Outreach9 min read
How to write a LinkedIn message with Claude that actually gets a reply
Most LinkedIn messages fail because they read generic. Here's the exact loop we run with Claude to make every message sound like it was written by someone who read the profile — because it was.
By Roki HasanApril 20, 2026
Why templates fail
You can feel a template before you finish reading the first line. So can your prospect. The moment someone reads "I hope this message finds you well" followed by something about their CEO title, they stop. The problem isn't that templates are bad writing — it's that they're obviously not written for the reader.
Claude solves a very specific part of this: it can read a profile, a few recent posts, and the company's latest news, and write a single paragraph that references something real. That's the whole trick. The rest is your offer.
The two-step loop
We run every outreach message through two Claude calls. The first gathers context. The second writes the message.
Step 1 — Gather
We paste (or pull via scraper) three things:
- The full LinkedIn profile — headline, about, experience.
- The latest three to five posts the prospect has written or reposted.
- A one-paragraph summary of the company, including any recent funding, launches, or news.
Then we ask Claude:
Read the profile, posts, and company summary below. Identify: (a) the single most specific thing this person cares about right now, based on their own words, not their job title. (b) one thing they're probably frustrated with. (c) one detail that would signal you actually read, not just name-dropped.
This gives you a short memo — three bullet points, maximum.
Step 2 — Write
Then:
Using the memo, write a LinkedIn connection message under 280 characters. Rules: open with the specific thing from (c). Never use "I hope this finds you well." Never mention their title. End with one small, low-commitment ask — not a pitch.
You will read the result and still edit one word, usually. But the structure will be right 80% of the time.
What makes this better than an AI outreach tool
Most "AI outreach" tools run a single prompt with a template baked in. The output reads like it's from a tool because it is. The loop above separates understanding from writing, which is how a person would do it.
It's also why we don't automate the send. A human reads every outgoing message. Claude drafts; we edit; we send. That's the deal.
The follow-up problem
Most outreach dies after the first message. Claude also handles follow-ups, but the prompt is different:
Here is the previous message sent, and the reply (or silence). Draft the next message. If they replied with a soft yes, propose a 20-minute call. If they replied with a question, answer it and ask one of your own. If they didn't reply, do NOT send a "just following up" message — send a second-first-message that opens differently.
That last rule alone changes the reply rate. "Just following up" is the worst sentence in outbound.
What to actually measure
Don't measure volume. Measure specificity — as in, how many of your last 50 messages reference something the prospect actually wrote? That number, more than anything else, predicts your reply rate three weeks later.
This is the same loop we run for our clients. If you'd rather we run it, come talk.
Keep reading
More from the Outreach series.
Want us to run this for you?
Thirty minutes, no slides. We'll tell you honestly whether this is the right first thing to fix.