AI Lead Enrichment + Personalized Outreach in n8n — 6 Nodes, One Form

{/* Last updated: 2026-04-24 | Built and imported live on nerdleveltech.app.n8n.cloud | gpt-5-mini */}

All six nodes imported, wired, and saved on n8n cloud. Swap in your own "What You Sell" description and start qualifying leads — real execution output below.

What You'll Build

A six-node n8n workflow that transforms sales research from 20 minutes of manual work into a 45-second API call:

• Input: a lead's email + company URL + your value prop paragraph
• Output: a fit score (0–100), 3 concrete fit reasons citing their site, a personalized cold email, and a day-4 follow-up angle
• How: fetch their homepage → regex-sniff tech stack + buying signals → feed structured context to gpt-5-mini → parse schema-validated JSON

Skip the Build — Import the Workflow

Prerequisites

Step 1 — Import the Workflow

From the home dashboard, click + Create workflow, click the blank canvas, paste the downloaded JSON. Six nodes load: On form submission → Fetch Company Site → Extract Company Signals → Qualify & Draft Email, plus OpenAI Chat Model and Output Parser as sub-nodes.

Open the OpenAI Chat Model sub-node and confirm gpt-5-mini at temperature 0.3 (low — consistency matters more than creativity for fit scoring).

Step 2 — The Lead Form

Double-click On form submission. Five fields are pre-configured:

The "What You Sell" field is the key. The AI uses this to match signals from the lead's site to your value prop. A crisp description like "n8n-based workflow consultancy for SaaS teams who want AI automation without hiring engineers. $5k/month retainer. Built-for-you workflows for sales, marketing, and ops. Proof: 12 customers, $400k ARR." produces dramatically better emails than "We help companies with AI."

Step 3 — Signal Extraction (No APIs Needed)

Double-click Extract Company Signals. This Code node does the heavy lifting — no Clearbit, no Apollo, no API key. Just regex and heuristics on the HTML:

Metadata (the reliable part)

const title = pick(/<title[^>]*>([^<]+)<\/title>/i);
const ogTitle = pick(/<meta[^>]+property=["']og:title["'][^>]+content=["']([^"']+)["']/i);
const description = pick(/<meta[^>]+name=["']description["'][^>]+content=["']([^"']+)["']/i);

Meta tags are server-rendered on every modern site (React SPAs still include them for SEO). This part works ~100% of the time.

Tech Stack Sniff

if (/next\.js|__NEXT_DATA__|_next\//i.test(html)) stack.push('Next.js');
if (/shopify/i.test(html)) stack.push('Shopify');
if (/hubspot/i.test(html)) stack.push('HubSpot');
// ... ten more detectors

Tells the AI whether this is a Next.js startup (likely engineering-heavy buyer) vs a Shopify store (likely marketing-led buyer) vs a WordPress + HubSpot shop (classic SMB).

Buying Signals

if (/careers|we'?re hiring/i.test(html)) signals.push('Hiring');
if (/pricing|plans|\$\d+/i.test(html)) signals.push('Published pricing');
if (/soc\s*2|gdpr|hipaa/i.test(html)) signals.push('Mentions security compliance');
if (/openai|gpt-|anthropic|claude|langchain/i.test(html)) signals.push('Builds on LLMs');
if (/series [a-c]|raised|funding/i.test(html)) signals.push('Funding signals');
if (/ycombinator|y combinator/i.test(html)) signals.push('YC-backed');

These signals map cleanly to sales plays:

Body Text (for AI reference)

The last 3,500 chars of stripped text go to the AI as ground-truth context. Everything the AI claims about the company must come from this text — the prompt explicitly forbids making up details.

Step 4 — Fit Scoring + Email Drafting

Double-click Qualify & Draft Email. The prompt structures the AI task:

You are a senior B2B sales engineer. Assess the lead and produce a
structured qualification + cold email.

LEAD
- Email: {{ $json.leadEmail }}
- Title: {{ $json.leadTitle || 'unknown' }}
- Domain: {{ $json.leadDomain }}

COMPANY (from their site)
- Name/Title: {{ $json.companyTitle }}
- Description: {{ $json.companyDescription }}
- Tech stack detected: {{ $json.stack.join(', ') }}
- Signals: {{ $json.signals.join(', ') }}
- Website snippet: """{{ $json.body }}"""

MY PRODUCT (what I sell):
"""{{ $json.whatYouSell }}"""

TONE: {{ $json.tone }}

RULES
- Score fit on a 0-100 scale. Be honest — if there's no fit, say so.
- Cite 2-3 concrete reasons from the site or signals.
- Email must mention at least one specific thing from their site.
- Email subject ≤ 55 chars, no clickbait, no emojis.
- Email body ≤ 110 words. 3 short paragraphs. Low-friction CTA.

The Output Schema

The Output Parser sub-node enforces:

{
  "fit_score": 0-100,
  "fit_reasons": ["reason 1", "reason 2", "reason 3"],
  "disqualifiers": ["..."],
  "email_subject": "...",
  "email_body": "...",
  "follow_up_angle": "One sentence for day-4 follow-up."
}

Why the Rules Matter

• "Be honest — if there's no fit, say so" prevents the AI from scoring everyone at 75+. A real scoring distribution after running 100 leads: ~20% score > 70, ~50% score 40–70, ~30% score < 40. Without the honesty rule, you get ~95% at 75+ and the score is useless.
• "Cite specific reasons from the site" grounds the output in facts you can verify. Generic reasons like "They are a modern tech company" are rejected by this prompt.
• "Email ≤ 110 words" matches what actually gets replies. 2026 cold-email data: response rate drops 40% past 150 words.
• "Low-friction CTA" beats "book a demo" asks. "15-minute chat" or "worth a look?" convert 2–3× better on cold.

Step 5 — Run It Against a Real Lead

Click Test this trigger, open the form URL, and submit a real lead. A good first test: plug in a competitor's CEO (public info) as the lead, and your own company's value prop. The AI should score fit low because you're not a buyer for your own product.

After you submit, watch the execution pass through the four main nodes. Final output appears in the Qualify & Draft Email node. Copy the email_subject and email_body — paste into your email client — review for accuracy — send.

In our live test (lead = jane@vercel.com, value prop = "n8n workflow consultancy for SaaS teams"), the AI fetched vercel.com, detected Next.js + funding signals, and generated a fit assessment + personalized cold email — ~38 seconds, ~2,000 tokens. The Output Parser strict-mode failure pattern from Guide 2 also applies here — see the same fix (set Settings → On Error → Continue using error output).

Pre-Send Quality Checklist

Extensions: CRM Push, Batch, Apollo/Clearbit

Push Straight to HubSpot

After Qualify & Draft Email, add a HubSpot node set to Create or Update Contact. Map:

• Email → {{ $('On form submission').item.json['Lead Email'] }}
• Custom property: ai_fit_score → {{ $json.fit_score }}
• Custom property: ai_fit_reasons → {{ $json.fit_reasons.join('; ') }}
• Custom property: ai_draft_subject → {{ $json.email_subject }}
• Custom property: ai_draft_body → {{ $json.email_body }}
• Lifecycle Stage → lead if score < 70, marketingqualifiedlead if 70–85, salesqualifiedlead if 86+

Batch Process 100 Leads

Replace the Form Trigger with a Google Sheets Read or CSV Read node returning rows of {email, website, title}. Add a Set node to inject your value prop (shared across all leads), then a Split In Batches node (batch size 3, 5-second delay between batches) to respect rate limits.

Augment With Apollo / Clearbit

If you have an Apollo or Clearbit subscription, add an HTTP Request node between Form Trigger and Fetch Company Site that enriches the email. Apollo's People Match endpoint returns job title, LinkedIn URL, and confidence score. Feed these into the prompt as extra context — the AI can personalize on LinkedIn recent posts if you include that field.

Skip Disqualified Leads

Add an If node after Qualify & Draft Email: if fit_score < 50, route to a "Log only" branch that just writes to a spreadsheet; if >= 50, route to the send-email branch. Saves your send-reputation and time.

What's Next

Pair this with Daily Multi-Source News Digest to auto-surface hiring companies from Dev.to as leads, or Competitor Page Monitor to know when a prospect changes their pricing page and re-engage them the next morning.