LinkedIn maxxing with Claude Code: a 12-prompt content system in one folder

LinkedIn maxxing with Claude Code means keeping your whole LinkedIn content system as plain Markdown files in one folder (business context, voice guide, research, drafts, reviews and reports) and running 12 prompts over those files. Claude Code reads the folder, CLAUDE.md and skills load your rules automatically, and you edit files instead of retyping instructions every week.

Ask an AI model for a post on a topic and it skips the work that makes a post worth reading: knowing your clients, picking an angle, gathering proof and knowing what you can honestly claim. The 12 prompts below turn that work into a routine, from market research to a content calendar to working out which posts led to actual business conversations.

By the end you will have:

• A voice guide drawn from how you really write and speak.
• A research routine that collects ideas before you need them.
• A draft-and-review process with clear quality standards.
• A reporting loop that tells you what to write next.

Keep this page open for your next content planning session.

What context should you load before the first prompt?

A job title and a request to "sound human" are not enough. Claude Code needs to know who you sell to, what you believe and which claims you can back up.

Paste this first:

Treat the details below as the standing context for my LinkedIn content. Store them in a reference file I can reuse and tell me its location. Company: [name, website, what we sell, usual length of a sale]. Audience: [client job titles, company sizes, industries, regions]. They struggle with [problems], push back with [objections], and start looking for help when [trigger events]. Positioning: we get [audience] to [result] by [method]. What sets us apart: [differences]. We are the wrong choice for [disqualifiers]. Opinions: I hold these three views about my industry: [views]. I disagree with [popular advice] because [reason]. Proof: draw only on these approved case studies, results, examples and personal stories: [sources]. Never make up customers, revenue numbers, anecdotes or quotes. Voice: these [10-20 posts] and [transcripts or other writing] sound like me. Stay away from [phrases, tones, formatting habits]. Goals: my main aim on LinkedIn is [qualified conversations / help with open deals / hiring / awareness]. I offer [offer], and I want interested readers to [next step]. Ground rules: only use connected sources I can access and have permission for. If you can't open a source, ask me for an export or pasted text. Publishing, commenting, connection requests and messages on LinkedIn stay with me. Put drafts in [review location], keep evidence apart from inference, and mark anything that needs my input.

Update this context whenever your offer, audience or opinions change. In Claude Code that is a single file edit: change context/business-context.md, and every later session picks up the new version.

Where does Claude Code keep this context?

Claude Code works inside whatever folder you start it in, so the context lives as files on disk. The folder doesn't have to be a code project; an empty folder named linkedin is enough.

A layout that covers all 12 prompts:

linkedin/
  CLAUDE.md
  context/   business-context.md, voice-guide.md, top-posts.md, sources.md
  calls/
  transcripts/
  data/
  research/
  briefs/
  drafts/
  review/
  published/
  .claude/skills/
    linkedin-calendar/
    linkedin-draft/

• The reference file. Start claude in the folder, run the prompt above and ask it to save the result as context/business-context.md. That path is your answer to "tell me its location".
• CLAUDE.md for rules that always apply. Put a CLAUDE.md at the root with a few lines: read @context/business-context.md and @context/voice-guide.md before any LinkedIn work, save drafts to drafts/, and never publish or send anything. Claude Code loads CLAUDE.md when each session starts, and @path imports bring in the files they name. Keep it to roughly 200 lines; longer reference material goes in skills or in .claude/rules/ files.
• AGENTS.md. If the folder already contains an AGENTS.md, Claude Code can load that file instead of CLAUDE.md.
• Auto memory. Claude Code also writes its own memory notes, which load before you type anything and survive /compact. /memory opens CLAUDE.md and those notes for editing, and /context shows what was actually loaded.
• User-level preferences. CLAUDE.md files stack across levels. Habits you want everywhere, such as spelling or no emojis, go in your user-level CLAUDE.md, while the LinkedIn rules stay in this folder.
• Output style. For writing sessions you can switch to an output style with a writing-assistant role, so the replies stop reading like an engineer's notes.

Part 1: research and content production

Prompts 1-8. Each one reads files from the linkedin folder and writes its result back there, ready for the next prompt.

1. How do you build a voice guide?

A voice guide earns its keep when it explains the decisions behind your writing: how you set up an argument, how you use examples, how you move from one idea to the next. A line like "conversational and authoritative" tells a model almost nothing.

Read the posts and transcripts I've given you and write a practical voice guide for my LinkedIn writing. Cover: how I start a post, explain an idea and close an argument; the words I tend to use, my usual sentence and paragraph length, and how often I use lists; how I show disagreement, doubt, humor and confidence; phrases that come naturally to me, quoted word for word with the file they came from; habits that appear in my weaker pieces and shouldn't become rules. Add a list of phrases and patterns to avoid, such as needless repetition, invented drama, certainty without support and too many one-line paragraphs. End with three before-and-after edits that move generic copy into my voice without changing what it says. Make the guide specific enough for another writer to follow.

In Claude Code:

• Input: your best posts in context/top-posts.md, one post per heading with its reactions and comments, plus transcripts as text files in transcripts/. Reference both with @ paths so Claude reads every line.
• Output: context/voice-guide.md. The "file they came from" references turn into file names and headings you can open and check.
• Where the rules go: copy only the five or six rules that apply to every single post into CLAUDE.md. The full guide stays in its own file and gets imported when a prompt needs it, so it isn't loaded in every session.
• Watch for: when Claude gets the same phrase wrong twice, put the fix in CLAUDE.md instead of correcting it a third time in chat.

Why it matters: a voice guide turns your feedback into something reusable. You fix the instructions behind every draft once, instead of fixing the same wording every week.

2. How do you find the client questions worth answering?

Sales calls and customer conversations surface questions no keyword tool will show you. The useful ones usually sit around a decision someone finds hard to make.

Go through the sales calls, support threads, customer interviews and lost-deal notes from [period]. Sort the questions that keep coming up into five groups: problems clients are trying to make sense of; worries that slow down a decision; alternatives they weigh against us; expectations that end in disappointment; questions asked just before they commit. For every theme give an anonymized quote, its source, how many separate conversations raise it, the client's role and how far along their decision they are. Propose two LinkedIn angles per theme, one that helps a reader diagnose the problem and one that helps them judge a solution. Rank the top 10 by how often they recur, how much they matter commercially and how strong our evidence is. Keep what clients actually said apart from your reading of it.

In Claude Code:

• Input: calls, support tickets and lost-deal notes exported into calls/ as text or CSV, one conversation per file, with the date and the customer's role in the file name.
• Run: forty transcripts would swamp your main session, so ask Claude to give the reading to a subagent. It works in a separate context window and hands back only the grouped themes and counts.
• Output: research/customer-questions.md, with each quote's file name as its source.
• Check: the "separate conversations" count is easy to verify, since you can search calls/ for the quoted phrase yourself.

Why it matters: you get a backlog of post ideas tied to conversations your business is already having, in wording you know and with evidence you can check.

3. How do you find topics moving from X into your market?

A thread picking up attention on X can turn into a useful LinkedIn post, as long as you explain why it matters to your clients and add something of your own beyond a rewrite.

Use web search, the pages I link and public sources you can reach to go over the past [seven days] of discussion on [topics] and from these accounts: [accounts]. Pick up to 10 developments or arguments that matter to [client role]. For each one give me: the original source, its author, the publication date and the link; a neutral summary of what happened or what people are claiming; the strongest competing readings; what it means for our audience day to day; an angle that our own experience or evidence supports. Treat like and repost counts as a snapshot, don't take popularity as proof, and check factual claims against the original source where you can. Recommend three ideas to develop and say what I would have to add before each could become a credible LinkedIn post.

In Claude Code:

• Search: the built-in web search tool works when you sign in with a claude.ai account or an Anthropic Console account. On Amazon Bedrock it isn't available, so hand the WebFetch tool specific URLs instead.
• X: Claude Code has no native X search. Keep the accounts and thread links you follow in context/sources.md and reference that file in the prompt. If a page won't load, paste the posts into research/raw/.
• Output: research/topics-YYYY-MM-DD.md, so every weekly sweep is kept and prompt 6 can read the newest one.
• Watch for: a summary citing a link Claude never opened. Ask it to mark which sources it read in full.

Why it matters: research hands you a starting idea, and your judgment gives readers a reason to follow you. A good LinkedIn version ties the discussion to a decision your readers are actually facing.

4. How do you analyze posts that beat their authors' usual results?

Measuring every post against a celebrity account's biggest hit skews your research. Compare each post with what that author normally gets first, then look at what might explain the gap.

Here is a LinkedIn swipe file with whatever performance data exists: [file]. Where there is enough comparable data, set each post's reactions against the median for the same author, format and period. Show the math, and flag thin samples, missing numbers and posts that are still very new. For the clearest outliers, describe: the topic and the reader it was written for; the opening, argument, evidence and close; the format and rough length; whether the comments suggest real interest from clients; other possible causes such as a launch, a news event or unusually wide distribution. Turn this into five structural patterns I can reuse, and suggest an original idea for each based on what I know first-hand. Don't copy anyone's distinctive wording, and don't present a pattern you noticed as a rule of the algorithm.

In Claude Code:

• Input: data/swipe.csv with columns for author, date, format, reactions, comments and reposts.
• Run: have Claude write and run a short script that calculates each author's median per format, so the numbers come from code and not from estimates in prose. "Show the math" then means a script you can run again as the file grows.
• Output: research/outliers.md with the five patterns and an original idea for each.
• Your own baseline: compare your posts with your usual results in the LinkedIn engagement rate calculator.

Why it matters: you end up with ideas worth testing, and you avoid assuming that copying a popular post's layout will copy its results.

5. How do you turn recorded conversations into original posts?

A transcript can hide a strong post inside a story, a passing explanation or a disagreement that deserved more time than it got in the original conversation.

Go through this transcript from an interview, podcast, webinar or meeting: [source]. Find eight moments that could carry an original LinkedIn post, favoring decisions, mistakes, surprising observations, practical explanations and opinions grounded in experience. For each moment give: the exact quote with its timestamp or location; the main idea and who it would help; any context needed so the speaker isn't misrepresented; facts already available to back it up; one follow-up question that would make the idea stronger. Choose the three best and draft a post from each in my voice, following the voice guide. Keep the speaker's level of certainty, take out anything confidential and leave placeholders wherever a result or example still needs confirming.

In Claude Code:

• Plan first: press Shift+Tab to switch to plan mode, so Claude reads the transcript and suggests the eight moments without writing any files.
• Then draft: once you have agreed on the three, leave plan mode and let it write drafts/YYYY-MM-DD-slug.md, one file per post.
• Placeholders: keep them in square brackets. Searching drafts/ for [ lists every result that still needs checking.
• Timestamps: you only get them if the transcript has them, so export it from your recorder with timestamps switched on.

Why it matters: you can publish from knowledge that already exists, without asking someone to come up with a fresh opinion every time the calendar has a gap.

6. How do you build a four-week content calendar?

A calendar has to balance what clients need to understand against what you have enough evidence to say. Filling every slot with another version of your offer makes the feed repetitive.

Using the approved ideas from our research, plan four weeks of LinkedIn posts at [three posts a week]. Mix in: practical explanations that help clients fix a problem; opinions backed by experience or evidence; customer examples we are allowed to share; answers to objections and decision questions; the occasional offer post with a fitting next step. For every post, name the reader, the single objective, the main argument, the source material, a suggested format and what the reader should do next. Check for repeated topics, overlapping arguments, claims without support and too much self-promotion. Write full briefs for week one and shorter outlines for weeks two to four. Keep one slot a week open for a timely swap instead of treating the plan as fixed.

In Claude Code:

• Input: the outputs of prompts 2-4: @research/customer-questions.md, the newest topics file and @research/outliers.md.
• Output: calendar.md as a Markdown table, plus each week-one brief in its own file under briefs/.
• Reuse: save the prompt as a skill in .claude/skills/linkedin-calendar/SKILL.md. Next month the whole step is one /linkedin-calendar command.
• Timing and formats: check the posting days with the best time to post on LinkedIn tool, and build any carousel slot in the LinkedIn carousel maker.

Why it matters: a good calendar cuts down weekly planning while leaving space for better ideas, new evidence and changes in the market.

7. How do you draft posts that have something to say?

The research should settle the argument before Claude Code starts polishing sentences. Skip that order and you get tidy writing that spends several paragraphs arriving at an obvious idea.

Turn these approved briefs into three LinkedIn drafts: [briefs]. Follow my business context and voice guide. In each draft: make one clear argument for one reader; open with a concrete observation, tension, result or question that the brief supports; walk through the reasoning with a real example; use bullets when parallel ideas are easier to scan that way; vary sentence and paragraph length, with an occasional one-line paragraph for emphasis; close with a useful implication or a natural next step. Aim for [180-300 words] unless the idea needs a different length. Avoid restating the ending, made-up personal stories, inflated promises and formula contrasts. Give two alternative openings and explain in a line how each shifts the emphasis. List missing evidence separately from the copy that is ready to publish.

In Claude Code:

• Skill: this is the prompt to turn into a /linkedin-draft skill that takes a brief file as its argument.
• Safety: since the skill writes files, set disable-model-invocation: true in its frontmatter so it only runs when you call it by name.
• Output: each draft in drafts/, with the two alternative openings and the list of missing evidence at the bottom of the same file.
• Openings: try a few extra hooks in the LinkedIn hook generator before choosing between the two alternatives.

Why it matters: tight constraints give you drafts that are easier to judge, so your review goes into the argument and not into cutting filler.

8. How do you set up a separate editorial review?

Give the reviewer one clear job: find the weak spots in the draft and say how to fix them. A friendly "This is strong" doesn't count as editing.

Check these LinkedIn drafts against their brief, the source material, the business context and the voice guide. Score each out of 100 with these weights: relevance to clients 25, concrete detail and evidence 25, strength and clarity of the argument 20, fit with my voice 20, readability and formatting 10. Back each score with evidence from the text. Verify every factual claim, mark generic passages, and flag repetition, abrupt one-line fragments and paragraphs that would read better as lists. Return a revision that is ready to publish, a short note on the changes that matter and any open questions. Treat made-up evidence, leaked confidential information or a misquoted source as a blocker whatever the score. Keep everything in the review file for a person to approve.

In Claude Code:

• Reviewer: a subagent starts with a fresh context that includes your CLAUDE.md but not the conversation that produced the drafts, so it judges the text itself and not the reasoning behind it.
• Output: review/YYYY-MM-DD.md with the scores, the revision and the open questions. The drafts stay as they are until you approve.
• Automatic checks: to check length every time a draft file changes, add a hook that runs a small script after each edit. Hooks fire on every matching event, with no decision from Claude.
• Last look: preview the approved text in the LinkedIn post preview and confirm its length in the LinkedIn character counter.

Why it matters: a scoring rubric puts your standards in writing. The score organizes the feedback; the source checks and your own judgment decide whether the post is ready.

What should you do in the first hour after publishing?

The review leaves you with an approved file in review/. Copy it into LinkedIn and publish it yourself, then add the live URL and the time at the top of published/YYYY-MM-DD-slug.md; prompt 11 uses that to compare posts at the same age. From this point the method is the same as any other: get as many reactions as you can in the post's first hour.

Start with people you know. Ask colleagues, friends, relatives and clients, anyone you can actually reach, to react as soon as the post is up.

Use a community when that circle is small. Pick one where people act by hand from their own accounts and where actions that never happened can be reported.

Use checked groups. The groups in the table below were checked on Oct 21, 2026.

LinkedInLikes is a free community where LinkedIn members support each other's posts:

• members like, comment, repost and save by hand from their own LinkedIn accounts;
• comments are written in each member's own words;
• if someone confirms an action they didn't do, you can report it within 72 hours and the points are taken back;
• it never asks for your LinkedIn password, cookie or a browser extension, only a public link.

After the post goes live, paste its link into a new task on LinkedInLikes and choose the actions you want; publishing, comments, invitations and messages stay with you. New accounts get 100 starter points after confirming their email. See how comments work on LinkedInLikes.

Which LinkedIn groups can you join for the first hour?

These free communities were checked on Oct 21, 2026, with the member counts each platform showed that day. Every group is run by its own admins with its own rules, so read the pinned message before sharing a link. Save the table as context/groups.md if you want it available inside your Claude Code sessions.

If you'd prefer not to go through groups for every post, LinkedInLikes for likes works from one public link.

Part 2: pipeline and performance

Prompts 9-12. These use sales data, so this is where file locations and .gitignore start to matter.

9. How do you write content around deals that went quiet?

Some of your most useful posts answer a worry several prospects raised in private. The aim is to explain the issue well enough that future clients arrive with better information.

Go through the CRM records and call notes for qualified deals that went quiet during [period]. Find the objections that repeat and separate: objections stated outright in conversations; possible explanations nobody has confirmed; internal or process delays that content probably won't fix. For each repeated objection, note the client's role, the deal stage, the evidence we have and the question a useful LinkedIn post should answer. Recommend five content briefs. Each should present the concern fairly, show where our approach fits, admit where it doesn't and name the proof it needs. Draft the two strongest posts without naming prospects or exposing deal details. Separately, list approved existing material a salesperson could send in a relevant conversation.

In Claude Code:

• CRM: if your CRM offers a remote MCP server, add it with its own name and URL and sign in through /mcp. If not, export the quiet deals to data/stalled-deals.csv.
• Privacy: deal notes are sensitive. If the folder is a git repository, add data/ to .gitignore so they never get committed.
• Output: briefs go to briefs/ and the two drafts to drafts/, so prospect details never sit next to copy you might publish.
• Watch for: a connected CRM server can write as well as read. Approve its tool calls one by one for the first few runs.

Why it matters: it links marketing to the questions salespeople keep having to answer, and it stops you from writing persuasive posts about problems your clients don't care about.

10. How do you prepare briefs for LinkedIn conversations?

When someone responds to a post, the right next step depends on what they wrote and whether a business conversation would be welcome. A reaction on its own doesn't show business interest.

Here are LinkedIn conversations I've collected: [comments, inbound messages or interactions I logged by hand]. Sort them into: questions that deserve a useful public answer; conversations where sharing a resource would help; clear interest in our offer; general discussion with no commercial next step. For each relevant conversation, sum up what the person said they need, the business context we have and what we still don't know. Draft a reply that deals with what they actually wrote before bringing up anything else. Where it fits, suggest one follow-up question or a resource we already have. Don't read business interest into a like, don't act familiar with someone you don't know, and don't force a meeting request into a conversation that can't carry one. Leave every reply as a draft for me to check and send.

In Claude Code:

• Input: by default Claude Code can't see your LinkedIn comments or inbox, so paste them into inbox/YYYY-MM-DD.md, one conversation per heading.
• Output: replies/YYYY-MM-DD.md, with each draft under the message it answers.
• Sending: you copy each reply and send it yourself.
• Second opinion: for short public replies, compare against suggestions from the LinkedIn comment generator.

Why it matters: preparation makes replies more considered without turning every exchange into a sales sequence. Sometimes the right outcome is just a good answer.

11. How do you report on business results with clear attribution?

Impressions show distribution. They can't tell you whether the right people got interested or whether a post helped close a deal.

Merge my LinkedIn analytics, tracked website visits, inbound inquiry records and the relevant CRM data for [period]. Report on: reach and interactions by post, topic and format; qualified inbound conversations; resource sign-ups and demo requests where we track them; opportunities with a documented contact through LinkedIn; time spent writing and reviewing. Keep three categories apart: conversions we tracked directly, discovery people reported themselves, and possible influence. Don't attribute a whole deal to LinkedIn because one contact reacted to a post. Call out posts that drew relevant prospects, posts that reached a wide but less relevant audience, and posts with too little data to judge. End with three findings, three limitations and one experiment to run. Where possible, compare posts at the same age so new posts aren't judged against older ones.

In Claude Code:

• Input: monthly exports in data/YYYY-MM/: LinkedIn analytics, website visits, inbound inquiries and CRM records.
• Run: ask Claude for one script that joins the files by post URL and date, so next month's report is the same script on new data.
• Output: reports/YYYY-MM.md. If you sign in with a claude.ai account, Claude Code can also publish it as an Artifact, a private interactive page you can open or share.
• Reach check: for a quick read on reach before the details, use the LinkedIn impressions calculator.

Why it matters: better reporting gives you a reason to repeat an approach or change it, and it keeps uncertainty visible instead of turning patchy tracking into a confident revenue claim.

12. How do you turn your edits into a better system?

Every draft you edit says something about your standards. Write the recurring lessons down so you spend less time fixing the same problems.

Compare the original drafts, my edits, the versions I published and the performance data available for [period]. Find: voice corrections I keep making; claims that needed more evidence; ideas that got clearer once I reframed them; formatting changes that made posts easier to read; topics that led to useful conversations with clients. Suggest updates to the voice guide, the research briefs and the editing checklist. For each proposed rule, give an example, explain what it's for and say when an exception makes sense. Keep style preferences apart from guesses about performance. Never turn one post that did well or badly into a permanent rule. Give me a short change log to approve and one controlled experiment for next month.

In Claude Code:

• Input: what you really posted, saved as published/YYYY-MM-DD-slug.md next to its draft. If the folder is a git repository, each edit is in its history as well.
• Output: changelog.md with the proposed rules, an example for each and when an exception applies.
• Apply: once you approve, Claude can edit context/voice-guide.md, the /linkedin-draft skill and CLAUDE.md. A correction you keep repeating belongs in CLAUDE.md, not in a one-off chat message.
• Keep apart: performance guesses go in experiments.md and stay out of the rules, so a single good post never hardens into policy.

Why it matters: this is how the system gets easier to run over time. Your corrections turn into written guidance, and performance ideas stay open to testing.

How should you roll out all 12 prompts?

Add the prompts in stages so you can check each output before taking on another recurring job.

• Week 1, the foundation. Load your business context, run prompt 1 and compare the voice guide with writing you would actually publish. In Claude Code terms, the folder, CLAUDE.md and context/ exist by Friday.
• Week 2, the research backlog. Run prompts 2-4 on a manageable set of customer conversations, market sources and LinkedIn examples. Fill calls/, context/sources.md and data/swipe.csv before you begin.
• Week 3, the first batch. Use prompts 5-7 to turn approved material into a calendar and a few drafts. By the end of the week, /linkedin-calendar and /linkedin-draft exist as skills.
• Week 4, the review. Run prompt 8, publish what you approve, work the first hour on each post and note the changes you make. Let a subagent handle the review and keep its files in review/.
• Weeks 5-6, conversations. Add prompts 9 and 10 once you have enough sales context and inbound activity to make them useful. If your CRM can connect to Claude Code, set that up now.
• Every month after that. Run prompts 11 and 12 to review results, improve the instructions and pick the next experiment. The prompt 11 script just runs again on the new month's data/ folder.

For each recurring job, write down its inputs, where the output goes, how often it runs and who owns it. Test one run before trusting the schedule, and have Claude Code report any missing access or unfinished work.

Automating it in Claude Code. When you sign in with a claude.ai account, /schedule sets up scheduled runs in the cloud that execute on Anthropic's side, with your laptop closed. Without that sign-in, /loop repeats a command at an interval inside an open session. For your own scheduler, claude -p "..." runs one prompt without the interactive interface, so a weekly job can run prompt 3 and save the result to research/; GitHub Actions is another route if the folder lives in a repository. All of these produce files for you to review, and none of them publishes to LinkedIn.

You don't need 12 separate agents on day one. Begin with the job that eats the most time and produces something you can reliably judge.

What can't Claude Code do for you?

Claude Code can organize research, pull out useful evidence, prepare drafts and learn from your edits. The quality of the posts still comes from the experience and judgment you put into the system.

That means holding a real opinion, being willing to explain it, and noticing when a polished draft says something you would never defend face to face. A folder of well-organized files doesn't change that; it just makes your judgment easier to apply every week.

Once a post from this folder is live, give it a first hour with some activity: share it with people you know, and with a free LinkedInLikes account if your circle is small. Otherwise, load your context and run prompt 1, because better inputs make every later step more useful.

More articles

• How to build an AI content system for LinkedIn from one empty folder (2026 setup)
• The 2027 LinkedIn marketing playbook
• LinkedIn maxxing with Claude
• LinkedIn maxxing with Codex