Why use Claude Code for LinkedIn posts instead of the Claude chat app?

The chat app works fine, but you re-upload your context files every conversation. Claude Code runs in a folder on your machine. It reads CLAUDE.md on its own at the start of every session, you can point at any file with @, and a skill turns your whole prompt into one command. Your context lives in files you can edit and version, not in a chat history.

Do I need to know how to code to use Claude Code this way?

No. You need to open a terminal, cd into a folder and type claude. Everything in this guide is plain Markdown files. If you can write a Google Doc you can write a CLAUDE.md.

What is the difference between CLAUDE.md and the @ file mentions?

CLAUDE.md is loaded automatically at the start of every session, so it holds the rules that apply to every post. An @ mention pulls a specific file into the current prompt. Keep the rules in CLAUDE.md and the bigger reference material (your posts, company profile) in separate files you mention or that the skill reads.

Why tell Claude not to start writing yet?

Because its first draft anchors everything that follows. If it writes before it knows your angle, you spend the next ten messages dragging a mediocre draft toward what you wanted. Two minutes of questions up front is cheaper than that. Plan mode (Shift+Tab) enforces the same thing: Claude can read and ask, but not write files.

Does the same pattern work for cold emails and other GTM content?

Yes. Swap top-posts.md for your best-performing cold emails, add a file with your ICP and objections, and make a cold-email skill next to the LinkedIn one. The ladder is the same: context first, questions second, writing last.

Learn GTM & Sales Automation Claude Code Prompts

AI for GTM

Bad, better, best: Claude Code prompts for LinkedIn

The difference between a generic AI post and one that sounds like you is not a clever prompt. It's the context you give Claude, and whether it asks before it writes. Here's how to set that up once in Claude Code.

3 prompt levels 5 files to copy ~6 min read

There's an image doing the rounds that shows three prompts for the same LinkedIn post. The bad one asks for "a brief social media post". The better one tells Claude to ask questions first. The best one uploads three files (about me, company profile, top posts) and tells Claude to read them fully before asking anything.

It's good advice. It's also a lot of typing and uploading every time you want a post. Claude Code fixes that, because it works in a folder on your machine. The files stay put, the rules load themselves, and the whole best prompt becomes one command.

The prompt ladder

Same request, three levels of context

Each step up removes a guess Claude would otherwise have to make.

Bad prompt

Write me a brief social media post.

What you get

A generic post about nothing in particular, written for nobody, with three emojis and a "What do you think? 👇" at the end.

What's still missing

Platform, topic, audience, your voice, your goal. Claude has to guess all five, so it guesses the average.

Better prompt

I want to write a LinkedIn post about [topic]. Don't start yet. Ask me clarifying questions first so we agree on my industry, my target audience and the key message.

What you get

A short interview, then a draft that is about the right thing for the right people.

What's still missing

You still type your background, your offer and your style from memory every single time. And you forget half of it.

Best prompt

Read @context/about-me.md, @context/company-profile.md and @context/top-posts.md completely before you respond. I want to write a LinkedIn post about [topic]. Do NOT start writing yet. Ask me clarifying questions so we can work out the hook, the structure and the call to action together, one step at a time.

What you get

Questions that are already specific to you ("Your top posts all open with a number. Want to keep that?"), then a draft in your voice.

What's still missing

Nothing, except that it is long to type. Which is why the last section turns it into a one-word command.

The best prompt works for two reasons. The files replace guesses with facts about you. And "don't start writing yet" stops Claude anchoring on a first draft before it knows your angle. Once a draft exists, every edit is a tug-of-war with it.

From chat habit to Claude Code setup

Every part of the best prompt has a permanent home in Claude Code. You stop repeating yourself and start editing files.

In a chat app	In Claude Code	Why it's better
Upload ABOUT ME.md, COMPANY PROFILE.md, TOP POSTS.md	`Keep them in context/ and reference with @context/about-me.md`	Files live on disk. Edit once, every future post uses the new version.
Paste your style rules into every chat	`Put them in CLAUDE.md`	Claude Code reads CLAUDE.md on its own at the start of each session.
"Do NOT start writing yet"	`Shift+Tab into plan mode`	In plan mode Claude can read and ask questions but won't write or change files.
Retype the whole best prompt	`/linkedin-post [topic]`	A skill stores the prompt in a file. The topic after the command fills in $ARGUMENTS.

The setup

One folder, five small files

Plain Markdown. No code. Copy the starters and fill in the brackets.

linkedin/
├── CLAUDE.md               ← rules, loaded every session
├── context/
│   ├── about-me.md
│   ├── company-profile.md
│   └── top-posts.md
├── posts/                  ← approved drafts land here
└── .claude/
    └── skills/
        └── linkedin-post/
            └── SKILL.md    ← the /linkedin-post command

1

Make the folder

mkdir linkedin && cd linkedin, then create context/ and posts/. Any folder works. It does not need to be a code project.
2

Fill the three context files

Use the starters below. Spend real time on top-posts.md. Pasting your actual best posts does more for your voice than any adjective.
3

Add CLAUDE.md and the skill

Copy both from this page. Tweak the rules to match what you actually care about.
4

Run it

Type claude in the folder, then /linkedin-post and your topic. Answer the questions. Pick a hook. Edit the draft like you would a colleague's.

context/about-me.md

Who is talking. Role, what you have actually done, opinions you hold, words you never use.

# About me

- Role: founder of [company], ex-[previous role] at [company]
- I have done: [3 concrete things, with numbers if you have them]
- I believe: [2-3 opinions people in my industry would argue with]
- Tone: plain, direct, a bit dry. Short sentences.
- Never write: "game-changer", "unlock", "in today's fast-paced world", hashtags

context/company-profile.md

What you sell and to whom. The ICP, the pain, the proof, the one call to action you want.

# Company profile

- What we sell: [one sentence a customer would say]
- Who buys it: [job title] at [company size / industry]
- The pain they feel before buying: [in their words]
- Proof: [customer result, metric, or case study link]
- Default call to action: [book a call / try free / reply "guide"]

context/top-posts.md

Your 3 to 5 best-performing posts, pasted in full, with a line on why each one worked.

# Top posts

## Post 1 (42k impressions, 180 comments)
[paste the full post here]

Why it worked: opened with a specific number, one story, one ask.

## Post 2
[...]

CLAUDE.md: the rules you'd otherwise repeat

Claude Code reads a CLAUDE.md in the folder at the start of every session. Put anything that applies to every post here: length, format, banned habits, where drafts get saved. Keep it short. If a rule only matters for one post, it belongs in the prompt, not here.

CLAUDE.md

# LinkedIn content workspace

Before writing any post, read everything in context/.

Rules for every post:
- Ask clarifying questions before drafting. Never write on the first reply.
- Match the voice in context/top-posts.md, not a generic LinkedIn voice.
- Under 1,300 characters. One idea per post.
- Hook in the first two lines, before "...see more".
- One call to action, taken from context/company-profile.md.
- No hashtags, no emoji bullets.

Save approved drafts to posts/YYYY-MM-DD-slug.md.

The best prompt, as one command

Turn it into a /linkedin-post skill

A skill is a Markdown file with a name and a set of instructions. Save this as .claude/skills/linkedin-post/SKILL.md and it becomes a slash command. Whatever you type after the command replaces $ARGUMENTS.

.claude/skills/linkedin-post/SKILL.md

---
name: linkedin-post
description: Draft a LinkedIn post in my voice. Use when I ask for a LinkedIn post.
---

Topic: $ARGUMENTS

1. Read every file in context/ completely. Do not skim.
2. Do NOT start writing. Ask me up to 5 clarifying questions about
   the hook, the structure and the call to action. Wait for my answers.
3. Suggest 3 hooks. Let me pick one.
4. Write the draft. Then list anything you assumed that I did not tell you.

Now the best prompt from the image is this:

/linkedin-post why we stopped cold calling

Step 4 matters more than it looks. Asking Claude to list what it assumed surfaces the made-up stat or the customer quote you never gave it, before it goes out under your name.

The same pattern works for the rest of GTM

Nothing here is specific to LinkedIn. Swap top-posts.md for your five best-replied cold emails and you have a /cold-email skill. Add a file of common objections and you have one for follow-ups. Paste in a few call transcripts and it can draft a case study.

The ladder stays the same every time: give it the facts, make it ask, then let it write. If you're automating the sending side of outreach too, our n8n vs Clay breakdown covers the tooling. For everything Claude Code can do beyond this, the official Claude Code docs are the place to go.

Common questions

FAQ: Claude Code prompts for content

Want this set up for your team?

Makeinfo builds Claude Code workspaces, skills and automations for GTM teams: content, outreach, enrichment and reporting. Tell us what your team writes every week and we'll come back with a plan.

Contact Makeinfo → Email [email protected]

Or keep reading: more GTM & sales automation guides

Bad, better, best: Claude Code prompts for LinkedIn

Same request, three levels of context

From chat habit to Claude Code setup

One folder, five small files

Make the folder

Fill the three context files

Add CLAUDE.md and the skill

Run it

CLAUDE.md: the rules you'd otherwise repeat

Turn it into a /linkedin-post skill

The same pattern works for the rest of GTM

FAQ: Claude Code prompts for content

Want this set up for your team?