Generate Personalized Cold Emails with Anthropic, GPT-4 & Google Sheets

Your Cold Email is Now Researched. This pipeline finds specific bottlenecks on prospect websites and instantly crafts an irresistible pitch

🎯 Problem Statement

Traditional high-volume cold email outreach is stuck on generic personalization (e.g., "Love your website!"). Sales teams, especially those selling high-value AI Receptionists, struggle to efficiently find the one Unique Operational Hook (like manual scheduling dependency or high call volume) needed to make the pitch relevant. This forces reliance on expensive, slow manual research, leading to low reply rates and inefficient spending on bulk outreach tools.

✨ Solution

This workflow deploys a resilient Dual-AI Personalization Pipeline that runs on a batch basis. It uses the Filter (Qualified Leads) node as a cost-saving Quality Gate to prevent processing bad leads. It executes a Targeted Deep Dive on successful leads, using GPT-4 for analytical insight extraction and Claude Sonnet for coherent, human-like copy generation. The entire process outputs campaign-ready data directly to Google Sheets and sends a critical QA Draft via Gmail.

⚙️ How It Works (Multi-Step Execution)

1. Ingestion and Cost Control (The Quality Gate)

• Trigger and Ingestion: The workflow starts via a Manual Trigger, pulling leads directly from Get All Leads (Google Sheets).

• Cost Filtering: The Filter (Qualified Leads) node removes leads that lack a working email or website URL.

• Execution Isolation: The Loop Over Leads node initiates individual processing. The Capture Lead Data (Set) node immediately captures and locks down the original lead context for stability throughout the loop.

• Hybrid Scraping: The Scrape Site (HTTP Request) and Extract Text & Links (HTML) nodes execute the Hybrid Scraping strategy, simultaneously capturing website text and external links.

• Data Shaping & Status: The Filter Social & Status (Code) node is the control center. It filters links, bundles the context, and critically, assigns a status of 'Success' or 'Scrape Fail'.

• Cost Control Branch: The If (IF node) checks this status. Items with 'Scrape Fail' bypass all AI steps (saving 100% of AI token costs) and jump directly to Log Final Result. Successful items proceed to the AI core.

2. Dual-AI Coherence & Dispatch (The Executive Output)

• Analytical Synthesis: The Summarize Website (OpenAI) node uses GPT-4 to synthesize the full context and extract the Unique Operational Hook (e.g., manual booking overhead).

• Coherent Copy Generation: The Generate Subject & Body (Anthropic) node uses the Claude Sonnet model to generate the subject and the multi-line body, guaranteeing coherence by creating both simultaneously in a single JSON output.

• Final Parsing: The Parse AI Output (Code) node reliably strips markdown wrappers and extracts the clean subject and body strings.

• Final Delivery: The data is logged via Log Final Result (Google Sheets), and the completed email is sent to the user via Create a draft (Gmail) for final Quality Assurance before sending.

🛠️ Setup Steps

Before running the workflow, ensure these credentials and data structures are correctly configured:

Credentials

• Anthropic: Configure credentials for the Language Model (Claude Sonnet).

• OpenAI: Configure credentials for the Analytical Model (GPT-4/GPT-4o).

• Google Services: Set up OAuth2 credentials for Google Sheets (Input/Output) and Gmail (Draft QA and Completion Alert).

Configuration

• Google Sheet Setup: Your input sheet must include the columns email, website_url, and an empty Icebreaker column for initial filtering.

• HTTP URL: Verify that the Scrape Site node's URL parameter is set to pull the website URL from the stabilized data structure: ={{ $json.website_url }}.

• AI Prompts: Ensure the Anthropic prompt contains your current Irresistible Sales Offer and the required nested JSON output structure.

✅ Benefits

• Coherence Guarantee: A single Anthropic node generates both the subject and body, guaranteeing the message is perfectly aligned and hits the same unique insight.

• Maximum Cost Control: The IF node prevents spending tokens on bad or broken websites, making the campaign highly budget-efficient.

• Deep Personalization: Combines website text and social media links, creating an icebreaker that implies thorough, manual research.

• High Reliability: Uses robust Code nodes for data structuring and parsing, ensuring the workflow runs consistently under real-world conditions without crashing.

• Zero-Risk QA: The final Gmail (Create a draft) step ensures human review of the generated copy before any cold emails are sent out.