AI automations for lead generation

Last quarter I needed a steady flow of fresh leads for my consulting offer without spending hours on manual LinkedIn searches. I built a small AI‑driven pipeline that pulls profiles, enriches them with contact info, and writes a personalized intro line — all while I slept. After reading this you’ll be able to assemble the same flow, spot where it usually breaks, and decide whether to build it yourself or grab a ready‑made blueprint.

How do you keep lead data fresh without manual exports?

Most operators start by exporting a CSV from LinkedIn Sales Navigator, then they try to enrich each row with an email finder. That works for a few dozen leads but collapses when you need hundreds per week because the export step is a manual bottleneck. Instead, let the scraper run on a schedule and push results straight into a Google Sheet or Airtable via a webhook. I use PhantomBuster to scrape LinkedIn profiles every morning at 6 AM, then I send the JSON payload to a the Make platform webhook that appends each record to a Sheet. The sheet becomes the living source for the next step.

One thing that tripped me up early was PhantomBuster’s rate limit change. I woke up to a stalled scraper and a sheet with only half the expected rows. The vendor had lowered the daily limit without updating their dashboard notice. I now keep a simple health‑check Make scenario that pings the PhantomBuster API and sends me a Slack alert if the response time exceeds two seconds. That tiny guardrail saved me from losing a whole day of outreach.

What most guides get wrong about AI prompts for lead scraping

Many tutorials tell you to feed the raw LinkedIn HTML into GPT‑4 and ask it to “extract the name, title, and company.” That approach is fragile because the HTML varies with LinkedIn’s A/B tests and the model often returns incomplete fields. A better pattern is to let the scraper output structured JSON (PhantomBuster already gives you name, headline, profile URL) and then use a short prompt to enrich only the missing pieces.

Here is the prompt I use for enrichment:


Given the following LinkedIn profile data:
Name: {{name}}
Headline: {{headline}}
Profile URL: {{url}}

If you cannot find a value, set the field to null.


I run this prompt through OpenAI’s gpt‑4o‑mini model via the Make.com HTTP module. The temperature is set to 0.2 to keep the output deterministic. The result is a clean JSON that I can directly map to columns in my Sheet.

Why does this work better? Because the model only has to reason about a small, predictable snippet instead of wrestling with a full page of markup. It also reduces token usage, which keeps the cost low.

A concrete named example: PhantomBuster + OpenAI enrichment

Let’s walk through the numbers I see on my own setup.

• PhantomBuster LinkedIn scraper: $29/mo for the “Starter” plan (500 ph per day).
• Make.com scenario: free tier handles up to 1 000 operations/mo, which covers my webhook and two HTTP calls.
• OpenAI API: gpt‑4o‑mini at $0.0005 per 1 k tokens. Each enrichment call uses about 300 tokens, so roughly $0.00015 per lead. At 2 000 leads/mo that’s $0.30.
• Google Sheets storage: free.

Total monthly cost: just under $30. I consider $29/mo fair for a fully automated lead feed that delivers 2 000 enriched prospects with email guesses and cleaned titles.

One concrete love: the OpenAI function call returns a JSON that already matches my Sheet column names, so I never have to write a mapping script. I simply enable the “Make JSON output” toggle and the data lands where I expect it.

One concrete gripe: PhantomBuster’s documentation assumes you know JavaScript to tweak the scraper’s output format. When I tried to add a custom field for “LinkedIn follower count,” I found the example code was outdated and threw a silent error. I spent two hours digging through community forums before realizing the platform had moved to a new SDK version. A version‑specific changelog would have saved me that frustration.

How to debug when the enrichment pipeline breaks

When the Sheet stops getting new rows, I follow a three‑step checklist.

1. Check the PhantomBuster agent log. If the agent shows “Idle” or “Error: Rate limit exceeded,” the problem is upstream.
2. If the agent looks fine, open the Make.com scenario history. Look for the webhook module — if it shows “No data received,” the webhook URL might have changed or the payload format shifted.
3. If the webhook fires but the OpenAI module returns an error, inspect the raw response. A common issue is the model returning stray text before or after the JSON; I added a simple regex extraction step to strip anything outside the first curly brace pair.

I keep this checklist as a note in my Make.com dashboard so I can run through it in under five minutes. Most of the time the fix is just restarting the PhantomBuster agent after a rate‑limit reset.

Pricing opinion and when to grab the blueprint

If you enjoy tinkering with APIs and don’t mind spending a Sunday afternoon setting up webhooks, the build‑it‑yourself path costs about $30/mo plus your time. I’ve seen similar setups sold as “AI lead gen courses” for $199/mo, which feels ridiculous for what you get — mostly a few recorded videos and a Discord invite.

On the other hand, if you’d rather skip the build and deploy a working version in an afternoon, we’ve packaged this workflow as a blueprint at deepusecase.com/vault/ai-automation-blueprint. The blueprint includes the PhantomBuster agent, the Make.com scenario JSON, the OpenAI prompt template, and a pre‑configured Google Sheet — all ready to import.

Adjacent reading: deeper coverage of AI agent platforms.

Whether you choose to DIY or grab the blueprint, the core idea stays the same: let the scraper handle the heavy lifting, use a tight AI prompt for enrichment, and automate the handoff to your outreach tool. That’s how you turn a manual lead hunt into a quiet, reliable flow that works while you focus on closing deals.