Handbook · by Dr. Louis Korczowski

The X Ghostwriter Handbook: grow on X from zero with an AI that writes in your voice, and a ranker that gives a profile click zero points.

X published the weights of its ranker: a copy of your post's link counts 20, a reply 5, a follow 4, and the profile click that brings you clients counts 0. This handbook turns that into eight recipes: a brand DNA your AI can check, hook forensics on your niche, a reply routine for cold accounts, a tracker for the clicks the ranker ignores, and one 30-day goal. Every recipe ends with one thing to do today and a prompt for your assistant.

Ask your AISummarize this handbook and apply it to your situation: ChatGPT ClaudeMarkdown AI skills on GitHub

TL;DR

An AI can ghostwrite your X account in your voice if you give it rules it can check. Write what you refuse to say as "I never" lines, your voice as a mix in percentages, and your pillars from what a buyer must believe before paying. Study ten viral posts of your niche and keep a formula file. Know what X's published ranker counts: a copy of your post's link weighs 20, a reply 5 (20 on your own posts shown to people who follow you back), a follow 4, a profile click 0. Under 1,000 followers, grow through replies and one X Article a week. Track the clicks the ranker ignores yourself, and judge the AI on one 30-day goal. It drafts and scores. You decide and you post.

The method

An AI drafts in your voice, the ranker decides the reach, and you decide everything else.

Eight recipes. Each one fits on a phone screen, ends with one thing to do today, and comes with a prompt for your AI assistant.

Fig. 1 · What X's ranker counts● a profile click weighs 0

When	What	Time
Once	Refusal list, voice mix, pillars, proof bank	1 h 45
Every weekday	3 replies under your target accounts	15 min
Every day you post	Draft, score, edit, post	20 min
Once a week	Hook forensics, tracker, one X Article	1 h 30
Every 30 days	Scorecard and the next goal	20 min

01Write your refusal list from your own posts45 min once · 10 min a month
02Set your voice mix, pillars and proof bank60 min once · 15 min a month
03Run hook forensics and keep a formula file40 min a week
04Know what X's ranker counts10 min once · re-read each quarter
05Draft with your AI, score before you post20 min a day
06Under 1,000 followers, grow through replies15 min a day
07Track the clicks the ranker ignores15 min a week
08Give your AI one 30-day goal and a scorecard20 min once · 10 min a week

The recipes

Eight recipes, from your first refusal line to the day-30 scorecard.

Write your refusal list from your own posts

45 min once · 10 min a month

Adjectives do not survive a deadline. "I never" lines that a stranger can check in five seconds do.

You need

Your last 90 days of posts and the replies under them
Under 50 posts: transcripts of your videos, sales emails, your about page

Steps

Ask your assistant to list the opening lines you used more than twice, the phrases you share with five bigger accounts of your niche, and the posts that drew pushback, each with its link.
Turn the list into 10 to 15 lines that start with "I never", in three groups: words, hooks, moves (framings, claims, behaviour).
Test each line: can a stranger check it in five seconds? "I never sound generic" fails. "I never open with a question" passes.
Link each line to the post that made you add it, and mark the lines your own posts break. You decide whether the line or the habit goes.

A model averages toward the most common phrasing unless it gets constraints it can test. Refusals are easier to check than qualities, which is why they work as a brief for a ghostwriter, human or AI.

Avoid

Writing the list from memory
Lines nobody can check ("I am authentic")
Keeping a rule your best posts break without deciding

TodayWrite three "I never" lines and link each to one of your posts.

For your AI

Copy this prompt into your assistant:

Here are my last 90 days of posts with links: [paste]. List every opening line I used more than twice, the phrases I share with 5 bigger accounts in my niche (name them), and the posts that drew pushback. Then turn it into a refusal list: 10 to 15 lines starting with "I never", in three groups (words, hooks, moves), each checkable by a stranger in five seconds, each linked to the post that justifies it. Flag the lines my own posts break.

Skill: statur-brand-dna-audit · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-brand-dna-audit

Set your voice mix, pillars and proof bank

60 min once · 15 min a month

Your voice is a mix of three modes in percentages. Your pillars are the beliefs a buyer needs before paying.

Steps

Define three modes. For each: what the first line does, sentence length, the kind of proof, how it ends. The ending decides the mode.
Write the target mix (for example 50 / 30 / 20) and ask your assistant to label your last 90 days. Compare the real split with the target.
List the 3 or 4 beliefs a buyer must hold before paying you. Each one becomes a pillar, named in four words or less, with the kind of post that proves it.
Write your one reader in a sentence that excludes people, and your one outcome: what that reader gets after a year of following you.
Build the proof bank: only numbers you can screenshot or link, with their source and date. Nobody cites a number that is not in it, you included.

Pillars that start from the offer keep reach and revenue in the same account. Posts that prove nothing a buyer needs to believe can still earn views, and they do not move anyone toward the offer.

Avoid

Pillars chosen from what you enjoy talking about
A voice mix nobody can label the same way twice
A proof bank with rounded or remembered numbers

TodayWrite the one reader sentence and the first two beliefs a buyer needs.

For your AI

Copy this prompt into your assistant:

Here is what I sell: [offer, price range, who buys]. Here are the questions people ask before they buy: [paste]. List the 3 or 4 beliefs a buyer must hold before paying me, turn each into a pillar named in 4 words or less, and give for each the kind of post that proves the belief and the kind that looks related but proves nothing. Then write my one reader in a sentence specific enough to exclude people, and my one outcome after a year.

Skill: statur-brand-dna-audit · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-brand-dna-audit

Run hook forensics and keep a formula file

40 min a week

Study ten viral posts of your niche case by case. A formula enters your file when two posts use it.

You need

10 to 20 posts from accounts your one reader follows, from the last 30 days
A few posts from the same accounts that did poorly

Steps

For each post, note the public counts and how many hours it has been live. Bookmarks above likes mean people saved it as a reference. Flag promoted posts and paid partnerships: their views tell you nothing about organic reach.
Describe the first line in your own words and what it does to the reader: a number, a confession, a curiosity gap, a conflict, a promise of a reference.
Note the moves further down: structure, proof, the objection answered in the post, where the link sits.
Give each repeated pattern a code (F1, F2...) and a verdict for your brand: use, test or skip, with the rule from your refusal list that decides.
Pick three formulas to test this week.

The idea comes from Avid's "hook forensics" bot (X, October 26, 2026), which studies viral creators and keeps a formula file he writes from. Any assistant can do it from public posts. Keep the claims you read in check: his own post credits the bot with "3.5M views in 14 days", a number nobody outside his analytics can verify.

Avoid

Copying a viral post's wording
Studying posts that were boosted with ads
Keeping a formula that needs a move you refuse

TodaySave five viral posts of your niche with their counts and the hours since posting.

For your AI

Copy this prompt into your assistant:

Here are 10 viral posts from my niche with their public counts and hours since posting: [paste links and numbers]. For each one, describe the first line in your own words and what it does to the reader, the moves further down, the actions it invites (replies, quotes, follows, link copies) and any claim to check. Then group repeated patterns into formulas with a code, and give a verdict for my brand using my refusal list: [paste].

Skill: statur-hook-forensics · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-hook-forensics

Know what X's ranker counts

10 min once · re-read each quarter

Write for replies, quotes, follows and link copies. A profile click earns no reach, so your business has to measure it.

Steps

Read the default weights X published in its ranker's code (listed below and in Fig. 1), and note the date: X can change them by configuration.
Write posts that invite a reply from your one reader: a number they can argue with, a mechanism they can add to, a question at the end when it is real.
Make the post worth sending: a table, a dataset, a short reference. A copy of the link carries the largest positive weight in the file, tied with a reply from a mutual follow.
Avoid anything that makes your one reader tap "not interested": bait, rage, topics outside your pillars.

In the code X published (xai-org/x-algorithm, param.rs), each weight multiplies the predicted probability of an action. Default values read on October 27, 2026: copy the link 20, reply 5, raised to 20 on original posts shown to mutual follows, quote 5, share by DM 5, follow 4, repost 1, like 0.66, open a link 0.2, profile click 0, not interested -52.3, mute -58.8, report -234. In the 2023 code (the-algorithm-ml, April 2023), a profile click weighed 12 and a reply 13.5: advice written from that version describes a ranker X has replaced.

Avoid

Quoting 2023 weights as current
Writing for likes
Bait that annoys the readers you want

TodayRewrite your next post so it ends on something a founder would argue with or forward.

For your AI

Copy this prompt into your assistant:

Here is my draft post for X: [paste]. Using the X ranker weights in the skill statur-x-viral-post-ghostwriter (references/x-ranker.md), tell me which ranked actions it invites (reply, quote, follow, link copy) and which it risks (not interested, mute). Rewrite it twice to invite more replies from [one reader], without bait.

Skill: statur-x-viral-post-ghostwriter · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-x-viral-post-ghostwriter

Draft with your AI, score before you post

20 min a day

Every draft names its pillar, its intent and its formula, and passes your refusal list line by line before you see it.

Steps

Split the week into reach posts (earn replies, quotes, follows) and revenue posts (earn profile visits, link clicks, messages about your offer). Label each draft.
Ask for two variants of each post, each on a different formula from your file.
Ask for a score line under each draft: refusal list pass or the broken lines, proof pass or the missing sources, the reply it invites.
Put links in the first reply with UTM parameters, so you can count the clicks.
Edit in your own words and post yourself. The assistant never posts.

A score line turns taste into a check you can run in seconds. It also catches the two mistakes that cost the most trust: a number with no source and a phrase from your refusal list.

Avoid

One kind of post only (all reach or all pitch)
A link in the main post
Letting the assistant post or reply for you

TodayAsk for this week's two revenue posts, with score lines.

For your AI

Copy this prompt into your assistant:

Using my brand DNA [paste refusal list, voice mix, pillars, one reader, proof bank] and my formula file [paste], draft this week on X: 3 reach posts and 2 revenue posts that point to the same next step [link]. Two variants each, on different formulas. Under each draft, a score line: refusal list pass or broken lines, proof pass or missing sources, intent, formula, the reply it invites.

Skill: statur-x-viral-post-ghostwriter · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-x-viral-post-ghostwriter

Under 1,000 followers, grow through replies

15 min a day

A cold account has no reach to spend. It earns it in other people's replies, and keeps it with one X Article a week.

Steps

List 10 to 20 accounts of roughly 5,000 to 100,000 followers that your one reader follows and that post on your pillars.
Every weekday, reply to 3 of their posts early in the post's life. Add a fact, a number or a mechanism the post missed. No compliment alone, no link.
Once a week, publish one X Article on your best method or dataset, and quote it the next day with a short demo or image.
Skip threads: on a small account each part is shown and read on its own.
Count follows from your one reader every week, and the replies that got a reply back.

In the published weights, the reply weight on your own original posts rises from 5 to 20 when they are shown to people who follow you back. Your replies under other accounts get the base weight. Their value is that the right readers see you and follow you, which turns your next original post into a conversation the ranker weighs 20. The weights multiply predicted probabilities, so never read them as counts of likes. The thread advice comes from our own test: on a 20-follower account, a 9-part thread fell from 59 views on its best part to 4 on the last.

Avoid

Replies under huge accounts only: views without followers
Generic praise
Automated replies: X's automation rules ban unsolicited ones and require X's written approval for AI reply bots

TodayWrite your list of 10 reply targets with their follower counts.

For your AI

Copy this prompt into your assistant:

My one reader is [sentence] and my pillars are [list]. Suggest 15 X accounts of 5K to 100K followers that this reader follows and that post on my pillars, and for each one an angle I can bring in a reply (a fact, a number or a mechanism from my proof bank: [paste]). I will check the follower counts and write the replies myself.

Skill: statur-x-viral-post-ghostwriter · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-x-viral-post-ghostwriter

Track the clicks the ranker ignores

15 min a week

The click that pays you carries no ranking weight. Count it yourself, every week, in a file you own.

You need

A sheet or a CSV file with one row per post
UTM parameters on every link you post (utm_source=x, utm_content=post id)
Your site analytics

Steps

One row per post: date, link, type (post, reply, quote, article), pillar, voice mode, intent, formula, views, replies, reposts, likes, bookmarks, profile visits, link clicks, follows.
Read the numbers once a week. Under 50 followers, X's analytics screen shows nothing per post: read the public counts on each post and the UTM clicks on your site.
Each Sunday, name the revenue post that brought the most link clicks and why, and the pillar that sends people to your offer.
Leave a cell empty when a number is not shown. Empty means unknown, never zero.

On October 27, 2026, X's analytics screen of a 20-follower account read "Detailed engagement metrics become available once you reach 50 followers". The ranker does not reward the profile click, and the platform hides it from small accounts, so the measure has to live on your side.

Avoid

Judging posts on likes
Typing numbers from memory
Links without UTM

TodayAdd UTM parameters to the link in your bio and your pinned post.

For your AI

Copy this prompt into your assistant:

Build my X post tracker from these posts [paste or screenshots]: one row per post with date, link, type, pillar, voice mode, intent (reach or revenue), formula, views, replies, reposts, likes, bookmarks, profile visits, link clicks, follows. Leave unknown cells empty. Then tell me which pillar sends people to my offer and which only earns reach, and what you cannot tell from this data.

Skill: statur-social-analytics · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-social-analytics

Give your AI one 30-day goal and a scorecard

20 min once · 10 min a week

One outcome, written like a contract. Judge the result, never the activity.

Steps

Write the goal as one business number: follows from your one reader, link clicks to your offer, demo requests.
Let the assistant measure the baseline and explain how. If a number cannot be read, set up the measure before day 1.
Write the limits: it drafts, scores and updates the tracker. It never posts, replies, messages, follows or edits a live page.
Every Monday: goal against result, what was tested, what to stop.
On day 30, fill the scorecard: goal, baseline and how it was measured, where it stands, what was tested, what moved, what did not, what to stop, the honest miss, the next 30 days.
If the goal moved and the business did not, diagnose the first weak step: views without profile visits (hook or pillar), visits without clicks (bio and pinned post), clicks without calls (offer page), one purchase and nothing after (missing next offer).

A goal gets a plan back. A task gets output back. The scorecard keeps the AI honest by asking for the miss in writing, and it keeps you honest by naming what you will stop doing.

Avoid

Ten goals at once
A baseline invented to make the result look good
Judging the month on views

TodayWrite your 30-day goal in one line and the date of the scorecard.

For your AI

Copy this prompt into your assistant:

Your goal for the next 30 days: [one number, e.g. 100 followers from my one reader]. Work out the baseline from my posts and tracker [paste] and tell me how you measured it. You may draft, score and update the tracker. You may not post, reply, message, follow or edit anything live. Send me your plan, the data you need and your questions first. Every Monday: goal vs result, what you tested, what you would stop. On day 30, fill this scorecard: goal, baseline, now, tested, moved, did not move, stop, the honest miss, next 30 days.

Skill: statur-x-viral-post-ghostwriter · install: npx skills add lkorczowski/content-strategist-AI-cookbook --skill statur-x-viral-post-ghostwriter

Worked example

I ran the handbook on my own X account first.

Before writing it for you, I ran the eight recipes on @LouisKorczowski. The account is a cold start, which is why recipe 06 exists.

Account	@LouisKorczowski, 20 followers, read on October 27, 2026
Posts and replies read	46, from August 25 to October 27, 2026
Median views per post	12.5
Best owned container	X Articles: median 268 views (2 Articles)
Best single post by views	a 4-word reply under SpaceX's Starship post: 3,028 views, 62% of the views recorded over two months, 2 likes
Threads	a 9-part thread fell from 59 views on its best part to 4 on the last
X analytics	locked under 50 followers
30-day goal set	100 followers and 30 link clicks to statur.co, scorecard on November 27, 2026

What changed after the run: no more threads, English only on this account, three replies a day under founders and operators my reader follows, one X Article a week, and UTM links everywhere. The scorecard is due on November 27, 2026. The honest part: one reply under a giant account made most of my views and brought almost nothing, and X hides the numbers that matter until 50 followers.

For your AI

The same method, written for your assistant.

Skills in the open Agent Skills format, maintained on GitHub so every update reaches you. Upload them in Claude, paste them in ChatGPT, or install them with one command.

statur-x-viral-post-ghostwriter

Draft and score X posts, replies, quotes and Articles in your voice, plan the weekly mix and the cold-start routine, run a 30-day goal with a scorecard.

statur-brand-dna-audit

Write your refusal list, voice mix, pillars, one reader, one outcome and proof bank from your own posts.

statur-hook-forensics

Study viral posts of your niche case by case and keep the hook formula file you draft from.

statur-social-analytics

Track X posts with profile visits, link clicks and follows in files you own.

Rules every skill follows. Drafts only: the AI never sends or publishes for you. No passwords: you sign in yourself. No promised views.

What to expect

The system, the volume and the measure are yours to keep. Reach is not guaranteed: a cold account grows slowly, one conversation at a time, and some months the scorecard will show an honest miss. The AI saves you the hours of reading, labelling and drafting. The ideas, the edits and the decision to post stay with you.

Credit

This handbook started from a public X Article by Avid ("Personal Brand 101: How to build a brand that makes money with AI", October 26, 2026), labelled on X as a paid partnership with Viktor, an AI coworker product. Elements taken from it: the refusal list of "I never" lines checkable in five seconds (recipe 01), the voice mix in percentages, pillars from buyer beliefs, the one reader, the one outcome and the proof bank (recipe 02), reach posts versus revenue posts with a weekly cadence of three reach and two revenue posts, and tracking the profile click yourself (recipes 05 and 07), the first weak step of the funnel and the 30-day contract with its scorecard (recipe 08). The hook forensics idea comes from his Grok bot post (recipe 03). Changed here: the sponsor's product is replaced by open skills that run in your own repository, the ranking claims are checked against X's published code (recipe 04), and the cold-start routine (recipe 06) comes from a test on a 20-follower account. The prompts and skills are written from scratch, in my own words.

The theory behind it

Why X gives a profile click zero points, what changed since the 2023 code, and what it means for a founder who sells something. The essay by Dr. Louis Korczowski on louis-korczowski.org comes out with this handbook. Why a ghostwriter alone makes founders sound like everyone else: the note on founder voice. Launching on X: the X launch playbook.

Questions

X ghostwriting with AI, answered.

What is an X ghostwriter?

A person or an AI that writes posts, replies and Articles for someone else's X account in their voice. In this handbook the AI drafts and scores, and the founder edits and posts every word.

Can AI write X posts that sound like me?

Yes, when it works from rules it can check: a refusal list of "I never" lines, a voice mix in percentages, your pillars and a proof bank. Without them it drifts toward the most common phrasing in your niche.

Does X's algorithm reward profile clicks?

Not in the code X published: the default weight of a profile click is 0, while a reply weighs 5 and a copy of the post's link 20. In the 2023 code it weighed 12. X can change these defaults by configuration.

How do I grow on X with no followers?

Reply every weekday under accounts of 5,000 to 100,000 followers that your reader follows, add a fact or a number each time, and publish one X Article a week. Track follows from your one reader rather than views.

Is it allowed to use AI to write on X?

Writing drafts with an assistant and posting them yourself is ordinary use. X's automation rules ban unsolicited automated replies and require X's written approval for AI reply bots, its rules ban bought engagement, and paid relationships need a clear disclosure such as X's Paid Partnership label or #ad. Read the current rules before automating anything.

Version 1.0 · last updated 2026-10-27: First edition: eight recipes from a paid-partnership X Article by Avid (credited), ranking claims checked against the weights X published in its code, and a cold-start routine tested on a 20-follower account. Three new companion skills in the public cookbook. Changes are logged in the changelog. Author: Dr. Louis Korczowski.

Bring the three ideas you repeat on every sales call.

Work with me

Two-minute application. I reply within 24 hours.