AI writes the hook, a human signs the message

July 28, 2026

The debate about AI in outreach is usually framed as all or nothing. Either you automate and accept that it sounds like a robot, or you write everything by hand and accept that you will send fifteen messages a week.

There is a third option, and it is the one that actually works: let the AI write the two sentences that must change for every person, and keep everything else fixed and human.

The structure we use

A first-contact message has five parts. Only two of them are generated:

  • The greeting. Fixed. One line, no exclamation mark.
  • The curiosity hook. Generated. One verifiable fact about them, naming where we saw it, then a question of eight to fourteen words that creates a bit of tension.
  • What you do. Fixed. One sentence. It never changes because it is not about them.
  • The closing teaser. Generated. One specific idea tied to what we noticed, announced without being revealed.
  • The ask. Fixed. Fifteen minutes, nothing more.

Two generated blocks out of five. That ratio is deliberate. It is enough for the message to be unmistakably about that person, and little enough that a human can read and approve it in fifteen seconds.

What makes a hook work

The fact has to be checkable and you have to say where it came from. “I saw on your site that you work with manufacturing SMEs” proves you looked. “I was impressed by your journey” proves nothing at all.

The question is the harder half. Most AI-written questions are disguised audits: “How do you handle lead generation?” Nobody answers that, because answering it means doing homework for a stranger. What works is a question that reframes something:

  • A paradox: do your clients buy the product, or the certainty of never being stuck?
  • An unexpected comparison: between word of mouth and your LinkedIn profile, which brings in more clients?
  • A blind spot: everyone visits your site, nobody tells you what it sets off.

What AI must never be left alone with

Three things, in our experience. Tone, because a model drifts toward enthusiasm and enthusiasm reads as sales. Facts, because a model will happily infer a funding round that never happened. And the ask, because a model will try to close, and closing on first contact is exactly what makes people stop reading.

The test to run on your own messages

Take a message you are about to send and remove the name. If it could go to anyone else on your list without editing, it is not personalized. It is a template with a merge field, and the person on the other end can tell.

Let’s talk about what LinkedIn could actually do for you.

We have a feeling that this is the beginning of something wonderful. Send us an email with your project needs, and we’ll get back to you within 1-2 business days (probably sooner).
© Copyright, Horizon B2B 2025