ChatGPT prompts for LinkedIn outreach work when they start from the prospect's profile and activity, not from a message template. LinkedIn is a worse channel than email for generic messages — the inbox is smaller, the social cost of a bad message is higher, and buyers have seen every "I came across your profile" opener hundreds of times. The prompts below run in a specific order: turn the profile into one specific observation first, then write the connection request, the first message, and the follow-ups from that observation. Write for one prospect at a time; the moment you batch, the output reads batched.
Why most AI-written LinkedIn messages get ignored
Buyers pattern-match. "Impressive background," "I'd love to connect," "quick question" followed by a paragraph that is not quick — these phrases mark a message as automated regardless of whether a human sent it. The failure is upstream of the writing: the sender never gave the model anything specific, so the model reached for the phrases everyone else's model reaches for. Every prompt below therefore starts with pasted profile material, and several explicitly ban the stock phrases.
The profile-to-personalization prompt
"Here is the prospect's LinkedIn material: [paste their headline, about section, recent posts or comments, and current role]. Identify the one detail most relevant to [your one-sentence value case] and write a single opening observation about it in under 25 words. It must be something that could only be said to this person — if it could be sent to anyone with the same job title, discard it and pick another detail."
Run this before anything else. The output becomes the seed for every message in the sequence, and the could-only-be-said-to-this-person test is the strongest quality filter in this whole workflow. Deeper research prompts — funding, hiring signals, tech stack — live in the Account Research & Buyer Intelligence category.
Connection request prompts: under 300 characters, no pitch
"Using this observation: [paste from previous prompt], write a LinkedIn connection request under 280 characters. It must reference the observation, say why I want to connect in plain words, and contain no pitch, no link, and none of these phrases: 'impressive background', 'came across your profile', 'love to connect', 'pick your brain'."
The request's only job is acceptance. A pitch inside a connection request converts the small number of people who were already looking to buy and burns everyone else. Sending a blank request is a legitimate alternative for well-known titles; a good note simply raises the accept rate for cold ones.
The first message after they accept
"[Name] accepted my connection request, which referenced: [observation]. Write a first message under 80 words that picks up that thread, adds one useful thing — an insight, a relevant example, a pointed question about their situation — and ends without a meeting ask. It should read like the start of a conversation, not a step in a sequence."
Not asking for a meeting in message one feels slow and is consistently faster: replies come easier when the recipient is not being closed, and a real reply is worth more than a polite decline. The meeting ask belongs one exchange later, after they have said something you can anchor it to.
The comment-then-DM play
Warming a prospect before the request measurably raises accept and reply rates. "Here is the prospect's recent post: [paste]. Write a comment under 40 words that adds a specific point from my experience with [topic] — not agreement, not praise, no emoji. It should be worth reading for anyone who sees the post." Comment first, wait a day or two, then send the connection request referencing the same thread. The sequence feels natural because it is: you showed up in their notifications as a person with a point of view before you showed up as a request.
Follow-up prompts inside LinkedIn
LinkedIn follow-ups run on a slower clock than email. One unanswered message needs four to seven days before the next, and the next needs new material: "Write a LinkedIn follow-up under 50 words. New input since my last message: [their new post, company news, or a relevant resource]. Reference the new input, not the silence. No 'just bumping this'." Two follow-ups is the practical ceiling in most sequences — past that, move the thread to email or the phone. For the email side of the cadence, see ChatGPT prompts for cold email outreach; the Outreach & Messaging category holds LinkedIn and social messaging prompts by touch.
What not to automate
Drafting with ChatGPT is fine; sending with automation tools is where sellers get hurt. LinkedIn's user agreement prohibits third-party software that automates activity — mass connection requests, auto-DMs, scraping — and accounts do get restricted for it. The safe line is simple: the model drafts, you read, you personalize the details the model could not know, you send. That last human pass is also where quality comes from. If a message is not worth thirty seconds of your review, it is not worth your prospect's attention either, and no prompt fixes that. A library of drafting prompts you trust makes the human pass fast; free vs paid prompt libraries covers what that costs. Start with a free prompt or see pricing.
Frequently asked questions
How do I write a LinkedIn connection request with ChatGPT?
First run a profile prompt that turns the prospect's headline, about section, and recent activity into one specific observation, then ask for a request under 280 characters that references the observation and contains no pitch and no link. Ban the stock phrases — 'impressive background', 'came across your profile' — inside the prompt itself.
Can ChatGPT personalize LinkedIn messages from someone's profile?
Yes, if you paste the profile material in — the model cannot browse the profile for you in a normal chat. Give it the headline, about section, and a recent post, and apply one test to the output: if the line could be sent to anyone with the same job title, it is not personalized, so discard it and pick a different detail.
Are AI-written LinkedIn messages against LinkedIn's rules?
Using AI to draft a message you personally review and send is not prohibited. What LinkedIn's user agreement does prohibit is third-party automation software — bulk connection requests, auto-messaging, scraping — and accounts get restricted for it. Keep the sending human and the volume realistic, and the drafting tool is not the risk.
How long should a LinkedIn cold message be?
Connection requests are capped at 300 characters; keep them under 280. First messages after acceptance work best under 80 words, and follow-ups under 50. LinkedIn is read on the phone between other tasks — a message that requires scrolling is a message that gets saved for later and never reopened.