Video walkthroughs · Claude Code Club

How To Grow LinkedIn Followers With AI (14,847 In 90 Days)

15 minute read · Companion guide to “4 Levels of Claude for LinkedIn Growth (1,200 to 14,847 Followers in 90 Days)”

Almost everyone who uses AI to write LinkedIn posts is trying to fix the wrong thing. They rewrite the prompt, then rewrite it again, and the posts still sound like nobody in particular. The account in this walkthrough grew from 1,200 followers to 14,847 in 90 days, and the thing that changed was not the prompt. It was what the model got to read before it started writing.

Watch the full walkthrough: the four levels, from a blank chat to a daily posting system

What the 90 days actually looked like

Here is the shape of it, with the real numbers. The starting point was 1,200 followers on an account that had been posting occasionally for a long time. The ending point 90 days later was 14,847. The cadence was one post per day, published in the morning. There was no ad spend, no engagement pod, and no viral moment carrying the whole thing.

The walkthrough breaks the tooling into four levels, from typing a prompt into a chat window up to a system that researches, writes, and publishes on its own. That progression matters, and a sibling guide covers the wiring. The growth itself came from a much smaller set of decisions, and all of them are about what you post rather than how you post it.

Those decisions are the subject of this guide. There are five, and they are worth naming up front so you can see where this is going:

• Feed the model three documents about you before you feed it a topic.
• Open every post with a hard number that was researched today, not remembered.
• Write to a three-beat skeleton that ends somewhere a reader can go.
• Source topics from what your audience complains about, not from what is trending.
• Post daily against a rotating list of themes so you never repeat yourself.

Why your AI posts sound like nobody

Open a fresh chat, ask for a viral LinkedIn post about how AI is reshaping the workforce, and you get something genuinely competent back. In the video, that first attempt produced a real line: most knowledge workers are one bad quarter away from being replaced, not by a smarter person, but by a $20 a month subscription. That is a usable hook. It is also completely anonymous.

The problem is not quality. The problem is that a blank chat has two things missing that no amount of prompt tuning replaces. It does not know what happened in your industry this week, and it does not know anything about you. A post with neither of those is a well-written post that belongs to no one, which is exactly why it gets scrolled past on a platform where people follow people.

Input 1: The three documents that make a post yours

This is the highest-leverage hour you will spend on the whole system. Write three documents once and every future post inherits them.

• Tone of voice. How your sentences are built, how long they run, what vocabulary you use and refuse to use, what your rhythm sounds like when you are talking rather than performing.
• Proof points. Every result, milestone, client outcome, revenue number, and credential you can honestly point at. This is what turns an opinion into a claim with weight behind it.
• Personal stories. The layoff, the failed launch, the moment something clicked. Specific scenes with specific details, not themes.

You do not have to write these from a blank page. If you have been talking to an AI assistant for a while, it already has a great deal of context about how you write. Ask it directly to build a tone of voice document from everything it knows about you from your past conversations, and edit what comes back. The video demonstrates this by building a deliberately absurd version, a voice document for Yoda, and the output flips instantly from corporate neutral to unmistakably that character. The point of the joke is that voice is a file, not a talent.

The decision worth making carefully is how much of the proof document is numbers versus adjectives. Numbers travel. "I have helped a lot of people with this" adds nothing to a post. "I did this 47 times last year and 3 of them failed for the same reason" is the whole post. If your proof document is mostly adjectives, it will produce mostly adjectives.

Input 2: Open with a number that was researched today

The single most repeatable pattern across the 90 days is this one. Before writing anything, the system searches the web for current information on the topic, specifically hunting for hard statistics that can be worked into the opening line. Posts that carry a real number in the first sentence consistently outperformed posts that opened with an opinion.

Two examples from the video show what this looks like in practice. One pulled research showing that 49% of companies using AI chat tools report having already replaced workers because of it. Another opened by pointing out that young people entering tech are the canaries in the coal mine, then immediately noting that this is not a metaphor, it is the phrase Stanford researchers used in their own 2025 paper. Both posts earn attention in the first line and then spend the rest of the post being interesting.

The mechanism is simple. A number is a claim that can be checked, and a claim that can be checked reads as a person who has done the work. An opinion in the first line reads as a person who has had a thought. On a feed where everyone is having thoughts, doing the work is the differentiator.

The hook patterns that carried the 90 days

When you ask for a post without specifying an angle, a good system will ask which one you want. That question is the useful part, because it forces you to pick a stance rather than defaulting to the flattest one. These are the angles that appeared repeatedly across the run, with what each is actually doing:

• The uncomfortable truth. State the thing your audience suspects but has not said out loud. Highest engagement, highest risk, and it only works if you can back it up in the next three lines.
• The contrarian reframe. Take the consensus position and show why it points the wrong way. "AI is not stealing jobs, it is splitting them" is this pattern in one line.
• The wrong question. Name what everyone is asking, say it is the wrong question, then supply the better one. This one reliably drives comments because it invites disagreement.
• The operator opportunity. Here is the gap this creates and here is who can walk through it. Lower reach, better quality of follower, because it filters for people who act.
• Borrowed authority. Lead with the researcher, the institution, or the report, then add your read on it. Safest of the five and the easiest to write when you have nothing personal to add.

The decision point here is reach versus fit. Uncomfortable truths and wrong-question reframes pull the most impressions. Operator posts pull fewer followers who are worth more. Running only the high-reach patterns grows a number that does not convert into anything, which is a failure mode people discover about four months in. A workable ratio is roughly three reach posts to two operator posts per week.

Input 3: The three-beat skeleton every post follows

The structure used across the 90 days is short enough to hold in your head, and the video states it plainly. Name the pain. Show the proof that you have solved it. Point at where the reader can go next.

1. Name the pain in the reader's own words, specifically enough that it is slightly uncomfortable to read. This is where the researched number usually sits.
2. Show your proof. Not a claim that you know the answer, but the actual thing you did and what it produced. This is the beat that the proof-points document exists to feed.
3. Point somewhere. A community, a resource, a next post, or a question worth answering in the comments. A post that ends nowhere converts attention into nothing.

That third beat is where most people go wrong in one of two directions. Either every post ends in a hard pitch, which trains an audience to stop reading, or no post ends anywhere at all, which is how accounts accumulate followers and no outcomes. The honest version of beat three is an offer of more, never a gate on what you just gave away. If the post itself was useful on its own, the pointer is welcome.

Input 4: Mine complaints, not trending topics

The topic source matters more than most people expect, and the approach in the video is deliberately unglamorous. Rather than writing about whatever is trending, the system goes to Reddit and finds what a specific audience is complaining about right now, filtered through a definition of who that audience is.

The difference is worth being precise about. Trending topics tell you what is getting attention, which means the post you write about them is competing with everyone else who noticed. Complaints tell you what is causing someone pain today, which is a much shorter path to a reader thinking that you are describing their week. A complaint is a hook that already has an audience attached to it.

How to run this by hand in about fifteen minutes:

• Write down who your audience is in one sentence. Not a demographic, a situation. "People in a corporate job trying to build something on the side" is usable. "Professionals" is not.
• Find the three or four subreddits and forums where those people talk when nobody is selling to them.
• Sort by recent and read the posts that are questions or frustrations, not the ones that are announcements.
• Collect the exact sentences. The phrasing is the asset, because using their words is what produces recognition.
• Generate three candidate angles from what you found and pick one. Choosing between three beats accepting the first.

Input 5: Daily cadence with a rotating keyword list

The cadence across the 90 days was one post per day, scheduled for the morning. Daily is the part that most people cannot sustain by hand, and it is also the part that produces the compounding, because the platform rewards consistency and because you cannot learn what works from four data points a month.

There is a trap in daily posting that the video names directly. Left alone, a system asked to write a post every day about your niche will circle the same three ideas until the audience stops seeing them. The fix is a rotating list of roughly ten keywords or themes that get cycled through, so each day starts from a different corner of the same territory.

Build the list the boring way. Write the ten questions your audience asks most often, phrase each as a theme rather than a title, and put them in a file the system reads before choosing a topic. Themes are broad enough to generate many posts each, which is what keeps a ten-item list running for a full quarter. Ten themes at one post per day is about three passes through the list in 90 days, and by the third pass you already know which themes perform.

The loop that compounds: read what worked, then do not repeat it

This is the part that separates the level-four version from everything below it, and it is a strategy decision rather than a tooling one. Before writing anything, the system reads two things: which of your past posts performed, and what you have posted recently.

Those two checks pull in opposite directions on purpose. The performance check pushes toward repeating what worked. The recency check stops that from becoming the same post four times. Without the recency check, a system optimizing for past performance converges on one post and then repeats it, which is the most common way an account that was growing quietly stops.

Running this by hand is genuinely feasible. Once a week, open your analytics, note the three posts with the highest engagement and the three lowest, and write one sentence about what separated them. Keep that file where you can see it while writing. Also keep a running list of the last fourteen topics you covered. Both take a few minutes and both change what you write next.

What did not work

These are the specific things the walkthrough shows failing, plus the ones that show up repeatedly among people running this system for the first time. They are worth reading before your first week rather than after your fourth.

• Prompting alone. A single well-written instruction with no voice or proof files produces posts that are technically strong and completely anonymous. This is the failure that stops most people.
• A system prompt with no files behind it. The video shows this exact intermediate step, and the output is described accurately as robotic. Structure without personal material is still nobody's voice.
• Opening with an opinion. Posts that lead with a stance and no evidence underperformed posts that led with a researched number, consistently enough to be a rule.
• The same topic on repeat. Daily posting without a rotating list produces variations on one idea, and audiences disengage before they unfollow, which makes it hard to notice.
• Manual copy and paste. Not a content failure but a cadence failure. The friction of moving text between windows is what quietly ends daily posting around week three.
• Ending nowhere. Posts with no next step gather impressions and produce no outcome, which is how an account grows a follower count and nothing else.
• Unverified statistics. One wrong number in front of a new audience costs more trust than the post earned in attention.

Your first week, in order

The shortest honest path from reading this to having a system that runs. The first three steps are the ones that take real thought. Everything after is repetition.

1. Write your audience in one sentence as a situation, not a demographic. Everything downstream reads this.
2. Build the three documents: tone of voice, proof points, personal stories. Ask an assistant to draft the voice one from your past conversations, then edit it hard.
3. Write the ten-theme rotation list from the questions your audience asks most.
4. Spend fifteen minutes in the forums where your audience complains, and collect exact sentences into a file.
5. Write one post using the three-beat skeleton with a researched number in the opening line. Open the source and verify it.
6. Publish it. Then do the same thing tomorrow, moving one step down the theme list.
7. At the end of week one, note which post performed best and write one sentence about why. Add that sentence to a file you read before writing post number eight.

Do that for a week and you have a system. Do it for a quarter and you have the 90 days described in the video, which is roughly 90 posts, three passes through your themes, and enough performance data that you are no longer guessing.

What to expect from the numbers

Setting expectations honestly matters here, because a 12x follower jump in a quarter invites the wrong reading. The account that grew from 1,200 to 14,847 was not starting from zero. It had an existing base, an established niche, and a person behind it who had already been posting. The system multiplied something that already existed rather than creating it from nothing.

What that means for you, depending on where you are starting:

• Under 500 followers: expect the first month to feel like posting into a room with nobody in it. That is normal, and the compounding starts when a post gets shared outside your network for the first time.
• Between 500 and 5,000: this is where the system moves fastest, because you have enough of a base for the algorithm to test against and enough room to grow quickly.
• Above 10,000: growth rate slows and the operator-angle posts matter far more than the reach posts, because the number stops being the point.

The follower count is a proxy, and a lagging one. The signal to watch weekly is comments, because comments are the only engagement that reliably turns into conversations, and conversations are what the whole thing is for. An account with 3,000 followers and twenty real comments a week is in better shape than one with 30,000 and none.

Everything in this guide can be run by hand starting today, with a notes app and thirty minutes a morning. Running it by hand for two weeks is also the fastest way to learn which parts actually matter, which is why the whole framework is written out here rather than held back.

Read this online at claudecodeclub.ai