How to train Claude with your best LinkedIn posts

Authors observed: Spanish AI consultants, trainers and independent advisors. Verified problem: 8 months of activity, 5-30 LinkedIn posts, 180-500 followers, 1-3 reactions per post. Solution: train Claude with your LinkedIn posts that work. Just once. Then Claude validates every new post.

The problem: you post on LinkedIn but nobody sees you

You have been posting on LinkedIn for months. It's not a lack of discipline. You post every week, sometimes every other day. You invest 30-45 minutes per post: you research, structure and review. Then you publish.

And nothing.

Two reactions. One from a bot. No comments. No conversation. No private message from an interested prospect.

Meanwhile, you see other freelancers with less experience getting 50 or 100 reactions with similar posts. Their content is not better. Their industry is not different. The difference is brutal: they have a voice. You have a format.

The problem? Your posts sound like a corporate brand. And LinkedIn is trained to ignore them.

When LinkedIn sees "5 AI strategies to transform your business", it classifies it as corporate noise. When it sees "Yesterday I told a client that AI was expensive. Here's what happened", it classifies it as an authentic conversation. One is invisible. The other goes viral for your audience size.

Why this happens: it's not an accident

1. You confuse content with voice

Content = what you say (topic, structure, data). Voice = how you say it (tone, point of view, risk, honesty). Most freelancers optimize content: "5 steps" frameworks, "3 reasons", "the inverted pyramid". They apply the structure, publish, and nothing.

Ruben Hassid (340,000 LinkedIn followers, €0 in ads) says it plainly: "If you train Claude on bad posts, you get bad posts. Voice is the input." Your input is generic structure. Your output is noise.

2. You over-optimize for "professionalism"

You are afraid of sounding "too personal". So you sand down every sentence until it offends nobody, surprises nobody and provokes nobody. Result: posts so polite that they are invisible.

LinkedIn prioritizes engagement. Engagement comes from friction: surprise, disagreement, emotion, vulnerability, risk. A post that says "here's my uncomfortable truth" generates 10x more conversation than one that says "here's the consensus you already knew".

Real example:

• "Professional" version: "AI implementation requires cultural change. Here are the steps."
• Authentic version: "I've watched 12 companies fail at AI because their leaders think it's a software button. It's not. Here's what's missing."

The second generates conversation. The first generates silence.

3. You don't audit your own voice

You publish 15 posts. Some work (8-15 reactions, 2-3 comments). Others don't (1-2 reactions). What is the difference? If you don't analyze it, you repeat the bad ones. And you blame the algorithm.

The truth: the posts that work are those where your corporate voice disappeared and yours appeared. But without an audit you don't see the pattern, so you go back to the safe formula because at least it "looks professional".

4. You confuse "teaching" with "sounding authentic"

Many freelancers assume: "My posts have to be educational, so I have to sound like a distant expert." False. The most educational posts come from real experience, not from a manual. "Here's what I discovered two weeks ago with a client" is 10x more educational than "the 5 steps of a framework". Because one is verifiable. The other is theory.

The solution: train Claude with your best LinkedIn posts

Today you write without voice validation:

1. You pick a topic (an idea that "should" work, not necessarily yours)
2. You structure (outline, headline, closing)
3. You write (you fill in the structure)
4. You publish (hoping it works)

There is no voice validation. Without validation, you repeat mistakes. The solution is to train Claude. But not as a generator: as a validator.

You will train Claude with your posts that ALREADY work. Then, before publishing each new post, you run it through Claude. Claude validates: "Does this keep your voice or are you being corporate again?" This is different from "Claude writes my posts". This is "Claude validates that I am following my winning formula".

How to train Claude: a 4-step system

Step 1: extract your best posts (2 minutes)

Download your last 15-20 posts with reactions, comments and timestamps. If you don't know how:

• Option A: use Apify (€0 with free credits, 2 minutes, .xlsx download)
• Option B: LinkedIn Data Export (settings → data privacy, slower)

Result: an .xlsx file with your posts, reactions and comments.

Step 2: upload to Claude and ask for an audit (30 minutes)

Open Claude, upload the .xlsx file and copy/paste this prompt:

Claude gives you back:

• Report: what works in your posts
• SOP: your personal formula
• Pattern: winning angle + tone + structure

Step 3: create a reusable Claude skill (15 minutes)

In the SAME chat (don't close it), ask for this:

Claude creates the skill. You save it. This skill is your "Claude trained on your posts".

Step 4: use the skill before every post (2 minutes)

Before publishing, you invoke:

Claude gives you validation in 30 seconds:

• ✅ Follows your voice
• ⚠️ Here you sound corporate again
• 🔧 Rewrite these sentences to sound like you

You adjust. You publish.

The complete system: train + validate with Claude

Week 1:

• Monday: extract your posts (2 min)
• Tuesday-Wednesday: train Claude with your posts (45 min)
• Thursday: create the skill (15 min)

Week 2 onward:

• Every new post: validate with the skill (2 min)
• Expected result: posts that sound like you, not like AI

What NOT to do: common traps when training Claude

Trap 1: "Now Claude writes my posts"

No. Claude doesn't have your voice. If you use it to generate posts from scratch, you get corporate noise. You train it with your posts that DO work. Its job is to validate, not to invent.

Trap 2: "I'll post EVERY day now"

No. Quality > quantity. Ruben Hassid posts twice a week and has 340,000 followers. Better: one weekly post with an authentic voice than seven generic ones.

Trap 3: "It works without data"

No. You need at least 5-10 published posts. If you start from zero, there is no voice to train. Training Claude with 2 posts doesn't work: the machine needs a pattern.

Trap 4: "My best posts are the ones with the most likes"

Not always. A post with 8 reactions but 3 valuable comments is better than one with 20 passive likes. When you train Claude, look at comments, not just likes.

Real case: training Claude worked

AI consultant, Madrid, 6 months on LinkedIn, 0 leads generated.

Before:

• 12 posts, 200 followers, 1-2 reactions per post
• Typical posts: "5 keys to AI", "Why digital transformation fails", "Success story: company implemented AI"
• Tone: professional, distant, no clear opinion
• Result: invisible

Trains Claude:

• Extracts 12 posts (Apify, 2 min)
• Uploads to Claude and asks for an audit (30 min)
• Creates the skill "validate-linkedin-consultant" (15 min)

Pattern identified:

• Best post: a real anecdote ("Yesterday a client told me AI was expensive. I showed him an invoice for manual errors: €4,000 a month.")
• Pattern: anecdote + specific number + clear opinion
• SOP: anecdote → number/evidence → why it matters → question

After (week 5):

• 8 new posts validated with the skill
• Average: 25 reactions + 4-5 comments
• 3 inbound leads from qualified prospects
• 1 client closed in week 8

The change: not more content. The same number of posts. An authentic voice instead of a corporate one. All because he trained Claude with his real posts.

The uncomfortable truth

Training Claude with your posts doesn't work if your real voice isn't differentiated.

If you are one more consultant among 50 saying the same thing, training Claude won't save you. You first need to discover what you think differently. Then you train Claude to validate that you publish it unfiltered.

But if you already have an opinion, a point of view, a truth you believe but don't say, training Claude works. Because the only thing missing is courage and validation. Claude gives you the validation. The courage is yours.

What to do now: train Claude today

Today (15 minutes): bring your last 10 posts. Identify which ones worked best. What do they have in common?

This week (1 hour total):

• Monday: extract posts with Apify (2 min)
• Wednesday: upload them to Claude and ask for the analysis (30 min)
• Thursday: create the skill (15 min)

Next post (2 minutes): use your skill. Validate against the pattern. Does it follow the pattern or are you being corporate again?

Next 4 weeks: publish 4 posts validated with the skill. See what happens. If engagement goes up, you now have a system: Claude trained on your best posts, ready to validate every new post.

Frequently asked questions

How many posts do I need to train Claude? At least 5-10 published posts with real interaction. With less data there is no voice pattern for the AI to identify.

Does Claude write my LinkedIn posts? No, and it shouldn't. The system uses Claude as a voice validator, not as a generator. Posts written by AI from scratch sound corporate and the algorithm ignores them.

How long does it take to set up the system? About 1 hour in total: 2 minutes to extract posts, 30 for the audit and 15 to create the skill. After that, each validation takes 2 minutes per post.

What tools do I need? A Claude account and a way to export your posts (Apify or LinkedIn's native data export). No risky automations or mandatory paid tools.

What results can I expect? In the reference case: from 1-2 reactions to 25 reactions and 4-5 comments per post in 5 weeks, with 3 inbound prospects and one client closed in week 8. The result depends on your voice being truly differentiated.