n8n ICP Website Scoring Workflow with GPT-4.1 CSV Export

Score ICP fit for every domain—then export a clean CSV/XLSX with GPT-4.1 guidance

This n8n ICP Website Scoring Workflow with GPT-4.1 CSV Export takes a list of company domains from an n8n form, refines your ideal customer profile (ICP) using OpenAI (GPT-4.1), scans each website for firmographic and intent signals, and produces deterministic ICP-fit scores you can download as CSV or Excel.

What this workflow does

• Collects inputs via an n8n form—upload a CSV/XLSX or paste a domain list, provide your ICP description, choose the output format, and decide whether to include AI-generated reasoning.
• Parses, normalizes, and deduplicates domains; if no valid domains are found, it stops with a completion message.
• Refines your ICP with GPT-4.1 by analyzing your ICP description and (if needed) asking follow-up questions via forms to produce a final structured ICP profile.
• Fetches website evidence for each domain in batches—validates domains and retrieves the homepage and /about pages via HTTP to gather evidence and technology fingerprints.
• Extracts structured signals using OpenAI (gpt-4.1-mini) from the gathered website evidence plus deterministic domain-based clues.
• Deterministically scores fit with a weighted engine, including penalties for exclusions, low confidence, and suspected prompt injection—then builds a gap/fit explanation from the score breakdown.
• Optionally generates match reasoning using gpt-4.1-mini, strictly based on the computed breakdown, and exports results to CSV/XLSX.

Use cases

• Automate ICP scoring for inbound lead lists before routing to sales.
• Validate whether target accounts actually match your SaaS ICP using website evidence.
• Create repeatable “account scoring + export” processes for marketing ops and SDR workflows.

Technical details

• OpenAI models: GPT-4.1 (ICP refinement), gpt-4.1-mini (signal extraction + scoring explanations)
• Workflow nodes/blocks: if, code, form, no op, wait, merge
• HTTP fetching: pulls homepage and /about pages; uses detected technology fingerprints
• Output: downloadable CSV or Excel export