scalematic.
BlueprintContentFounder Brand 2 min read · v1.0 · July 2026

Human-in-the-Loop Content Operating System

A content system where AI does the labor and humans keep the judgment — ideation, drafting, development, management, and performance analysis with approval gates throughout.

Executive summary

What problem does this solve?

Fully automated content sounds like nobody. Fully manual content does not ship. This system uses AI for the labor at every stage while keeping a human approval gate on anything that carries the brand.

Business outcome

  • Publish consistently without the founder writing every post
  • Keep the founder's actual voice in AI-drafted content
  • Feed performance data back into what gets produced
  • Cut content production time without losing judgment
Revenue maturityLvl 3–6

Pairs with the 90-Day Content Plan — install the plan first if you are starting cold.

Implementation effort
16implementation hours
People required
FounderMarketingSalesRevOpsDeveloper

Ongoing: ~1 hour of founder input per week.

DifficultyIntermediate
Business impactHigh
Time to install1 week
Automation65%
MaintenanceMedium
OwnerContent Engineering
Required software
FavikonSybillbeehiivSandcastles AIPerplexityMagicPostClaudeFigmaAdobe PremiereNotionWebflow CMSBuffer
Required integrations
Transcript captureContent calendarNewsletterBlog CMSYouTube
Architecture

How the system fits together

Human-in-the-Loop Content Operating System — system architecture 1 AI steps
100%
01 · Ideate
02 · Draft
03 · Develop
04 · Manage
05 · Distribute
06 · Analyze

Hover a node for detail, or tap a tool below to see where it runs.

Highlight tool
ScaleMatic original

ScaleMatic-designed system. The source playbook documents an AI-assists-writing model; this is our own AI-drafts / human-approves design with a Notion system of record. Not a reproduction of the original.

The problem it solves

Fully automated content sounds like nobody. Fully manual content does not ship. The teams that win run a system where AI accelerates every stage but a human approves the moments that carry the brand.

Expected outcomes
  • Consistent publishing without the founder writing every post
  • AI drafts in the founder's actual voice, not generic thought-leadership tone
  • Performance data feeding back into what gets produced next
Who it's for
  • Founders building a personal brand
  • Content teams adopting AI
  • Marketing leads
Before / After

What changes when this is installed

Before — manual
  • Founder writes every post from a blank page
  • Publishing stops the week things get busy
  • AI drafts sound like generic thought-leadership
  • No idea which posts actually produced pipeline
After — installed
  • AI drafts from the founder's own transcripts
  • A human approves in 20 minutes a week — publishing never stops
  • Every draft carries the founder's phrasing and stance
  • Performance data routes back into what gets produced next
Implementation

5 steps, start to finish

Everything downstream reads from one table. Build it before you capture a single idea — retrofitting properties after you have 50 rows is the most common reason this system stalls.

Where
Notion → new page → /database - full page. Name it "Content OS — Intake".
Configuration
  • Idea — Title
  • Source — Select: Sales call | Newsletter | Competitor | Trend | Founder note
  • Source URL — URL
  • Original transcript — Text (long)
  • Content pillar — Select (your 4 pillars)
  • Funnel stage — Select: TOFU | MOFU | BOFU
  • Format — Select: Text | Carousel | Video | Image
  • Draft status — Select: Raw idea | Research | Draft | Review | Approved | Scheduled | Published
  • Approval status — Select: Pending | Approved | Rejected
  • Owner — Person
  • Publish date — Date
  • Published URL — URL
  • Impressions — Number
  • Comments — Number
  • Leads generated — Number
  • Repurpose status — Select: Not yet | Queued | Repurposed
Expected result
A single row can move Raw idea → Research → Draft → Review → Approved → Scheduled → Published without leaving the table or losing its source transcript.
Test it
Create one dummy row. Set Draft status to each value in order. If any stage has no home for the data it needs (e.g. nowhere to paste the transcript), a property is missing — add it now.
If it fails
If the board view looks empty, you grouped by the wrong property. Group by Draft status, not Approval status — approval is a gate inside Review, not a pipeline stage.

Prompt library

Copy-ready prompts

Tuned for Claude, GPT, Gemini, and Grok. Copy and run.

Draft in founder voice
Below is a transcript of the founder speaking about a client situation. Draft a LinkedIn post using ONLY their phrasing, examples, and point of view. Do not add generic business language. Do not soften a strong opinion. Open with the most specific detail in the transcript. Keep it under 200 words.
Your turn

Build it now, not later

0/6 · 0%

Run one full cycle of the operating system right now — from raw input to an approved draft. Follow the decision points: they are the OS.

  1. 1

    Decision 1 — does a transcript exist? If yes, paste its source below and skip to step 3. If no, do step 2 first.

  2. 2

    Self-interview (15 min, voice memo, then transcribe): "What did a client ask this week that surprised you?" · "What do most people in your market get wrong?" · "What did you change your mind about recently?" Answer out loud, unfiltered.

  3. 3

    Run the drafting prompt from this resource's prompt library against the transcript. Paste the strongest draft it produced.

  4. 4

    Decision 2 — does it need editing? If the fix is one line, make it and continue. If you're rewriting it, note what the prompt got wrong (this feeds the next prompt revision) and re-run.

  5. 5

    Decision 3 — approve or reject. If approved: schedule it and note the date. If rejected: write one sentence on why — that sentence is training data for next week's cycle.

Progress saves in your browser — no account needed
Troubleshooting

When it breaks, here's the fix

Claude output sounds generic despite the transcript

The transcript is too clean — a summarized transcript loses the founder's phrasing. Feed the raw, verbatim transcript (filler words included) and instruct the model to reuse exact phrases, not paraphrase them.

No transcript exists this week

Run the self-interview template in the exercise below — 15 minutes of voice memo answering three questions produces enough raw material for a week of drafts. Never draft from nothing.

The approval queue keeps backing up

The gate is too heavy. Approval means yes/no/one-line-edit — if the founder is rewriting drafts, the drafting prompt needs fixing, not the approval step. Track which drafts get heavy edits and adjust the prompt with those examples.

Everything gets approved but nothing performs

The feedback loop isn't closed. Pull the top and bottom three posts monthly, feed both sets back into the ideation stage, and bias the next batch toward what the winners share.

FAQ

Common questions

Related resources

WorkflowMarketing

Track Employee LinkedIn Engagement

Turn the engagement on your team's LinkedIn content into a scored, tiered pipeline — capturing profile visitors and post engagers, then routing them by intent.

LevelAdvancedInstall1 weekAuto80%
ClayTeamfluenceTrigify+3
Install this system

Build it yourself, or have it installed

The documentation above is complete — everything you need is on this page. The only question is whether you want to spend the time.

Do it yourself

Free · 1 week

Have ScaleMatic install it

Done with your team

Follow the documentation
Complete implementation
Configure every tool yourself
Tool configuration included
Troubleshoot issues yourself
Tested and supported setup
Train your team on it
Team training and SOPs included
Time investment: several hours or days
Guided implementation

We diagnose the constraint first — if this system isn’t what you need, we’ll say so.