June 30, 2025
AI Icebreaker Builder: Scrape Sites with Dumpling AI and Save to Airtable
Introduction
Sending cold emails without personalization is a waste of time. But writing custom messages for each lead manually is nearly impossible at scale. That’s where this automation comes in.
This workflow scrapes your lead’s website using Dumpling AI and feeds the data to GPT-4o to automatically generate:
- A personalized icebreaker
- A cold email message
- A short summary of the lead’s website
The final output is saved in Airtable, giving you a growing list of outreach-ready leads without lifting a finger. Ideal for SDRs, founders, agencies, and anyone who wants to automate high-quality cold emails.
Workflow Breakdown
⏰ Trigger: Schedule Node
Purpose: Starts the automation on a recurring basis.
Details: This node is set to run the entire automation daily. You can adjust the interval as needed (e.g., every few hours or once a week) depending on your lead volume. A scheduled run ensures the process stays consistent without you needing to click a button.
Step 1: Airtable – Search Records
Purpose: Fetch new leads that haven’t yet been processed.
How it works:
- Searches the “cold leads” table in Airtable.
- Filters for records where the “Ice breaker” field is empty.
Why it matters: This step ensures the automation only processes fresh records. That way, existing leads with completed data aren’t touched.
Step 2: Loop node
Purpose: Loops through each lead one at a time.
Details: This node enables one-by-one processing so that each record gets handled individually. It’s especially useful when dealing with APIs like Dumpling AI or OpenAI that may have rate limits.
Step 3: Wait
Purpose: Pause before scraping.
Details: Adds a delay (e.g., 1 second) before making a request to Dumpling AI. This helps manage API load, especially when processing many leads in a row.
Step 4: HTTP Request to Dumpling AI /scrape
Purpose: Scrapes the lead’s website for text content.
Payload Example:
{
“url”: “{{ $json.Website }}”
}
Details: Sends a POST request to the Dumpling AI /scrape endpoint with the lead’s website URL.
Why it matters: This gives GPT-4o real, live content from the lead’s actual website. Instead of relying on just the domain name or company name, GPT now sees the tone, services, and branding of the business.
Step 5: OpenAI GPT-4o (Chat Model)
Purpose: Generates an icebreaker, email message, and website summary.
Prompt Design: GPT receives a structured prompt that asks it to generate:
- A personalized sentence that could start an email (the icebreaker)
- A short, first-contact email (2–3 sentences max)
- A website summary that describes the business in 2–3 lines
Output format expected:
{
“icebreaker”: “Nice to see how your clinic offers both in-person and virtual therapy—great blend of modern care.”,
“email_body”: “We help mental health professionals get more clients with location-targeted ads and automated booking systems.”,
“website_summary”: “A therapy clinic offering in-person and virtual services for individuals and couples across California.”
}
Why it matters: GPT-4o brings empathy, tone, and relevance to each message. This makes the email feel real, not robotic.
Step 6: Airtable – Update Record
Purpose: Writes the AI-generated data back into the Airtable record.
Fields Updated:
- Ice breaker
- Email Body
- Website Summary
Details: Maps the outputs from GPT directly to the appropriate fields in the same record where the lead was found.
Why it matters: Once saved, your Airtable is now fully enriched with personalized outreach messages. You can now sync these to your CRM, outreach tools, or export them as CSV.
Loop Logic
After updating Airtable with one lead’s data, the automation returns to the batch step and picks the next lead. This continues until all new leads are processed.
Conclusion
Personalized cold outreach doesn’t need to be slow or painful. With this automation, you can scrape real website content and use AI to write messages that feel human, thoughtful, and relevant.
It saves hours of writing and ensures your leads get something better than a copy-paste template. If you’re ready to turn cold contacts into warm conversations, this setup is your new best friend.
Download the blueprint used in this blog post
Click here to access the blueprint.
