The system I'm running on a brand new account

Start here

When I started four years ago I had no system. I posted, hoped, and read the tea leaves when something worked. It eventually got me to 700k, but most of that was luck stacked on volume, and I couldn't tell you why half of it hit.

This time I'm treating the account like a machine with four moving parts. Each part feeds the next. You study the people already winning so you know what good looks like. You capture ideas the second they hit so none of them die in your head. You wire analytics into the loop so the account tells you what to make more of. And you start collecting emails on day one so you own the audience instead of renting it.

None of these are exotic. What's different is that they run as one loop instead of four disconnected habits. Below is each part with the exact setup I use. This is the real stack, not a beginner version — some of it assumes you're comfortable in a terminal and willing to run an AI coding agent like Claude Code on the $100/mo tier. Where a part has its own full build guide, I link it so you can go as deep as you want.

1. Study the top performers

Before you make anything, you reverse-engineer the people already winning in your space. Not to copy them, but to build a library of moves you can bend to your own content. Pick five to ten accounts you'd kill to trade places with, and for every video of theirs that pops, you pull three things out of it.

The idea. What is non-obvious about their take? Anyone can say "post consistently." The winners say the thing you haven't heard framed that way before. Write down the actual claim, not the topic.

The tactical layer. How do they make it real enough that a viewer can go do it? A take with no mechanism is a fortune cookie. Note the specific step, number, script, or example they hand the audience so the advice can actually be applied.

The hook structure — spoken and visual. There are two hooks in every good video, and most people only study one. The spoken hook is the first line out of their mouth. The visual hook is what's on screen in the first second: the text overlay, the framing, the motion. Note both. How did they grab you? Then try to apply the shape to your own idea.

How I run it. I keep a single swipe doc. Every time a video stops my thumb, I paste the link and fill four fields: idea, tactical, spoken hook, visual hook. Over a few weeks you stop seeing individual videos and start seeing patterns — the same hook shapes and structures recurring across your whole niche. Those patterns become templates. The trick I keep coming back to is that a hook is a reusable move, not a set of words. You transfer the structure and swap in your own subject.

Once you have enough of them, this library stops being a study exercise and becomes the raw material your scripting system pulls from (part 2). I've packaged the two most useful cuts of mine as their own free resources so you don't start from a blank page:

• The Hook Decoder — 10 viral hook patterns pulled from 69 talking-head videos, each with a fill-in-the-blank template and why it works.
• 5 Frameworks for Talking-Head Videos — the video structures the best short-form creators use, beat by beat.

You can keep this swipe library by hand in a doc. I run mine inside Melda — it breaks down every video I save into its hook, its structure, and why it worked, so what I'm building is a library of breakdowns, not a pile of links.

2. A system to capture ideas

The best idea you have this week will show up while you're walking, driving, or halfway through someone else's video. If capturing it takes more than ten seconds, you'll lose most of them. So the capture system has to live where you already are, which for me is my phone.

I built a Telegram agent for exactly this. I text it a reel link or a voice memo of a raw thought, and it does the work I'd otherwise put off: it pulls the transcript, researches the idea, and comes back with a full idea packet I can riff on later. When I sit down to write, I'm not staring at a blank page — I'm editing something that's already 70% there.

The flow, in plain English.

1. You send a URL or a voice note to a Telegram bot.
2. It pulls the transcript (Apify or Whisper) so the source is captured, not paraphrased.
3. It runs an EXTRACT pass that turns the transcript into a brief: the core idea, the proof points, the unique angle, the format. This brief becomes the single source of truth.
4. It pauses and waits for your SPIN — your take, your lived example, your override ("only cover these three, add my story here"). This is the one human decision in the loop.
5. It generates ranked HOOKS from your swipe library (part 1), you pick one.
6. It writes the SCRIPT from the brief plus the hook, runs a GATE check for AI tells and fabrications, and drops a Google Doc that's ready to record.

The reason it works is that the brief is locked at EXTRACT and nothing downstream re-reads the original video. Every later stage reads your brief, with your spin baked in. So the output sounds like you reacting to an idea, not a robot summarizing a transcript.

This is the deepest part of the stack, and it has its own complete build guide — from a blank server to your first script, every command and click:

• Build a Telegram Scripting Agent — the full build for the phone-first agent above.
• Build Your Content Second Brain — the next move: wire every script, blog, and video you've ever made into one searchable vault, so the agent checks what you've already said before it writes a new hook.

3. Tie analytics to the system

Most people check their views in each app, feel something, and move on. That's not analytics, that's mood. The point of tying analytics to the system is to make the account tell you what to do next, in one place, on a schedule.

I run everything through Metricool. Every time I post, it goes out through Metricool's API, and Metricool pulls the views and engagement back for every single post across every platform. One dashboard instead of five apps. That's the unlock: you can finally compare a Reel to a Short to a TikTok without doing mental currency conversion.

What you do with it. Two moves, run weekly.

1. Find the outliers and repurpose them. Once every post sits in one table, the winners are obvious — they're the rows two or three times above your median. Those aren't one-offs. Re-cut the winner for the platforms it didn't run on, turn it into a carousel, expand it into a newsletter or a long-form video. A proven idea deserves more than one shot.
2. Find the flops and re-angle them. A post that died usually didn't have a bad idea — it had a bad hook or a bad framing. Take the same idea, change the angle, and run it again. You'll be shocked how often the second attempt doubles the first.

The setup.

1. Create a Metricool account and connect every platform you post on (Instagram, TikTok, YouTube, and so on).
2. On a paid plan, grab your API access token and the "blog" (brand) ID from Metricool's settings. These two values let you script against your own account.
3. Route your posting through the API so publishing and measuring are the same pipe. I use a small script that takes a video plus a caption and schedules it to every platform at once, then a second script that pulls the metrics back into a Google Sheet on a weekly cron. The exact posting script I run is the same one behind my distribution workflow.
4. Sort that sheet by views descending every Monday. The top rows are your repurpose queue. The bottom rows are your re-angle queue.

You don't need the scripts to start. You can do the whole loop by hand inside Metricool's dashboard on day one. The scripts just remove the friction once the habit sticks.

4. Start the newsletter early

This is the one everybody delays, and it's the one I'd start on day one if I could do it over. Followers are rented. The platform decides who sees you, and it can change its mind overnight. An email list is the only audience you actually own.

But the real reason to start early isn't insurance, it's the conversation. When someone's on your list, you can ask them what they're actually struggling with and which topics they want you to cover. Every reply is two things at once: a content idea you didn't have to invent, and a signal about what product would genuinely serve these people. My best videos and my clearest product ideas both come straight out of reader replies.

The setup.

1. Create a Kit account. There's no reason to wait for "enough" followers.
2. Build one opt-in form. Give people a concrete reason to join — a specific resource, not "subscribe to my newsletter." This whole page you're reading is an example of that: a lead magnet worth an email.
3. Put the form everywhere the audience already is: your link in bio, the end of every video ("comment the keyword and I'll send it"), the footer of your site.
4. Send one email a week. It does not need to be long. End every issue with a single question and actually read the replies.

The last step is where the loop closes. Those replies feed part 1 (what's my audience telling me the top performers are getting right?) and part 2 (straight into the idea capture system). The newsletter isn't a separate channel. It's the account's feedback organ.

The order to build it in

You don't build all four at once. You'd stall. Here's the order that keeps you shipping while the system fills in underneath you.

1. Week one: start studying and start the list. Open a swipe doc (part 1) and open a Kit account with one form (part 4). Both take an afternoon and both start compounding immediately.
2. Weeks two to four: post, and track by hand. Use what you're learning from the swipe doc to make videos. Log every post's views in one place (part 3), by hand, inside Metricool. You're looking for your first outlier.
3. When the manual parts start to hurt: automate. Once capturing ideas and pulling metrics feels like a chore, build the Telegram agent (part 2) and script the Metricool pulls (part 3). Automate the habit you already have, never the one you wish you had.

That's the whole system. Four parts, one loop, built in the order that keeps you moving. Now go make the first thing.