3	Pitch a new project	Idea → folder → brief → page → post with UTM; the one-week check on the calendar

---

Related skills

• Before this → S2: Make information usable
• Next → S4: Multiply your revenue streams — once your voice is established, peers start asking for help: the door to B2B

---

SOVREN Framework is open-source. The English edition covers S1–S6; S7–S9 and the milestones are published in Japanese. The method is practiced in weekly live classes in Yokohama, Japan, with a report after every session.