The lead-qualifying agent: turn your dream customer into a scoring spec Claude writes

What you're actually building

Most "lead gen agent" content skips the hard part and just shows a tool spitting out names. The build here is smarter and it is the bit nobody shows you: you use Claude as the brain that decides who counts, and a scraping or enrichment tool as the hands that go fetch them.

The flow has four moves:

1. 🗣️ You describe your dream customer to Claude in plain words.
2. 🧾 Claude turns that into a scoring spec: the exact signals that mean "hot lead", what to ignore, and which fields to pull.
3. 🛠️ You hand the spec to a tool (Apollo, Clay, or an Apify actor) that actually pulls the contacts.
4. ✍️ Claude writes the filters and a tiered outreach angle so you know exactly who to message and what to say.

The point: the tool executes, Claude judges. That judgment is the reusable asset, you can run it again next month with fresh data.

Step 1: turn your dream customer into a scoring spec

Open Claude and paste this. Fill the brackets with your real business, the more specific you are, the sharper the spec.

You are my lead-qualification strategist. I sell [what you sell] to [who you sell to]. My best-ever customers looked like this: [describe 2-3 real past customers, their role, company size, what made them a great fit, why they bought].

Build me a lead-scoring spec I can hand to a scraping or enrichment tool. Include:
1. A tiered scoring model (Hot / Warm / Skip) with the exact signals for each tier (job title, company size, industry, funding stage, tech they use, hiring signals, recent triggers).
2. The specific data fields I need to pull for each lead to score them (e.g. title, headcount, LinkedIn URL, company domain, last funding, job postings).
3. Hard disqualifiers, the signals that mean "do not bother", so I stop wasting credits on bad fits.
4. For each tier, a one-line reason it qualifies, so I can sanity-check the logic.

Ask me up to 5 clarifying questions first if it would make the spec sharper.

Step 2: pick the tool that does the pulling

Claude writes the spec, but it can't scrape the live web for verified contacts on its own. You pass the spec (the fields and filters) into one of these. Here is what each one is, when to use it, and the honest catch. Free tiers verified July 2026.

Apollo (apollo.io)

• What it is: a B2B contact database with verified emails and filters for title, company size, industry, and buying signals.
• Use it when: you want ready-made, verified B2B contacts fast and your dream customer is a defined job title at a defined company type.

Clay (clay.com)

• What it is: an enrichment and scoring workbench that pulls from 150+ data providers into one table and can run your scoring logic on each row (including its Claygent AI research agent).
• Use it when: you want to enrich AND score in one place, or blend multiple data sources instead of relying on one database.

Apify (apify.com)

• What it is: a marketplace of pre-built scrapers ("actors") that can pull leads from niche directories, maps, or sites where the big databases have thin coverage.
• Use it when: your dream customer lives somewhere Apollo and Clay don't cover well, like a local directory, a niche marketplace, or a specific site.

Step 3: have Claude write the filters and the outreach

Once you know your tool, paste the spec back into Claude with this:

Here is my lead-scoring spec: [paste the spec from step 1].

I'm pulling contacts in [Apollo / Clay / an Apify actor]. Do two things:
1. Translate my Hot and Warm tiers into the exact filter settings I should enter in that tool (job titles, headcount ranges, industries, keywords, any exclusions).
2. For each tier, write a 2-line opening outreach message that references the specific signal that made them qualify, so it doesn't read like a mass blast.

Keep the outreach human and specific, no "I hope this email finds you well".

Now the whole loop is closed: Claude decided who counts, told you exactly how to find them, and wrote the first line of the conversation.

3 bonus prompts to run next

1. Score a list you already have.

Here is a CSV of leads I already pulled: [paste 10-20 rows]. Score each against my spec as Hot, Warm, or Skip, and tell me which single field is doing the most work to separate good from bad, so I can tighten my filters next time.

2. Find the trigger signals I'm missing.

Based on my dream customer, list the "buying trigger" events I should watch for (new funding, a specific hire, a tool switch, a job posting) and where I could realistically detect each one for free or cheap. Rank them by how strongly each predicts a hot lead.

3. Build a weekly refresh routine.

Design a simple weekly process where I re-run this pull, dedupe against the leads I already contacted, and only surface net-new Hot leads. Give me the exact checklist so it takes me under 15 minutes a week.

How to get the most out of it

When this isn't the move