ChatGPT and Claude prompts for LinkedIn prospecting

If you ask an AI to write for a general audience, it writes for nobody. The output comes back full of phrases like solutions that drive results, because the model has nothing specific to hold on to. Give it a narrow target: the industry, the job title, the company size and the problem that person is dealing with this quarter.

The prompt below works because it names a pain the buyer would recognize in their own words. Union compliance and manual timesheets are things a construction CFO complains about at lunch. Efficiency is not.

Run this once and save the output in a note. You will paste the same ICP description into every other prompt in this guide, and rewriting it each time is how the details drift.

AI will not hand you names with LinkedIn URLs attached, and expecting that wastes an afternoon. What it can do is turn your ICP into search phrases and a scoring checklist you apply by hand in LinkedIn search.

Ask for the words a buyer would put in their own headline, then run those phrases through the people search filters you already have. The prompt is short because the value is in the specificity you already supplied.

Do not ask the AI to write a message straight after you paste a profile. It will summarize the profile back at you in slightly different words, and the message will sound like a summary. Use it as a research step first.

Copy the About section and the last three posts. Posts matter more than About, because posts carry opinions and opinions are what you can reply to.

A connection request has one job: get accepted. Pitch in the note and most people ignore it, because the note arrives before any trust exists.

Constraints do most of the work here. Give the model a character limit, a tone and a list of words it may not use. Without them you get three sentences of throat clearing before the point.

Once they accept, the connection note has done its job. Now you move from the thing you both noticed to the thing you sell, in one step.

Keep the first message a question about their current setup. Leave out links and calendar suggestions. The exchange below is what a working thread looks like when the lead is a construction CFO.

When someone says tell me more, the model tends to dump the whole product into one message. That kills the thread. The reply should name the outcome in a sentence and offer a time.

The prompt for this step: Here is the reply I got: [paste]. Write a two-sentence response that names the outcome in plain words and offers two specific times next week. No feature lists, no apologies, no I'd be happy to.

Kinlead handles the search and scoring without a prompt. You describe the customer or paste your website, and it scores each lead from 0 to 100 against that profile. Leads below the fit line are not kept, so the list you work from is already filtered.

The Steal Customers agent watches competitor company pages and optional founder or employee profiles, reads the comments on their recent posts, and turns commenters who look like buyers into leads. Each lead keeps the post URL, what the post was about and what they commented, which is the research step above done for you.

Connection notes and follow-ups are written from the campaign's goal, tone and your own sales playbook. Queued messages can be read and edited on the Actions page before they send. There is a walkthrough at book a demo.