How Do You Write LinkedIn Posts with AI Without Sounding Like a Robot?

Ground the AI in your own data. Flux's MCP tools connect your posting history to AI assistants like Claude, Cursor, and Claude Code, so the model writes about topics your audience actually rewards, in your voice, and you can score the draft against your own performance before publishing. The result is AI-assisted, not AI-generated: you supply the perspective and data, the AI handles structure and first drafts.

What you'll learn

Why most AI-generated LinkedIn posts fail and how to fix it
How to find topics worth writing about with get_topic_insights
How to generate data-informed angles with suggest_angles
How to make Claude write in your actual voice using get_posts
How to score a draft before publishing with score_draft
How to pick the right time and format for the post
The complete writing workflow inside an AI assistant

Why do most AI-written LinkedIn posts fail?

Generic prompts produce generic posts. When you ask an AI to "write a LinkedIn post about leadership," it has no idea who your audience is, how you write, or what's worked for you before. So it falls back on the average of everything it's seen: tidy three-point lists, an em-dash-heavy hook, a closing question that asks for engagement. Readers have learned to recognize that shape, and they scroll past it.

The problem isn't AI itself, it's writing without data. An AI assistant is a strong drafting partner once it knows your topics, your voice, and your track record. The fix is to feed it that context before it writes a single word, which is exactly what Flux's MCP tools do. They expose your posting history and performance data to any MCP-compatible AI client, so the model writes from evidence instead of averages.

Step 1: Find what works with `get_topic_insights`

Before writing anything, pull your topic performance so you write about themes your audience already rewards. get_topic_insights ranks your recurring topics by engagement and shows how each one performs against your own baseline. Because Flux measures engagement (likes, comments, and shares), not impressions, this reflects what people actually reacted to.

The signal is simple. If "hiring" runs 1.5x your baseline but "culture" sits at 0.7x, point the AI at hiring and away from culture. Asking Claude to write about a 1.5x topic starts you from a theme your audience has already validated, instead of a guess. Run this first so every later step is anchored to a winning subject.

Step 2: Generate data-informed angles with `suggest_angles`

Once you've picked a topic, suggest_angles turns it into five concrete angles built from your top-performing posts on that theme. You give it a topic and it pulls your relevant high-engagement posts as context, so the suggestions reflect what's worked for you rather than a generic brainstorm. Each call costs 5 credits.

For example, suggest_angles({ topic: "remote work" }) might return an angle like "The hidden cost of async meetings — why your remote team is burning out on Slack." That's specific, opinionated, and grounded in your style, which is a far better starting point than "share your thoughts on remote work." Pick the angle that matches what you actually believe and carry it into the next step.

Step 3: Write in your voice with Claude

Flux's MCP tools work in Claude Desktop, Claude Code, Cursor, or any MCP-compatible AI client, so you can do the writing right where you already work. The key to your voice is get_posts, which pulls your recent high-performing posts so the AI can see your real writing patterns: your typical length, tone, structure, and hook style.

Feed those posts to the model as style examples. A prompt like "Write a LinkedIn post about [the angle from step 2], matching the voice and style of these posts: [paste your top posts]" gives the AI both the substance and the style to imitate. Because it's working from your actual writing rather than a generic LinkedIn template, the draft comes back sounding like you wrote it on a good day.

Step 4: Score before publishing with `score_draft`

Don't publish on faith. Paste the draft into Score My Post at fluxgraph.app/score, or call the score_draft MCP tool, to get a predicted engagement percentile against your own history. Scoring is free and iterative, so you can score, revise, and score again until the prediction lands where you want it.

The output is built to drive revisions. You get a factor breakdown showing what's helping or hurting the post (length, timing, topic, format), similar past posts for comparison, and concrete improvement suggestions. Treat a low score as a prompt to tighten the hook or adjust the format, not a reason to scrap the post, then re-score the new version.

Step 5: Pick the right time and format

A strong post still underperforms at the wrong time or in the wrong format. Check your Heatmap for the times your audience is most active, and use the Content Type Breakdown to see which format (text, image, carousel) earns you the most engagement. Write toward the format that wins for you, not the one that's trendy.

score_draft makes timing part of the decision: it accepts day_of_week and hour parameters, so you can score the same draft for a few candidate slots and see which one the model prefers. Pair that with your Heatmap and you're choosing a posting time backed by your own data instead of conventional wisdom.

The complete workflow in Claude

Put the tools together and the whole process runs as a short conversation inside your AI assistant:

"What topics are working for me?" → get_topic_insights
"Give me angles on [top topic]" → suggest_angles
"Write a post in my voice about [angle]" → Claude, with get_posts for style examples
"How will this perform?" → score_draft
"When should I post it?" → get_posting_insights or check the Heatmap

Each step hands its output to the next, so you move from "I have no idea what to post" to a scored, voice-matched draft with a posting time in one sitting. The same flow works in Cursor and Claude Code, since they share the same MCP connection.

Keeping your voice authentic

The goal is AI-assisted, not AI-generated. Use the AI for ideation and first drafts, then edit heavily: cut the lines that don't sound like you, sharpen the ones that do, and add the specific detail only you would know. Your unique perspective and experience can't be generated — the AI structures it, but you have to provide it.

This is also why score_draft is a useful guardrail rather than a crutch. Its model is trained on your posts, so it rewards drafts that match your proven style and flags the ones drifting toward generic. When the AI overshoots into corporate-speak, your edits and the score pull it back toward something that reads like you.

Frequently asked questions

Can AI write LinkedIn posts that don't sound AI-generated?

Yes, if you ground it in your own data. Generic prompts produce generic posts, but feeding the AI your top-performing posts as style examples (via get_posts) and a data-backed angle (via suggest_angles) gives it the substance and voice to write something that reads like you. Editing the draft yourself closes the gap the rest of the way.

Which AI tools work with Flux?

Flux's MCP tools work in any MCP-compatible AI client, including Claude Desktop, Claude Code, and Cursor. Once connected, the AI can call tools like get_topic_insights, suggest_angles, get_posts, and score_draft directly in your conversation.

Does Flux write the post for me?

Flux supplies the data and scoring, not the prose. It surfaces your winning topics, generates angles from your best posts, exposes your writing style to the AI, and predicts how a draft will perform. The AI drafts and you edit, while Flux keeps both grounded in your actual performance.

How much does this cost in credits?

suggest_angles costs 5 credits per call. get_topic_insights, get_posts, and get_posting_insights are standard read tools, and score_draft (and the Score My Post page) is free, so you can score and re-score drafts as much as you like.

Does Flux use impressions to score my draft?

No. Flux measures engagement (likes, comments, and shares), not impressions, because impressions aren't reliably available across LinkedIn. Every prediction and recommendation is grounded in what your audience actually reacted to.

On this page