Why Your AI-Generated LinkedIn Posts Sound Like AI (And How to Fix It)

The real problem? You're using the wrong AI tool for the wrong task — and you haven't given any of them enough of yourself to work with.

After years of building AI content tools since GPT-3 launched, testing endless configurations, and eventually moving our entire platform from GPT to Claude, I've learned something that took a long time to accept: the question "which AI writes the best LinkedIn posts?" is the wrong question.

Here's what actually works.

The Generic AI Content Problem Everyone's Facing

Here's what most people do. They open ChatGPT, type "write me a LinkedIn post about leadership," hit enter, and get something like this:

Leadership isn't about having all the answers. Real leadership shows up when the plan breaks, the team disagrees, the outcome is uncertain.

It's not terrible. It's formatted nicely. You could technically post it. But it's completely generic.

It's the kind of writing that fills a very large share of AI-generated LinkedIn content. It's all true. It's also all been written a million times before. Most importantly, it has nothing to do with you or your actual experience of leading anything.

So how are you supposed to build a personal brand out of phrases with no connection to your own expertise?

Can Better Prompts Fix This?

You might think you just need a better prompt. So you try:

"Write in a professional but conversational tone. My target audience are founders and entrepreneurs. Make it engaging."

The result is marginally better. Some formatting improvements. Maybe the hook now targets founders: "Most people think leadership is about having the right answers. Founders know better."

But it's still generic. Anyone could have written it. There's no connection to you, your experience, or your particular perspective — and it will do nothing for your personal brand.

This is the context problem, and it's why most people conclude that AI content doesn't work for them.

The Real Answer: Each AI Excels at Different Tasks

After a lot of testing, here's where I landed: "which AI is best?" is unanswerable. "Which AI is best for this specific task?" is not.

Each model has distinct strengths. What follows is our own working view from building on all of them — not a benchmark.

Claude: the best writing model, by some distance

Use Claude when you need words that sound like a human wrote them.

• Natural flow and tone that doesn't announce itself as AI
• Cleaner prose without the telltale patterns
• Better voice preservation
• Less reliance on buzzwords and overused constructions

When we first built ContentIn we used GPT. We eventually switched the writing path to Claude — same prompts, different model — and the difference was immediate and one-directional. The output stopped reading like a competent assistant and started reading like a person. We've never considered switching back.

ChatGPT: the strategy engine

Use GPT when you need to work out what to write about.

• Brainstorming angles and post ideas
• Building content frameworks and structures
• Analytical thinking about your content strategy
• Digging through analytics data
• Finding posts that match your topics and style

GPT is strong at the strategic, analytical end even where it isn't the best writer.

Gemini: the visual one

Use Gemini for images and infographics. Its recent image models are what finally made AI-generated infographics good enough for us to ship — something we'd wanted to offer for two years and couldn't.

The Real Problem: You Need All Three, Plus Proper Context

Here's where it gets impractical. To create the best possible LinkedIn content on your own, you'd need to:

1. Use GPT to brainstorm angles and build your strategy
2. Use Claude to write the posts, with proper context
3. Use Gemini for visuals and infographics
4. Stitch it all together
5. Schedule and post it
6. Analyse the results
7. Feed those learnings back into your prompts

And you need the right prompts for each, carrying context that actually sounds like you.

For a busy founder, this isn't realistic. Nobody is actually doing all of it. Depending on your copywriting, you might not even be faster than writing the post yourself.

The Context Engineering Problem Nobody Talks About

Even if you pick the right model for each task, the context problem remains. This is the part most people miss, and it is a genuine skill in its own right.

When you tell an AI to "write in a professional but conversational tone," you have just described most of LinkedIn. The model has nothing distinctive to work with.

And here's the awkward part: you don't actually know the patterns in your own writing. You think you write a certain way. Your published posts may say something else entirely.

• When you describe yourself to an AI, you describe an idealised version
• When an AI reads your actual content, it sees what's really there

That gap is the whole problem. To close it, the AI needs your actual voice rather than a generic tone setting, your real patterns rather than your self-description, your specific phrasing, LinkedIn-specific structure rules, and guard rails against the telltale AI tics.

Which raises the obvious question: what would a real voice profile actually contain?

What a Real Voice Profile Contains

Most AI writing tools ask three questions about your voice:

• Professional or casual?
• Short posts or long?
• Who's your audience?

That's not a voice profile. That's a costume. The AI puts it on and writes something that vaguely resembles your direction, and it still isn't you.

What we build instead is Voice DNA: a structured profile of you as a writer, extracted in stages from your own material. Not just tone. Not just style. The whole structure. Among the dimensions it captures:

• Core beliefs: what you actually stand for professionally
• Audience pain points: what your readers are dealing with
• Vocabulary: the specific words you reach for
• Avoidance list: words and phrases you would never use
• Jargon level: how technical you're willing to get
• Opening patterns: how you hook a reader
• Closing patterns: how you land a post
• Humour style: whether you use it, and what kind
• Vulnerability: what you share and what you protect
• Formatting identity: emoji use, line-break style, down to whether you break between every sentence or only between paragraphs
• Substance anchors: the numbers, projects and proof points that are actually yours

Most of this you have never consciously decided. You don't sit down and rule that you'll use dry wit but never sarcasm, or that you open most posts with a question. But run the analysis across your posts and the pattern is there every time. That's your voice — not a tone setting.

Where the data comes from

• Your actual writing and past posts
• Your conversations with the AI
• Your editing patterns — what you change before publishing
• What actually performs once it's out there

You could map all of this by hand. Go through your last 20–25 posts and write down every pattern you notice; that genuinely works. We built something to do it automatically and update it every time you write, edit or publish.

The Four Stages of Voice Confidence

The profile carries a confidence score, and the product tells you in plain language where you are. The bands are worth understanding, because each one unlocks something different.

Stage 1: "I'm starting to sound like you"

The foundation is in — tone, basic vocabulary, post-length patterns. Show a post from this stage to someone who doesn't know your writing and they might believe it's yours. Someone who does know it won't. You'll want to edit and sharpen.

The trick: actually do that editing. It's what moves you up.

Stage 2: "Your voice is coming through clearly"

Formatting identity is locked: how you use emoji, or whether you avoid them entirely; your line-break style; whether you open on a question or a statement. Someone who knows your writing starts to recognise it here.

Stage 3: "I could probably fool your colleagues"

This is where it gets specific, because the substance anchors are in: the numbers you reference are present and correct, the company names and projects are real, the proof points are ones you actually own. Your avoidance list is locked, too.

The AI stops writing "leverage" if you always say "use." It stops writing "game-changer" if that phrase makes you wince. Posts begin to feel like you wrote them rather than like a capable assistant wrote something in your general direction.

Stage 4: "Practically indistinguishable"

The full picture. Vulnerability ceiling calibrated, story patterns locked in, every edit you've made absorbed back into the profile. At this point the AI is correcting itself before you do — not mimicking your style, but carrying enough signal to make judgement calls in your voice.

How to Get There Faster

Four things, in roughly this order.

1. Do the onboarding interview

It sets the foundation: your professional context, what you stand for, what you're writing for. It takes about ten minutes, and people skip it because it looks optional. It isn't.

2. Keep writing with the AI — and keep editing

Every edit is a signal. Change a word, cut a paragraph, rewrite a hook, correct a number, and all of it is captured. We're not only watching what you publish; we're watching what you change before you publish. That edit data is some of the richest signal there is, which means the more you edit, the smarter the profile gets. Your corrections aren't wasted effort — they're the training.

3. Keep talking to the AI

Not only to write posts. Drop in a hot take. Explain a framework you use. Describe a conversation you had last week. There are prompts built into the chat to pull this out of you, but you don't have to wait for them.

Those off-hand bits — the half-formed opinions, the things you'd say to a colleague over coffee — fill the substance layers that analysing your published posts alone will never reach.

4. Upload your knowledge resources

Decks. Old articles. Interview transcripts. Newsletters. Anything with your thinking in it. That's what populates the substance anchors: the specific ideas and proof points that make your writing distinctly yours.

The Bottom Line: AI Is What You Put In

If AI doesn't sound like you, it's because you haven't given it enough of yourself. That's the whole explanation.

The model is no longer the limiting factor. Every serious tool is building on roughly the same underlying models. What differs is how much of you is in the context.

Most people do step one and then wonder why the output still feels flat. A voice profile built on five posts is a rough sketch. One built on twenty-five posts, hundreds of small edits and a proper onboarding interview is something else entirely.

Garbage in, garbage out. Gold in, gold out. That isn't motivational — it's just how the data works.

A Better Way: Let the Right Tool Handle the Complexity

Or you could use something that has already solved this. To make it easy to see the difference, we built a free tool — as much as we could reasonably pack into one.

1. Paste in your LinkedIn profile URL
2. Click "Generate Posts"
3. The AI reads your profile and your existing posts
4. It writes new posts aimed at your ICP, knowing what you do and learning from your style

When I tested it on myself, it immediately picked up on quirks I hadn't mentioned — including the lowercase formatting I'd been playing with:

most people think they need perfect timing to start posting. your competition isn't other creators. it's the voice in your head saying you're not ready.

It already knew what I write about. It addressed my ICP's problems. And it sounded like me.

That's the starting point, not the destination — the free tool reads your profile once. Compare it against a raw ChatGPT post and the difference proper context makes is obvious.

Try the free LinkedIn post generator

Want the Full System?

If you want the profile to keep learning — improving with every post, plus scheduling, carousels, rich media and analytics — that's the full product. At the heart of it is the AI Ghostwriter, which is where Voice DNA actually gets used:

• Multiple models working together, each doing what it does best
• Claude for writing, GPT for strategy, visual models for carousels and images
• Orchestrated automatically, so you're not switching platforms or maintaining prompts
• One tool that knows which model to use for what
• You focus on being you, give it input, and let the system do the assembly

With ContentIn I'm down to a few minutes per post, media and idea generation included. You can see what's in each plan on the pricing page.

Your Challenge This Week

Stop using raw ChatGPT for your LinkedIn posts.

At minimum, try the free tool and see what proper context does to the output straight away. It takes about a minute to generate your first contextual post. If you're still copying between three AI tools trying to get something usable, there's a better method.

And if you do sign up: do the onboarding interview. Ten minutes at the start is worth more than any prompt you will ever write.

Start your free trial at ContentIn →