Short answer: an AI writes a good LinkedIn message if you give it one verified fact about the person and strict writing rules, then proofread the result. Left alone, it produces sentences that sound fake: "your background caught my eye", "I'd love to connect". The gain is not in the prose, it is in the time saved reading profiles.
Why does an AI-written message stand out?
An AI writes like the average of what it has read, and what it has read is thousands of failed prospecting messages. Hence the same tics everywhere: a compliment on the background, a connector at the start of a sentence, an offer to "hop on a quick call", an exclamation mark. Someone who receives ten messages a week recognizes the shape before reading the content.
The second sign is length. An AI does not know when to stop: three sentences requested, five delivered, plus a closing formula. A first-contact message runs between fifteen and thirty-five words. Beyond that, it looks like a letter.
The third sign is the missing fact. "I saw you work in logistics" is not a fact, it is the reading of a job title. A fact is "your Tuesday post on year-end delivery delays". An AI given nothing makes things up, and a made-up fact that gets spotted costs you the conversation.
What should you give the AI so it writes right?
Three things, and not one more. One specific, verified fact about the person: a post, a comment, a job change, a hiring in progress. An honest description of what you do, in two sentences, for whom and for which problem. And writing rules, spelled out in black and white.
The fact matters most, and it is what takes time. Without it, the AI writes the same message for everyone, swapping the first name. With it, it has something to say, and the question it asks follows from it. Our method of one specific fact, then one question applies word for word to an AI.
The description of your business weighs more than you think. An AI that reads "innovative digital transformation solution" writes empty messages, because it has nothing concrete to connect to the person's problem. We detailed what to write in describe your business so your messages sound right.
Which writing rules should you impose?
The same ones you would give a human who is starting out, and they must be written in the prompt, not assumed. Here are the ones we use:
- Two or three sentences, forty words at most, no sentence over twenty-two words.
- One question only, at the end, answerable in one line.
- No flattery: no "impressive", no "inspiring", no "congrats".
- No learned formula: no "I hope this finds you well", no "I wanted to reach out", no "looking forward to connecting".
- No exclamation mark, no emoji, no long dash.
- No "It's not X, it's Y": that turn gives a generated text away.
- Match the person's register: formal if her posts are formal, casual if they are casual.
A prompt is not enough, the AI forgets it one time in five. You have to check the output against those rules, and reject what fails. That is tedious by hand; it is the kind of check a tool does well.
Before and after examples
Same person, same fact, same offer. Only the prompt changes.
Before, with no fact and no rules:
Hi Claire, your background in the logistics sector really caught my eye! I wanted to reach out because I help companies like yours with their transformation. I'd love to hop on a quick call to explore how we might collaborate. Looking forward to connecting!
Forty-eight words, three tics in the first sentence, no precise question. This message has been sent to thousands of Claires.
After, with a fact and the rules:
Hi Claire, I read your post on December delivery delays and the cost of penalties. Do you absorb them, or bill them back to the carriers?
Twenty-six words, a verifiable fact, a question answered in five words. Nothing about the offer: that comes if Claire replies.
Another case, a comment rather than a post:
Hi Mark, your comment under Julie Perrin's post on quotes that drag on for three weeks struck a chord. Is that the case on your side right now?
The limit of these examples: they work because the fact is real and recent. A six-month-old fact is a dead fact. And an AI has no way of knowing whether the post exists: you, or your tool, must have read it.
Which messages should you not hand to the AI?
The reply to an objection, the negotiation, and anything that commits to a price or a deadline. An AI can propose a draft, but the text that goes out must be proofread, because a written promise is a promise. Same for a message to someone you already know: she will feel the change of voice.
The first message and the follow-up, on the other hand, suit an AI well, provided you proofread them. They are short, repetitive in form, and the fact does all the work. That is where the time is saved.
How does Sales Farmer write the opening line?
The opening line is the first sentence of the message, written for one person from what she did: a post liked or commented, a publication, a visible change. It is written on demand by a single model call, then kept on the person's record, never generated at send time. The context comes from what the extension read, not from an invention: if the topic of a post could not be read, it is not made up.
Each opening line goes through a tic detector that scores the text on measurable criteria: forbidden expressions, flattery, length, an overlong sentence, an exclamation mark, the same opening as the previous message in the batch. A text above the threshold does not go out; the tool falls back on a neutral sentence built on the signal. You proofread and edit each message person by person before clicking Launch. The number of opening lines written per month depends on the plan, detailed on the Pricing page.
What the tool does not do: it does not guarantee a reply, and it does not read the person's mind. A well-quoted fact gives a reason to reply, not an obligation.
FAQ
Can people tell a LinkedIn message was written by an AI?
Often, yes, when the message carries the usual tics: flattery, a learned politeness formula, an offer to hop on a call. A short message, with one specific fact and a simple question, is indistinguishable from a handwritten one, because the fact is what counts, not the prose.
Can you use ChatGPT to write your LinkedIn messages?
Yes, with a prompt that contains the fact, your business in two sentences and the writing rules above. The result needs proofreading every time. The real cost is not the writing, it is reading the profile to find the fact.
Should you tell the person the message was generated?
Nothing requires it if you proofread it and stand behind it, like a letter prepared by an assistant. What is a problem is a message that claims to have read a post no human has read.
Can an AI write in several languages?
Yes, and that is one of its real advantages. Check the register: formal and informal address, the German "Sie", the opening formula, vary from one language to the next, and a mistake there shows immediately.