Most AI-written LinkedIn posts have a tell. They start with "In today's rapidly evolving landscape." They use phrases like "leverage" and "synergies." They have a 3-bullet middle and a question at the end. People scroll past them. Here's the workflow for posts that read like you actually wrote them.
One-time setup. Find 10–20 of your past LinkedIn posts that performed well — meaning real engagement, not just impressions. Paste them into a Claude Project knowledge base.
Write a voice guide: sentence-length patterns you use, words you favor, words you'd never use, the kind of opener that hooks for you (story, contrarian take, specific stat — whichever maps to your style).
System prompt: "Match the voice in the example posts in the knowledge base. Avoid the AI tells listed in the voice guide. Vary sentence length. No exclamation points. No bullet-list middles. No \"What do you think?\" closes."
The four LinkedIn angles that consistently drive engagement:
1. Specific story. Something that actually happened with named details. Not "lessons learned" — the actual event, told as a story.
2. Contrarian take. A widely-held belief in your industry that you think is wrong, with the reason.
3. Specific number. A stat from your own work or research that surprises people. The number has to be real and you have to be able to defend it.
4. Behind-the-scenes detail. What you actually do during a specific work activity that people don't usually see.
Pick one. Tell Claude which one. Generic "thoughts on AI" posts default to category 5, which is "doesn't work."
Write a LinkedIn post using angle: [SPECIFIC]. Topic: [SPECIFIC]. Match the voice in the Project knowledge base. Constraints: - Open with a specific moment, stat, or contrarian sentence — never with "In today's landscape" or "I've been thinking about" - 150-250 words total - No bullet lists in the middle - No "What do you think?" close — use a specific question or a confident statement - 0 emojis - 1 specific detail that proves authenticity (named project, real number, actual quote) Draft 2 variants with different openers.
Read the draft out loud. Anywhere it sounds like a corporate communications team wrote it, rewrite.
Specific phrases to hunt and kill: "I've been thinking about," "It's time we talk about," "Let's be honest," "Here's the thing," "rapidly evolving," "navigate this complex landscape," "unlock," "leverage," "synergize."
Anywhere the post sounds like Claude wrote it, you have to inject voice — usually with a specific detail only you would know, or with a colloquial phrase you actually use.
Save the post + engagement numbers back into your Project knowledge base. Tag what worked.
After ~10 posts, you'll see patterns. The angles that worked. The openers that landed. The topics that resonated. Feed that signal back into your system prompt.
This is the difference between teams that ship LinkedIn content for 6 months and stop because nothing landed, vs. teams that ship for 18 months and build a real following: the second group treats each post as data, not a one-off output.
1. Letting Claude open the post. AI defaults to "In today's rapidly evolving landscape." Write the opener yourself; let Claude help with the middle.
2. Using generic AI-flavored prompts. "Write me a LinkedIn post about X" produces forgettable content. Specific angle + voice match is what makes it work.
3. No personal specifics. AI can't invent your real customer story or your actual number from last quarter. Those specifics are what differentiate your post from any other LinkedIn post on the topic.
4. Long bullet lists in the middle. They're the visual tell of AI-written posts. Use prose, not bullets, for body content.
5. "What do you think?" closes. Engagement bait that everyone recognizes. Use a specific question or a confident statement instead.
Yes — but only with voice training. A blank prompt produces AI-shaped output. A Project loaded with your 10–20 best past posts + a voice guide produces posts that sound like you wrote them. The training is the whole game.
For thought-leadership posts in B2B: 150–250 words. Long enough to make a point, short enough that scrolling readers actually finish. The exception: a true story-driven post can extend to 350–400 words if the story is compelling.
Personal opinion: not necessary as long as the post is genuinely your perspective and you did the editorial work. If the post is 90% AI-generated with no real voice or insight, the issue isn't disclosure — it's that the post is bad.
2–4 quality posts per week beats daily mediocre. The compound effect of LinkedIn is voice consistency over time, not volume.