The content flywheel playbook

The usual failure mode looks like this: every video starts from zero, and nothing carries over from the last one that worked. Pick a topic, write a hook, film, edit, post, repeat. The playbook below fixes that in six steps — the first five are unglamorous, and the sixth is where the time comes back.

1. Source ideas from long-form, not short-form

Ideas come from YouTube videos, newsletters, long articles, books — not from your feed. Short-form is already downstream of long-form; if your ideas come from other short-form, you're producing a copy of a copy.

Then pair what you learned with something only you know: your experience, your data, your contrarian read.

Long-form input + your unique insight = a novel idea

The insight is the part that can't be scraped. The long-form input is what keeps it grounded in something real.

2. Source formats from the platform you post on

An idea and a format are two separate decisions. Keep a running list of formats that are working on the platform you're targeting right now, e.g.

• Listicles — "10/10 signs of…", "how to X in 60 seconds", "welcome to ___ 101"
• Sit-down talking head — rant, explaining a niche obsession, philosophy/aspirational
• Storytelling with B-roll overlay
• Whiteboard / 16:9 explainer
• Slideshow or carousel
• 7-second single-point video

Tag every idea in your backlog with a content type (educational, storytelling, transformational, entertainment, day-in-the-life) and a format from the list. The idea decides what; the format decides how it holds attention.

3. Find the outlier video in that format

For each format, find the creator who does it best and the specific video of theirs that over-performed relative to their normal numbers. That outlier is your validated starting point. You're not looking for a video to copy — you're looking for a structure that has already proven it holds attention.

Keep these organised: format → three example creators → their outlier videos.

4. Winning format + your novel idea

Take the elements that make the outlier work — the hook shape, the pacing, the way it closes — and run your own idea through them. You get what already works without taking anyone's content.

Winning format + novel idea = high-performing content

5. Repeat until you have data

Post across a handful of idea × format combinations until you have enough results to see a pattern. Look at shares and saves, not just views. Then double down on the combination that works for you — not the one that works for someone else.

This is the slow, boring part. There's no shortcut through it, but it's finite: you're collecting data points, not grinding forever.

6. The flywheel

This is where the time saving is. Once you have even one over-performing video, you can stop rebuilding from scratch. Three moves, in a loop:

Break down the outlier

Take your best-performing video (or the outlier you're modelling) and give it to an AI that has persistent memory — a system prompt, a memory file, a project. Ask it to analyse the video against a structured prompt. A useful breakdown covers roughly eight things:

1. The hook — the first line, and why it stops the scroll
2. The promise — what the viewer expects to get
3. Structure — how the points are ordered, where the payoff sits
4. Pacing — where it speeds up, where it slows down, and why
5. Language — specific words, phrases, and sentence shapes; tone
6. Mechanism — what psychological lever it pulls (identification, curiosity, mild provocation)
7. The ending — how it closes, what emotion it leaves, whether it creates a desire to follow, share, save, or comment; the implicit call to action even if none is stated
8. Shareability — why someone would send this to a friend; what sharing it says about the sharer; whether it makes them look smart, informed, or ahead

Save the formula

Turn the analysis into a script formula and store it in the AI's memory so it's loaded every time. It should read like a set of rules, not a summary. An example of what one looks like once it's been refined:

• Hook format: "10/10 [signs/habits] of [concept] — but it gets progressively more niche"
• The "progressively more niche" framing creates a watch incentive: the viewer stays for the point that catches them off guard
• Listicle with a payoff: numbered list, final item lands hardest, signposted with "and finally"
• Social mirror: every video makes the viewer self-identify ("that's me") or tag someone ("I know this person"). That's what drives shares and saves
• Every point follows behaviour + because / without realising + underlying principle. Never just a list; always explain why
• Charged vocabulary: real psychological or technical terms (agency, first principles, loss aversion, dopamine), each with a one-clause explanation
• Light antagonism: mildly provocative framing that triggers a reaction without being mean
• Diagnostic beats prescriptive: "here's how to spot it" outperforms "here's what to do"
• Sunk-cost pacing: first three points rapid-fire, one sentence each, no explanation — early watch time for the algorithm. Back half slows down and explains — the viewer is invested and stays for the payoff

Swap the idea, optimise the draft

For every new video:

1. Keep the formula, swap in a new idea (from step 1)
2. Write a rough script yourself, in your own voice — don't skip this; it's where the insight lives
3. Ask the AI to optimise this script against the formula in your memory

It rewrites your draft into the proven structure, using rules derived from your own performance data instead of generic advice. Then film it.

Every time a new video over-performs, feed it back in. The formula sharpens; the flywheel gets faster.

Why the playbook works

The playbook is a feedback loop with memory. Most people re-prompt from scratch every time and get random quality. This encodes what works once, in a place the model reads on every run, and refines it only when reality disagrees. The same shape as a well-maintained instructions file for any AI workflow: write the rules down, let the tool apply them, edit the rules when the output is wrong.

The playbook, one line each

1. Ideas from long-form + your own insight
2. Formats from the platform you're on
3. Find the outlier video in that format
4. Winning format + your idea
5. Repeat until the data tells you what works for you
6. Break down the winner → save the formula to memory → write rough, optimise against the formula → feed every new winner back in