How to write LinkedIn posts with Claude
Claude gets better at your LinkedIn posts once you stop starting from nothing each time. Here is how to set up a project, memory, a skill and a connector. A project holds your past posts and your rules, memory carries corrections forward, a skill runs your edit the same way every week, and a connector lets Claude work inside the tool that publishes. None of them changes the essentials: you still bring the material, check the facts and decide what goes out.
Give LinkedIn a project of its own
A project in Claude is a workspace with its own chats and its own knowledge base, and whatever you add to that knowledge is used in every chat inside it. Start small: eight to ten posts you wrote yourself and would publish again, a paragraph on the one kind of reader you write for, a short note on what you sell and what you believe about your field. Leave out anything you would not want imitated — one weak post in the knowledge is an instruction too.
Write instructions a draft can pass or fail
Project instructions apply to every chat in the project, so they are the place for rules — and a rule is only useful if a draft can be checked against it. “Professional but warm” cannot be. “British spelling. First line under 80 characters, then a blank line. No question as the last line. Never ‘excited’, ‘thrilled’ or ‘game-changer’. One idea per post; if my notes contain two, ask me which” can. Add the reason where there is one; Anthropic's own guidance is that knowing why a rule exists helps Claude apply it. Six or seven rules are plenty.
Bring one post's material to each chat
The project supplies the constants; you still supply the post. Open a new chat inside it, paste what happened — notes, numbers, the awkward detail — and ask Claude what is missing before it drafts: “What would a sceptical reader ask that these notes do not answer?” Answering three questions is quicker than repairing a draft built around a hole. Keep to one chat per post, so that each starts from the project's knowledge and instructions rather than from the full text of last week's draft.
Use memory for corrections, and read what it kept
Claude builds a memory from your chats and brings it into new ones, and each project has a memory of its own. Use it for corrections: when you fix the same thing twice — it keeps ending on a question, it calls your customers “clients” — say “remember this”, and the next chat starts with it. Memory is not a black box: Claude's settings list what it remembers, topic by topic, and you can edit or delete any of it. Check now and then; a wrong memory quietly bends every draft that follows.
Turn your edit into a skill
A skill is a set of instructions Claude loads when a task calls for it: a procedure, where the project is background. The one worth having for LinkedIn is your own edit — check the first line against the fold, cut sentences that repeat the one before, flag every figure that is not in your notes, list any banned phrase, then suggest changes. A simple skill is a short text file — a name, a line saying when to use it, the steps — packaged and uploaded once in Claude's settings (skills need code execution switched on). It takes no programming, but it is the most technical step here, and the same checklist at the end of your project instructions is a fair substitute. Ask for it by name: “run my LinkedIn edit on this draft”.
Add a connector for what Claude cannot do alone
On its own, Claude does not know how your posts performed, cannot put one on LinkedIn and, by Anthropic's own account, does not generate photos or illustrations. It will not always say so, either: the help centre warns that Claude can claim abilities it does not have when no tool for them is connected. Connectors close the gap. They let Claude read from, and act in, a service you have signed in to, and custom connectors are available on every plan, the free one limited to a single connector. With MediaPilot connected — MCP is part of its Growth and Scale plans — Claude can read your topic pool and analytics, save a draft with its image to your workspace and schedule it for a named account. The server address is https://mcp.media-pilot.io/mcp.
Keep the three jobs that stay yours
No set-up supplies the material, verifies it or decides. The material is what happened in your work this week, and no project knows it until you say it. The facts need checking: Claude can produce a quotation or a figure that sounds right and is not, and Anthropic itself advises against treating it as a single source of truth. And the decision to publish is the one step worth doing slowly — read the final text once at phone width, as the feed will show it. The free preview at media-pilot.io/tools/linkedin-post-preview shows where the post gets cut.