strategy18 min read

How to Use AI to Write LinkedIn Comments That Sound Like You (Not a Robot)

FliesReplies Team

May 5, 2026

You copied a ChatGPT reply into a LinkedIn comment once. You pasted it in, re-read it, and cringed. Maybe you posted it anyway. Maybe you deleted it. Either way, you knew — that's not how I talk.

You're not alone. Thousands of LinkedIn consultants, coaches, and founders have been through the exact same moment. And most of them walked away thinking: "AI just doesn't work for this."

That's the wrong conclusion. The right conclusion is: generic AI doesn't work for this.

There's a meaningful difference. And once you understand it, LinkedIn commenting gets a lot less painful.

This post is for you if:

You've tried using ChatGPT (or similar tools) to write LinkedIn comments and hated the output
You spend more time than you'd like staring at posts trying to think of something genuine to say
You're active on LinkedIn for business reasons and understand that comments matter — they build relationships, surface your expertise, and drive profile views
You want AI to help, but only if it actually sounds like *you*

Let's get into it.

Why ChatGPT-Generated Comments Sound Like ChatGPT

Before we talk about the fix, let's be precise about the problem. Because "it sounds robotic" is too vague to be useful.

Here's what's actually happening when you ask ChatGPT to write a LinkedIn comment and it comes out wrong.

1. It Doesn't Know Your Voice

ChatGPT is a general-purpose language model. It was trained on a vast range of text from the internet — not on your writing specifically. When you give it a prompt like "write a comment on this LinkedIn post about supply chain disruption," it has absolutely no reference for how you sound.

Does you tend to open comments with a direct assertion? A question? A personal anecdote? Do you use casual contractions or more formal syntax? Do you swear occasionally? Do you reference your specific industry background? Do you pepper in dry humor?

ChatGPT doesn't know any of this. So it defaults to a middle-of-the-road "professional and engaging" register that nobody actually talks in.

That register sounds like: "Great insights here! Totally agree that resilience is key. In my experience, companies that prioritize agility tend to outperform. What are your thoughts?"

It's not offensive. It's just not you. And on LinkedIn, where the entire point is building a recognizable personal brand, "not you" is a real problem.

2. It Defaults to "Professional and Engaging" — Which Is the Opposite of Distinctive

Here's the irony: the thing that makes ChatGPT so capable — its ability to be helpful across every possible context — is exactly what makes it bad at sounding like a specific person.

It's been trained to produce output that would be acceptable to the largest possible audience. It avoids strong opinions, hedges when uncertain, and gravitates toward phrases that signal engagement without actually engaging. Words like "insightful," "actionable," "key takeaway," and "great point" show up constantly because they're safe.

On LinkedIn, safe is invisible. The people who build audiences on this platform are the ones with a distinct perspective, a recognizable cadence, an unmistakable style. "Professional and engaging" is the enemy of that.

3. It's Trying to Sound Good — Not Sound Like You

When you give ChatGPT a prompt without voice context, its implicit goal is to produce a response that a reader would evaluate positively. It's optimizing for quality in the abstract.

But the bar you're actually trying to clear is different: does this sound like something I would write? A comment can be objectively well-written and still be completely wrong for your voice. The model doesn't know which one you need, so it guesses — and it guesses "polished" every time.

4. It Has No Context About You, Your Relationships, or Your Niche

LinkedIn comments aren't just replies to posts. They're often:

Continuing an ongoing conversation with someone you know
Demonstrating expertise in a specific niche to a specific audience
Signaling your positioning (what you stand for, what you push back on)
Nurturing relationships with potential clients or collaborators

ChatGPT knows none of this. It doesn't know that you've been commenting on this person's posts for three months. It doesn't know that you have a strong POV on this topic that you've written about before. It doesn't know that your audience is mid-market operations leaders, not startup founders.

Generic AI produces generic output because it's operating without the context that makes a comment meaningful.

The Core Insight: The Problem Is Generic AI, Not AI

This is the reframe that changes everything.

The frustration you felt with ChatGPT wasn't evidence that AI can't do this. It was evidence that AI without your voice data can't do this.

Voice-trained AI is a completely different experience.

When an AI tool has been given examples of your real writing — dozens of comments you've actually posted, calibrated against feedback about which suggestions felt right and which didn't — it's no longer guessing at your style. It's working from a model of how you communicate.

The output shifts from "sounds like a polished LinkedIn user" to "sounds like me on a good day."

That's the goal. Not AI writing instead of you. AI that's learned to draft the way you write, so you can review and post in seconds instead of minutes.

What "Voice Training" Actually Means

The phrase "voice training" sounds more technical than it is. Here's what it means in practical terms.

Giving the Model Examples of Your Real Writing

The foundation of any voice-trained approach is examples. Not instructions about your style — actual samples of you writing.

There's an important distinction here. Telling an AI "I write in a casual, direct tone and I like to push back on conventional wisdom" is far less effective than showing it 20 comments where you actually do that. Language models learn by example, not by description.

Good examples for voice training are:

Comments you've actually posted on LinkedIn (not drafts you decided against)
Comments that you felt good about after posting — ones that got replies, reactions, or private messages
Comments across the range of topics you write about
Comments that show different registers — when you're being sharp, when you're being warm, when you're being analytical

Bad examples are:

Comments you posted but felt were "meh" — they're not representative of your best voice
Comments that were unusually long for you (they'll skew the model toward length)
Comments on topics outside your normal range
Comments you've heavily edited to the point where you don't recognize your original voice in them

Setting Constraints (Content Pillars and Topic Focus)

Voice isn't just style — it's also substance. The topics you care about, the angles you take, the positions you hold.

Defining 3–5 content pillars gives AI suggestions something to anchor to. If you're a sales coach, your pillars might be: outbound strategy, mindset for quota carriers, sales leadership, the future of SDR roles, deal strategy. If you're a supply chain consultant, they might be: nearshoring, inventory optimization, supplier relationships, demand planning, risk management.

These pillars do two things:

They help the AI understand what a relevant comment looks like from you (not just a stylistically accurate one)
They prevent you from commenting on posts where you don't have anything genuinely useful to say

Creating a Feedback Loop So It Gets Better Over Time

The third element is iteration. A voice model that never receives feedback will plateau. One that gets signal — "this suggestion felt right, this one felt off" — will keep improving.

This is why thumbs-up/thumbs-down mechanisms in AI tools aren't just UX polish. They're how the system learns to distinguish between suggestions that are stylistically accurate and ones that are exactly right. Over time, the gap between "plausible for me" and "this is exactly what I would have written" should close.

Step-by-Step: How to Get Any AI Tool to Write More Like You

Even if you're using ChatGPT or another general-purpose tool, you can get significantly better results by engineering your inputs carefully. Here's how.

Step 1: Build a Voice Prompt

Before you ask ChatGPT to write anything, feed it a voice definition. This is a structured prompt that gives the model your style parameters.

Here's a template you can use right now:

I'm going to ask you to write LinkedIn comments in my voice. Before you do, here's how I write: STYLE: [e.g., direct, occasionally blunt, use contractions, short sentences, no exclamation marks unless I mean it] TONE: [e.g., confident but not arrogant, I push back on fluff, I avoid corporate jargon] TOPICS I CARE ABOUT: [e.g., B2B sales, founder-led selling, outbound strategy] THINGS I NEVER SAY: [e.g., "great insights", "totally agree", "actionable", "key takeaway"] THINGS I DO SAY: [e.g., I often open with a direct statement, I like to add a contrarian angle, I sometimes ask a specific question at the end] EXAMPLES OF MY ACTUAL COMMENTS: [Paste 5–10 real comments you've written] When I share a LinkedIn post with you, write 2–3 comment options in my voice. Keep them under 100 words unless the post warrants a longer response.

This won't be perfect — and it will drift over multiple conversations because ChatGPT doesn't retain memory across sessions by default. But it will be dramatically better than a cold prompt.

Step 2: Use the "12 Examples" Method

The 12 examples method is specifically about quality over quantity when it comes to voice samples.

Rather than dumping 50 mediocre comments into a prompt, curate 12 that you're genuinely proud of. These should be:

Comments where you felt you said exactly the right thing
Comments that generated a meaningful reply from the post author
Comments that prompted someone to visit your profile or reach out
Comments that you've re-read and thought "yes, that sounds like me"

Why 12? It's enough to establish pattern without overwhelming a prompt. More importantly, the curation process itself is valuable — you learn what "your voice" actually means by having to choose.

When you feed these 12 examples as style anchors, the AI has a much clearer target to hit.

Step 3: Be Specific About What the Post Is About

One of the most common prompt mistakes is pasting a post and saying "write a comment on this."

Instead, give the AI:

The post text or key argument
Your genuine reaction (do you agree, disagree, want to add nuance?)
Any personal experience or data point you want to include
The relationship context (do you know this person? Is this a prospect? A peer?)

The more context you provide, the more the output sounds like you actually engaged with the content — because the AI has something real to work with.

Example prompt:

Here's a LinkedIn post from a sales VP I follow. I want to comment in my voice. POST: [paste post] MY REACTION: I mostly agree with her point about discovery, but I think she's underselling the role of pre-call research. I've seen deals won and lost on how much prep the rep did before the call. I want to make that point without being dismissive of her take. RELATIONSHIP: I don't know her personally but I've commented on her posts before. Write 2 comment options in my voice (see style guide above). Keep it under 80 words.

Step 4: Edit, Don't Just Paste

Even with a good voice prompt and specific context, the AI output will probably need light editing. This isn't a failure — it's the workflow.

Read the suggestion out loud. Does it sound like you? Where does it break down? Make those edits. Over time, you'll notice patterns in what needs fixing, which helps you refine your voice prompt.

The goal is to spend 30 seconds editing rather than 3 minutes writing from scratch.

Why Dedicated Tools Outperform ChatGPT Prompts for This Use Case

The approach above will get you better results than a cold ChatGPT prompt. But it has real limitations.

Memory. ChatGPT doesn't remember your voice between sessions unless you explicitly recreate the context. Every new conversation, you're starting over. This means your "voice prompt" either lives in a document you have to paste every time, or you forget it and get generic output again.

Context window limits. Fitting a full voice definition, 12 example comments, a post, and your specific reaction into a single prompt gets unwieldy fast. The more context you add, the more the model can degrade at following all of it.

In-feed friction. You're on LinkedIn, you see a post you want to comment on. Now you have to: switch to ChatGPT, open your voice prompt document, paste everything together, iterate. By the time you've done all that, the moment has passed. Speed matters in comment-based networking — commenting early on a post that's gaining traction dramatically increases visibility.

No feedback loop. ChatGPT has no way to get smarter about your voice over time based on which suggestions you used. Each session is a clean slate.

Dedicated tools built specifically for LinkedIn commenting solve these problems architecturally:

Your voice is stored and applied automatically — no pasting required
The tool lives in your browser, adjacent to the posts you're reading
Feedback signals are captured and used to improve suggestions
The model is specifically fine-tuned for short-form professional engagement, not general writing

How FliesReplies Co-Pilot Handles Voice Training

FliesReplies is built specifically to solve the problem described in this post: helping LinkedIn users comment with their own voice, at speed, without the copy-paste-cringe cycle.

Here's how the voice training process works:

Step 1: Import your past LinkedIn comments. FliesReplies pulls in your comment history from LinkedIn. These form the raw material of your voice model. The more comments you've written, the richer the baseline.

Step 2: Curate 12 manual examples. Beyond the import, you select 12 specific comments that best represent how you want to sound. These are your anchor examples — the model treats these as the target output style.

Step 3: Define your content pillars. You set 3–5 topics that represent your professional focus. This ensures suggestions stay relevant to your expertise, not just stylistically similar to your writing.

Step 4: Comment naturally. Rate suggestions. When you encounter a LinkedIn post in your feed, the Co-Pilot surfaces 1–3 reply options directly in the feed. You read them, pick the one that fits (or edit it), and post. Thumbs up/down feedback trains the model — over time, it learns not just your general style but your preferences in specific contexts.

The result: after a few days of regular use, the suggestions start to feel like something you would have written yourself on a day when you had more time to think. Not a template. Not a placeholder. An actual draft that needs a light edit or nothing at all.

There's a free trial — 15 replies, 3 days, no credit card required — so you can see what the suggestions feel like before committing.

The Human Review Principle

One thing worth being direct about: you should always read an AI-suggested comment before posting it.

Not because the AI will necessarily produce something wrong. But because your name is on it. The person reading it doesn't know it was AI-assisted — they're assessing it as an expression of your thinking. That's a responsibility worth taking seriously.

The goal of a good Co-Pilot isn't to remove you from the process. It's to remove the blank page problem from the process. You still bring the judgment, the authenticity, the decision about whether a given post is even worth commenting on.

What changes is the work of going from "nothing" to "a draft." That's where most of the time goes. That's what good AI Co-Pilot removes.

When a suggested comment is good, you'll know in 5 seconds. When it's off, you'll know in 5 seconds — and you can edit or skip. Either way, the process is faster than writing from scratch.

The human review principle also has a practical upside: because you're reading before posting, you catch any suggestion that doesn't quite land. You never post something that makes you cringe. You stay in control.

Frequently Asked Questions

Will people know I used AI?

If the tool has been properly trained on your voice — and you're reviewing suggestions before posting — almost certainly not.

The giveaway with generic AI comments isn't AI itself; it's the tell-tale phrases, the hollow enthusiasm, the lack of specificity. When a comment is grounded in your real voice, your real opinions, and actual engagement with the post content, it reads like you. Because it is you, with drafting assistance.

That said: the more niche your voice, the more distinctive your style, the better voice-trained output will be. Generic voices are harder to train because there's less distinctiveness to learn from.

Is this cheating?

This is a values question, and it's worth taking seriously.

The parallel that helps most people: do you think it's cheating to run spellcheck? To use a thesaurus? To ask a colleague "does this sound right?" before sending an important message?

AI comment drafting is on that spectrum. It's a tool that helps you communicate more fluently. The opinion being expressed is yours. The relationship being built is yours. The voice — when the tool is trained correctly — is yours.

What would be "cheating" is outsourcing the thinking: posting AI-generated opinions you don't hold, building a persona that isn't yours, using AI to fake engagement at scale with no real presence behind it. That's bad practice. It's also detectable — eventually.

Using AI to reduce the friction of expressing what you actually think? That's just a better workflow.

Can AI really learn my specific voice?

Yes — to a meaningful degree, and it improves with use.

Voice has multiple layers: vocabulary, sentence structure, tone, cadence, opinion style, humor. AI tools that have been trained on your specific examples can replicate most of these with reasonable accuracy. The more examples, the more accurate.

What AI struggles with: very context-specific references ("inside jokes" with your audience), hyperspecific technical language from niche fields, deeply personal experiences. These you'll need to add in editing. But the structural voice — the feel of how you write — is learnable.

The feedback loop is crucial here. Every time you indicate a suggestion was good or bad, the model recalibrates. Think of voice training as something that gets noticeably better over the first few weeks of use, then levels off at a high baseline.

What if I'm still finding my own voice on LinkedIn?

That's actually a great use case for this kind of tool. The act of curating 12 examples of your best comments forces you to notice what you do well. The pillar-setting process forces you to articulate your professional focus. And getting suggestions back from the model acts like a mirror — when something feels off, it tells you something about what you don't want to sound like, which is also useful signal.

Voice training an AI and developing your own LinkedIn voice are complementary processes.

Does it work for X (Twitter) too?

Yes — FliesReplies Co-Pilot works on both LinkedIn and X. The voice model is trained on your writing across both platforms (or you can keep them separate if your style differs significantly between networks). The in-feed suggestion mechanism works the same way on X.

The Bottom Line on AI LinkedIn Comments

The sequence that works:

Accept that generic AI comments will always feel wrong — that's a feature, not a bug. Your discomfort was your quality filter.
Understand that the fix is voice training, not better prompts (though better prompts help as an intermediate step).
If you're starting with ChatGPT: build a proper voice prompt with real examples, be specific about context and your reactions, edit before posting.
If you want this to be a sustainable, frictionless part of your LinkedIn workflow: use a dedicated Co-Pilot tool that lives in your feed, remembers your voice, and gets better with feedback.

The goal was never to have AI comment for you. It was to have AI help you say what you wanted to say, in the way you would have said it, in a fraction of the time.

That's completely achievable. You've just been using the wrong tool for the job.

Train Your Co-Pilot in 10 Minutes

If you want to try the voice-trained approach: FliesReplies has a free trial — 15 replies, 3 days, no credit card.

The setup takes about 10 minutes: import your comments, pick your 12 examples, set your pillars. Then the next time you're scrolling LinkedIn and a post deserves a reply, you'll have a draft waiting.

Train your Co-Pilot in 10 minutes. Then let it handle the blank page problem.

Start your free trial of FliesReplies Co-Pilot →