// AI SDR AND AI AGENTS FOR SALES
AI Lead Scoring: Qualify LinkedIn Leads by ICP
Cold outreach underperforms mostly because it reaches the wrong people. AI lead scoring rates every prospect against your ICP before you send anything, so your requests and your personal notes go only to people who can buy.
Why score before you send
- LinkedIn limits how many requests you can send. Every request to a poor fit is one a good fit does not get.
- Personalization only pays off for people who might buy.
- Low reply rates push teams to send more, which pushes rates lower.
Do not pay to score the whole list
- Free rules remove obvious non-fits: wrong seniority, wrong region, clearly off-ICP.
- A cheap AI call scores the rest 1 to 10 with a short reason.
- Only validated leads get a full profile read and a personal note.
What makes a scoring prompt good
- What you sell and who buys it, in plain words.
- Roles and company types that fit, and the signals that mean they need you now.
- Disqualifiers and edge cases: agencies, students, recruiters score low even with a matching title.
- A score plus a one-line reason: the score routes, the reason lets you audit.
Example to start from: "Score 1 to 10. 9 to 10: [your exact buyer role] at [company type] who [the signal that means they need you]. Cap students, recruiters and job seekers at 3."
Keep the rejects and learn from them
Read the reasons on a sample batch. Good fits that were rejected mean the prompt is too strict. Poor fits that passed mean it is too loose. Tighten until the scores match your own judgment.
On your current stack
On a typical stack, scoring means exporting the list into Clay or a spreadsheet, paying credits to enrich and score every row, including the half that free rules would have dropped, then exporting the winners into an outreach tool. Every change to the ICP means re-running the whole table and re-importing the result, so most teams score once and never refine it.
With AtomSell free rules run first, AI scores only what passes, and the scored leads go straight to outreach from the same table. Changing the ICP is one sentence in your next prompt.
The result
Scoring alone took our Sales Navigator outreach to 33% acceptance on 7,152 requests, against 10 to 20% for unscored cold lists. With AtomSell, search, scoring, notes and sending run from one prompt in Claude.
Run it yourself
Start a 3-day free trial with full access at app.atomsell.ai/register, connect AtomSell to Claude, ChatGPT or Cursor, and run the playbook on your own buyers.