# How to Build an AI Content System for LinkedIn from One Empty Folder (2026 Setup Guide)

> Build an AI content system for LinkedIn from one empty folder: a 30-minute interview, a rulebook, two content banks and your voice. Runs in 10–30 minutes a day.

Source: https://linkedinlikes.com/blog/how-to-build-an-ai-content-system-for-linkedin · Updated: 2026-11-12

## Key takeaways

- An AI content system for LinkedIn is a folder with your rules, stories, research and corrections that Claude reads before drafting.
- Setup takes one 30-minute interview; running the system takes 10 to 30 minutes a day.
- Your voice comes from two layers: a list of banned AI patterns, then 20 to 30 of your own posts or messages.
- Every correction you make is saved as a rule, and a mature rulebook runs past 100 lines.
- Claude drafts and tracks results, while you approve, read the post aloud and publish it from your phone.

An AI content system for LinkedIn is a folder of your rules, stories, research and past corrections that the AI reads before it drafts anything. You can set it up from an empty directory with a 30-minute interview, and after that it takes 10 to 30 minutes a day to run.

Three things keep it working: the material comes from you, the voice is yours, and you make the last edit on every post.

The guide has twelve parts: why single-prompt posts all read the same, the folder and the interview, the rulebook and skills, strategy, the knowledge bank, the research bank, voice, corrections, outlier posts, routines and the dashboard, approval, and how much time it all takes.

## Why do AI posts written from one prompt all sound the same?

![Empty folder on a desktop next to a chat window, the first step in building an AI content system for LinkedIn](https://linkedinlikes.com/img/content/blog/20261112_085743_01.jpg) Ask a chatbot to "write me a viral post" and it hands you roughly the text it hands everyone else, which explains why so many LinkedIn feeds sound like one person wrote them. Most networks now push back against autoposting and AI slop, and posts sent out by autoposting tools often reach far fewer people.

### Four inputs and one habit take the place of the single prompt

- Goals, so each post has a job to do: growth, trust or clients.
- Outlier posts from your niche, which show the topics and formats that work at the moment.
- Two banks of material, one for your own stories and one for research, so drafts start from something real.
- Tracking of results, so the drafts you get next week reflect this week's numbers.
- The habit: you correct each draft, and every correction is stored as a rule.

Posts with concrete details earn trust, and a high volume of vague posts does not, so the system aims for fewer and sharper posts. A feed packed with near-identical posts also trains readers to scroll past yours, and removing that cost is the reason the system exists.

**Do this now:** Go through your last five posts and flag every one you could have written with no experience of your own. The rest of this guide is about getting rid of posts like those.

## How do you set up the folder and the first interview?

![Claude desktop app with the Code tab open on a new LinkedIn content folder](https://linkedinlikes.com/img/content/blog/20261112_085802_02.jpg) Create one empty folder, open it in the Claude app and paste a single setup prompt; it builds the folder structure and then starts interviewing you. The interview runs about 30 minutes, and no other half hour in the setup matters as much.

### Five setup steps, one of which you do by hand

- Get the Claude desktop app from claude.ai/download and sign in. Running the system every day needs a paid Claude account.
- Make a new folder on your desktop and give it a name that says what it does, such as LinkedIn Content Engine. Keep it empty: the setup prompt creates all the other files.
- In the app, open the Code tab, start a new chat and select the folder. From that moment Claude can read and write files inside it.
- Paste the prompt below after replacing everything in {curly braces} with your own details.
- Answer each question with concrete detail, since vague answers come back later as vague drafts.

```
Turn this folder into my LinkedIn content system. Create CLAUDE.md as the rulebook and open it first in every session. Add these folders: knowledge_base, research, voice, drafts, results. Niche: {your niche}. Audience: {your audience}. Offer: {your offer}. Goal for the next 90 days: {your goal}. Next, interview me one question at a time about my work, my stories and my opinions. Store each answer in the matching folder and tell me what you stored.
```

**Do this now:** Put 30 minutes for the interview on your calendar and get through it in one go.

## What goes in the rulebook, and which skills should you start with?

![CLAUDE.md rulebook file shown above three skill files for drafting, hooks and review](https://linkedinlikes.com/img/content/blog/20261112_085804_03.jpg) CLAUDE.md is the rulebook, and it is the first file Claude opens in the folder. The most useful rule in it says the system gets better each week from your feedback: when Claude gets something wrong, say what exactly was wrong and why.

### The rulebook holds the habits, and skills divide the work

- Rulebook: a single file that Claude reads first and that you both edit. Some folders start out empty and fill up in later steps.
- Skill: an instruction file for one job, which loads whenever that job comes up. Think of each skill as one person on your content team.
- Begin with three skills, such as drafting posts, writing hooks and reviewing drafts. Each one needs several rounds of your corrections before it is any good, and three is about as many as you can give that attention.
- Add more once the first three work. The last few rely on the data you collect in the routines step.

After that, put them to work: ask for "write posts about {idea}" and keep correcting drafts until they read like you.

**Do this now:** Write down the three content jobs you repeat each week and ask Claude for a skill file for each.

## How many LinkedIn posts do you need, and what should each one do?

![30-day LinkedIn content calendar split into growth, trust and client posts](https://linkedinlikes.com/img/content/blog/20261112_085812_04.jpg) Every post has one of three jobs: it grows your audience, it builds trust with people who already read you, or it brings in clients. The strategy step assigns one of these jobs to each post. Claude then uses your goals to work out the weekly number of posts and their types, which keeps your content from drifting toward whatever crossed your mind that morning.

### Three jobs on one calendar

- Growth posts are for people who don't follow you yet, so they depend on sharp first two lines and a topic your niche already reads about.
- Trust posts are for current followers and show how you work, through your stories and the decisions you made.
- Client posts describe a problem your future clients have and what you do to fix it.
- Calendar: have Claude build a 30-day plan in a single session, and limit topics to a short, fixed list of pillars so each week follows the same rhythm.

A single session gives you a month of plans; later sessions are only for adjusting it.

**Do this now:** Put your 90-day goal into two sentences and let Claude turn it into post types and a weekly count.

## How does a daily note make AI posts sound like you?

![Voice note being turned into a text entry in the knowledge_base folder](https://linkedinlikes.com/img/content/blog/20261112_085815_05.jpg) What makes a post sound like you is your own material, such as stories, wins, mistakes and opinions, and something worth posting about happens most days. Spend about two minutes once a day telling Claude what happened; a voice note does the job as well as typing.

### Three kinds of raw material, one folder

- Daily notes: what happened, what it taught you, what you would change next time.
- Calls and meetings: record them or write them up, then put them in the same folder so Claude can turn them into posts.
- Notes and courses you already have on your topic: drag them into knowledge_base, since years of old material make a ready-made backlog.

Your idea backlog grows out of this bank. One note can turn into several post ideas, and Claude sorts them by your strategy.

**Do this now:** Take one note from today and put it through our [LinkedIn post ideas generator](https://linkedinlikes.com/linkedin-post-ideas-generator) to count how many angles a single day gives you.

## How do you keep AI posts tied to real sources?

![Research folder holding a pasted YouTube transcript and an article with its source link](https://linkedinlikes.com/img/content/blog/20261112_085821_06.jpg) When you come across something strong, whether a YouTube video, an article or a study, save its best lessons into the research bank. Your posts then rest on things you have read, and Claude has no reason to fill gaps from memory.

### Three ways to add a source

- YouTube video: click "Show transcript" below the video, copy the text and paste it with a request to save the best lessons.
- Article or study: copy the text and paste it in with the same request.
- Check before publishing: open the original for every number and quote, because an AI can state an invented fact in a confident tone.

Plain version: the bank stores what you read, and the check protects what you publish.

**Do this now:** Save one article from today's feed and open its original source before you reuse any number from it.

## How do you teach the AI your writing voice?

![Two layers of voice rules: a list of banned AI patterns above a stack of the author's own posts](https://linkedinlikes.com/img/content/blog/20261112_085825_07.jpg) Voice comes in two layers. The first is a set of general rules that bans the most common slop patterns, and you paste it into the chat before anything personal. The second is 20 to 30 pieces of your own writing, which give Claude a voice to match.

### Layer one: eight bans to paste before anything else

- Don't open with a sweeping scene-setting line about the modern world.
- Don't add lines that announce what comes next to keep people reading.
- Don't string em dashes together.
- Don't close with a question whose only purpose is to collect comments.
- Don't list three adjectives in a row.
- Don't use emoji as bullet markers.
- Don't pair sentences that reject one idea and then assert another.
- Don't use words you would never say out loud to a client.

### Where layer two comes from

- If you have posted before: copy the text of your last 20 to 30 LinkedIn posts, or request your data archive from LinkedIn to get all of them.
- If you have never posted: copy 20 to 30 messages you sent to friends or coworkers, because the way you text is the way you should post. A 3-minute voice note about your work, transcribed and pasted in, works too.
- Other writers: feeding in posts by authors you admire only produces an imitation of them, so keep them out.
- Do-not-use list: every phrase in a draft that makes you wince goes on it.

**Do this now:** Generate one draft with our [LinkedIn post generator](https://linkedinlikes.com/linkedin-post-generator) and mark each phrase that doesn't sound like you. Those phrases start your do-not-use list.

## Why should every correction become a rule?

![Rulebook growing past 100 lines, each line a saved correction to an AI draft](https://linkedinlikes.com/img/content/blog/20261112_085834_08.jpg) Whenever you change a draft, tell Claude what you changed and why, and it adds a new rule. A mature rulebook is longer than 100 lines, and each line is a correction, sometimes as small as one banned word. This list is where the quality builds up.

### How the rule list changes over time

- Weeks 1 and 2: plan on a lot of fixing, since the rulebook is almost empty.
- From then on: each fix improves the next draft, and the edits per draft go down.
- How to phrase a rule: state the change and the reason, such as "cut the closing question, it reads as filler".
- When to log it: on the same day, while you still remember why you made the change.

**Do this now:** Open your latest draft, find three edits you made on autopilot, and give Claude each one with its reason.

## How do outlier posts show what works in your niche?

![LinkedIn posts from accounts in one niche, with outliers at five times the usual reactions highlighted](https://linkedinlikes.com/img/content/blog/20261112_085836_09.jpg) Choose 5 to 10 accounts that write for the same audience as you, and look for their posts that did at least 5 times better than usual. Those posts tell you which topics and formats are working right now.

### What to take from an outlier post

- Mix of accounts: include big ones and small ones, because one breakout post from a small account shows what is within your reach as well.
- How to measure: compare each post with that account's typical reactions and comments, which anyone can see on the post.
- What to keep: the topic, the format and the first two lines. Leave the text itself alone.
- How to open: start with a story of your own. A post gets expanded when its first two lines work, and LinkedIn reads an expanded post as a strong reason to show it higher in other people's feeds.

Each week, paste the outliers into the research bank and Claude tells you which ones fit your pillars. Plain version: you borrow what works in the niche, retell it through your own experience, and the post stays yours.

**Do this now:** Take the topic of your strongest outlier, write five openings for it in our [LinkedIn hook generator](https://linkedinlikes.com/linkedin-hook-generator), and keep the one that matches your story.

## Which routines and dashboard keep the system running?

![Dashboard page with morning picks, weekly outliers and post results in one view](https://linkedinlikes.com/img/content/blog/20261112_085839_10.jpg) The system becomes useful once research and logging run on a schedule. Routines run on your computer while the Claude app is open; if the app is closed when a routine is due, it runs the next time you open the app.

### Two daily routines, a weekly review and one page

- Morning routine: collects fresh posts worth quoting, each with a suggested angle, and lists this week's outliers. The times can be changed.
- Evening routine: records how your posts did, from the numbers you paste in from each post's analytics.
- Sunday review: updates the list of what works; you then say "plan my week" and get a plan with ideas.
- Dashboard: a single page with the picks, the outliers and the results, so you don't open five files before you start writing.

**Do this now:** Have Claude set up the evening routine today, and paste in the numbers from your latest post so it has its first entry.

## Who approves and publishes the posts?

![Person reading a LinkedIn draft aloud before posting it from a phone](https://linkedinlikes.com/img/content/blog/20261112_085841_11.jpg) You do. Claude writes drafts, and the decision about what goes live stays with you. Full automation is the wrong setting: check everything it writes and rework it into the way you normally write.

### Five checks before a draft goes live

- Rubric: give the draft a score from 1 to 5 for specifics, relevance to your audience, credibility and novelty. A low score on any of them sends it back.
- Read it aloud: say the post out loud once, and rewrite any sentence you trip over.
- Draft check: look at the hook, the length, the line breaks and where the "see more" cut falls.
- Your edit: keep changing the draft until it reads as if you wrote it yourself.
- Your phone: post it by hand. Publishing from your own phone is the recommended habit, since most networks push back on autoposting and AI slop, and autoposting tools can cut reach hard.

**Do this now:** Paste the draft into our [LinkedIn post analyzer](https://linkedinlikes.com/linkedin-post-analyzer), fix whatever it flags, and publish the final version from your phone.

## How much time does an AI content system for LinkedIn take?

![Daily schedule for an AI content system: morning picks, midday post, evening log](https://linkedinlikes.com/img/content/blog/20261112_085848_12.jpg) Setup costs about 30 minutes for the interview, and running the system after that takes 10 to 30 minutes a day. You could spend a full day on content, studying viral posts and training the AI, though the system doesn't ask for that.

### A normal day with the system

- Morning: open the dashboard, read the picks and outliers, and choose today's post.
- Midday: rewrite the draft until it sounds like you, then publish it.
- Comments: let Claude draft comments on other people's posts, and edit each one to fit the post. A comment of 6 words or more that responds to the post counts for something, and a bare "nice" counts for nothing.
- Evening: log the results and add today's note to the knowledge bank.

Spending fewer minutes means you learn less about your niche, and the results get weaker. Plain version: the time you put in is the part of the system that does the learning.

**Do this now:** Write three replies to recent posts in your niche with our [LinkedIn comment generator](https://linkedinlikes.com/linkedin-comment-generator), then rework each one so it fits its post.

## Related reading

- [The 2027 LinkedIn marketing playbook](https://linkedinlikes.com/blog/the-2027-linkedin-marketing-playbook)
- [How to get more comments on LinkedIn](https://linkedinlikes.com/blog/how-to-get-more-comments-on-linkedin)

**In short:** an AI content system for LinkedIn is one folder holding your rules, two banks of material, your voice samples and your corrections, with a set of routines around it. Claude drafts and keeps track, and you choose, rewrite and post from your phone. Set up the folder in 30 minutes this week, then run it for 10 to 30 minutes a day.

## Frequently asked questions

### Does AI content work on LinkedIn?

It can, when the drafts start from your own material and you make the final edit. Posts you publish yourself from your phone keep their reach, while autoposting tools tend to cut it.

### How long does it take to set up an AI content system for LinkedIn?

Plan on about 30 minutes for the interview. Once everything is in place, the system needs 10 to 30 minutes a day, plus extra time for corrections in the first two weeks.

### Can AI write LinkedIn comments for me?

It can write a draft, which you then adjust to the post you are replying to. Replies of 6 words or more that react to what the post says carry weight; generic ones such as "nice" carry none.

### How many samples of my writing does the AI need?

Your last 20 to 30 posts are enough. If you have never posted, record a 3-minute voice note about your work and give Claude the transcript.

### Do I need a paid Claude account?

For daily use, yes. The setup runs in the Claude desktop app, and running the system every day needs a paid Claude account.

### What does an AI content system give you?

A regular posting rhythm and a growing rulebook that brings each draft closer to your voice. How your account grows still depends on the quality of what you post and on keeping the last word yourself.