Skip to main content
Normalized for Mintlify from knowledge-base/aiconnected-apps-and-modules/modules/funnelChat/legacy-funnelChat-conversation-flow-sample.mdx.
Got it — you’re referring to your funnelChat philosophy: “The best funnel is the one leads don’t know they’re in.” And the goal is to collect lead data passively and naturally during a rich, helpful, Perplexity-style conversation — not through obvious form-filling behavior. Let’s reframe the solution around this invisible lead capture idea.

​
✅ funnelChat Lead Capture – Passive & Conversational Flow

Instead of direct prompts like “What’s your name?”, the AI should:
  1. Guide the conversation like a helpful research assistant.
  2. Casually prompt for info only when it feels relevant or earned.
  3. Use AI tools to extract structured data from unstructured, free-flowing dialogue.

​
🧠 How to Implement This in n8n

​
1. Define Passive Goals for the Agent

In the AI Agent node: {
"persona": "An ultra-helpful, respectful research assistant who offers solutions without pressure.",
"goal": "Guide the user through an insightful conversation and, where natural, capture their name, email, and business interest for follow-up.",
"tone": "Engaged, light, helpful — never pushy or robotic.",
"rules": [
"Do not directly ask for contact info unless it arises naturally.",
"If the user asks a detailed question, you may offer to send more info by email.",
"Make email or name collection feel like a service, not a requirement."
]
}

​
2. Enable Soft Prompts During Key Moments

Examples of prompts the AI can drop only after giving value:
  • “I actually wrote a short summary of this last week — want me to send it over?”
  • “If you’d like a downloadable version of these steps, I can email it to you. Totally optional.”
  • “Happy to turn this into a checklist — want me to email it over?”
These feel like bonuses, not asks.

​
3. Extract Lead Fields Invisibly

Use OpenAI / Claude / Regex to extract:
FieldExtraction Approach
Name“My name is…” / Email signature / Greeting
EmailDetect using regex or prompt: “Can I send this to your email?”
IntentInferred from multi-turn topic (“I’m trying to solve…” / “My business needs…”)
You don’t need a form — just pass messages through a parsing node (e.g. JS + LLM) every 2–3 turns.

​
4. Save to Memory + CRM Silently

Once fields are inferred:
  • Store them into memory
  • Update user context
  • Save lead quietly in background (e.g. Airtable, GHL, Supabase)
If user returns later, the agent can say: “Welcome back! Last time we were talking about your sales funnel. Want to pick up from there?”

​
5. Show Empathy, Not Automation

When it’s time to ask for info, don’t “collect,” just offer help. ✔️ Instead of: “Please provide your email.” Try: “Would you like a quick PDF of these strategies? I can email it to you.” Even if they say “Sure,” that’s enough to extract and store — without breaking immersion.

​
⚙️ n8n Node Tips

  • LLM Entity Extractor: Custom JS node or OpenAI node to pull name/email from chat stream.
  • Memory Check: If node to detect what’s missing and route the conversation softly.
  • Lead Saver: Background webhook or HTTP Request node that pushes to your CRM.
  • Re-engagement Prompt: Use stored memory to tailor follow-ups if user returns.
Last modified on April 17, 2026