People can’t always prove a message was written by AI. They can always tell it wasn’t written for them.
Open your LinkedIn inbox and you’ll find the same message from five different people. “I came across your profile and was impressed by your journey.” A compliment, a question about your “current priorities,” a request for fifteen minutes. Each one technically personalized. None of them written for you.
AI made this kind of message cheap to produce, so now there’s a lot more of it, and buyers have learned to spot it in a sentence. The answer isn’t to stop using AI. It’s to change what you ask it to do.
The short answer
Use AI to find the reason to reach out and to draft a first version in your own voice, then edit it like a human before it goes out. Personalize the reason for the message, not the greeting. Start from a real signal, such as a post, a job change or a hiring push, keep it short, make one specific point, and ask for nothing big in the first message.
Why AI outreach sounds like AI
It’s rarely one word. It’s a pattern. Here are the tells people notice most:
The empty compliment. “Impressed by your journey” or “love what you’re building” with nothing specific attached.
The template opener. “I came across your profile,” “I hope this message finds you well,” “I wanted to reach out.”
The fake-specific detail. Mentioning someone’s job title or company as if it were an insight. They know where they work.
Perfect structure. Compliment, problem, solution, ask. Three tidy bullet points. Every sentence the same length.
Corporate vocabulary. “Leverage,” “streamline,” “unlock,” “in today’s fast-paced landscape.”
The big ask too early. A meeting request from a stranger in the first message.
Length. A real person sending a connection request writes two sentences, not two paragraphs.
Notice that most of these aren’t really about AI. They’re about the message having no reason to exist other than the sender’s quota. AI just made it faster to produce.
The fix: personalize the reason, not the greeting
There’s a big difference between a message that mentions you and a message that’s about you. Merge fields and profile facts mention you. Context, like a post they wrote last week or the team they’re hiring for, gives the message a reason. A point of view, something useful that connects their context to your experience, is what makes them want to reply.
So the job of AI in outreach isn’t to write clever sentences. It’s to find the context that makes a message worth sending, and to save you the blank-page time of writing a first draft.
A workflow that keeps your voice
1. Start from a signal, not a list
Before anything gets written, there should be a one-line answer to “why this person, why now?” A post about a problem you understand. A new role. A team they’re hiring for. If there’s no answer, don’t send anything. See 15 B2B buying signals you can actually act on for ideas.
2. Show the AI how you actually write
Generic models write in a generic voice. Give it five or ten messages you’ve sent that got replies, and it will pick up your sentence length, your openings, your habits. This is the idea behind Manet: it drafts based on how you start conversations, and it keeps learning from the edits you make.
3. Give it hard constraints
Connection request notes under 200 characters, ideally two sentences.
One point per message.
No compliments without a specific reference.
No meeting request in the first message.
A list of phrases you never use.
4. Edit like a human
Three edits fix most drafts. Cut the first sentence, because it’s almost always a warm-up. Add one detail only you would know, like a similar mistake you made or a number from your own work. And make it a little uneven: real people write short sentences, start with “And,” and don’t wrap everything up neatly.
5. Review before it goes out
Anything that goes out under your name should get a glance from you. Not every like or profile view, but every comment, connection note and message. It takes seconds, and it’s the difference between automation that helps and automation that slowly erodes your reputation. We make the full case in AI outreach with human approval.
Before and after
Three common moments, with a typical AI draft and a version that sounds like a person.
A connection request after someone posts
Before: “Hi Sarah, I came across your profile and was impressed by your journey in sales leadership. I’d love to connect and explore potential synergies.”
After: “Your point about SDRs burning out on volume targets is the thing nobody says out loud. We cut daily sends by half last year and replies went up. Would be good to be connected.”
A comment on a post
Before: “Great insights, Sarah! This really resonates. Thanks for sharing!”
After: “The 90-day ramp number surprised me, ours was closer to five months for enterprise. Did the shorter ramp hold for reps without prior SaaS experience?”
A first message after they accept
Before: “Thanks for connecting! I’d love to schedule a quick 15-minute call to show you how we help sales leaders like you streamline prospecting.”
After: “Thanks for connecting. Genuinely curious, when you cut volume, what did you replace it with? We’ve been testing comments before connection requests and I’m not sure yet if it scales.”
None of the “after” versions are clever. They’re specific, short and about the other person. That’s all it takes to stand out right now.
A prompt you can use
If you’re drafting with a general AI assistant, a prompt like this gets you much closer to your own voice:
“Here are eight LinkedIn messages I’ve written that got replies: [paste]. Write a connection note to [name], who posted [summary of post] on [date]. Match my style. Two sentences, under 200 characters. Reference one specific point from the post and add one thing from my own experience. No compliments, no meeting request, no words like leverage, synergy or streamline.”
Then edit it. Every time.
The “does this sound like AI?” checklist
Could this message be sent to anyone else with the name swapped? If yes, rewrite.
Does the first sentence say something only this person would recognize?
Is there one point, not three?
Would you say this out loud to them at an event?
Is the ask small enough to answer in one line?
For where this fits in the bigger process, see the LinkedIn prospecting playbook.
Frequently asked questions
Can people tell when a LinkedIn message is written by AI?
Often, yes. Not because of a single word, but because of patterns: empty compliments, template openers, perfect structure and no real reason for the message. Specific context and your own voice fix most of it.
Is it okay to use AI for LinkedIn outreach?
Yes, as long as it helps you send fewer, better messages rather than more generic ones. Use it to find context and draft, and review anything that goes out under your name.
How long should a LinkedIn connection request note be?
LinkedIn's help pages put the limit at 200 characters, and free accounts can only add a note to a few invitations each month. Two sentences is usually ideal: one specific reference and one reason to connect.
What's the best way to personalize at scale?
Personalize the reason, not the greeting. Start each message from a real signal, such as a post, job change or hiring push, and let software do the research so you can spend your time on the sentence that matters.
How does Manet keep messages sounding human?
Manet drafts based on how you start conversations, uses context like recent posts, roles and signals, lets you review and edit important actions, and learns from your edits over time.
Should I pitch in my first LinkedIn message?
Usually not. A first message works better when it continues a conversation or asks a question the person would enjoy answering. Let the pitch come once there's interest.
If you’d like help with the research and the first draft, without giving up the final say, try Manet free.
[
ready to start?
]
