Personalize cold email openers from lead websites
Deep research + outreach system with AI. Task: Deep research + outreach system with AI TLDR: Building an AI-powered cold email system with ideas I got from viral videos on YouTube. I rebuilt it for my own business, added advice from two other popular videos, and documented the parts the videos skip. Part 1 is the what, why, and how, and Part 2 puts my results next to the creators’. If you’re weighing a system like this for your own business, you can book a free 30-minute AI diagnostic. Every week I see another video about how AI found 1000s of leads, added tens of thousands in profit, or made this or that go viral. I’m tempted by them too. But I exist in this weird space between “Yeah, that’s possible...” and “No way, that’s too good to be true.” I know AI is powerful, and yet I know things can’t be as easy as most of these videos make them out to be. And then I remembered I have free will, a reasonably capable brain, and a powerful AI. Not to mention, I build and advise businesses, and write content as well. Some of these builds look like they’d save me hours. So I’m starting a series where I take viral AI builds and strategies, build them into my own business, and report back on the parts the videos don’t cover. I follow their advice as closely as I can. Then I put my numbers next to theirs, and you can decide what you think. The first one is an AI cold email system. This is Part 1, where I explain what I built, why, and how. Part 2 will have the results. Know that AI can help, but don’t know where to start? AI can take a lot of work off a small business. Knowing which part to hand over first is the hard bit, and so is getting it to work the way you work. I offer a free 30-minute AI diagnostic. We look at where your week goes, where you’re stuck, and what AI could realistically take off your plate. Then I tell you the first move I’d make. Afterward I send you a written plan, so you can decide in your own time. Book a free 30-minute AI diagnostic → A viral AI outreach researcher When I looked at my business, the biggest lever I could pull was more leads. I have an offer and I have some proof, but almost all my leads come from inbound, people who find me through my writing. In $100M Leads, Alex Hormozi lists four core ways to get leads, and cold outreach is the one I’ve barely used. His own benchmark for it is about 1 customer for every 100 personalized cold emails, which only works if you can send enough of them and still make each one feel personal. That’s why I pay close attention when people online say AI can do the sending, the research and the personal touch for you, thousands of emails at a time. Videos like these get a lot of views: This n8n Sales Automation Won $5,000 in Just 14 days (Nate Herk) This n8n Sales Automation Won $5,000 in Just 14 days (Nate Herk) I Let Claude AI Get Me Clients for 30 Days (240 Meetings Booked) (Jordan Platten) I Let Claude AI Get Me Clients for 30 Days (240 Meetings Booked) (Jordan Platten) I Sent 1,500,000 Cold Emails Last Month, Here’s What Works in 2026 (Aaron Shepherd) I Sent 1,500,000 Cold Emails Last Month, Here’s What Works in 2026 (Aaron Shepherd) For this experiment, I pulled from three of them. Each one covered a different part of the system. Here are my inspirations. Nick Saraev: the build The main build comes from Nick Saraev’s video, “I Deep-Personalized 1000+ Cold Emails Using THIS AI System.” It has about 119k views. He published it in May 2025, so it’s about 17 months old now. His system runs in n8n. It pulls leads from Apollo using a scraping tool called Apify, keeps the ones with an email and a website, and then reads several pages of each person’s website. AI summarizes each page and writes a few personal opening lines for the email. Those lines go into a spreadsheet, and he loads the spreadsheet into a sending tool. What he says it does for him: It “routinely” gets him 5 to 10% reply rates on his campaigns. It “routinely” gets him 5 to 10% reply rates on his campaigns. One pitch got a 4% reply rate and over 30 qualified leads who wanted calls. One pitch got a 4% reply rate and over 30 qualified leads who wanted calls. It’s one of the ways he scaled his agency to $72,000 a month. It’s one of the ways he scaled his agency to $72,000 a month. I like that it reads the whole site, but dislike that it pretends the sender actually did it themselves. Nate Herk and Suvam Khadka: the sending; tools: n8n. Human role: Writes a draft opening line that I review.; limitations: Prerequisites include n8n, AI API access (via OpenRouter), and a sending tool like Smartlead ($39/month). Inbox warm-up takes weeks, delaying launch. Lead quality varies significantly by source; the author found Podcast Index yielded 40% usable leads vs. 6-8% from Apollo. Human review is mandatory for every email. The system does not automate sending without manual approval..
Tools
Published on Agent Work Atlas:
Overview
Deep research + outreach system with AI
Workflow
Deep research + outreach system with AI TLDR: Building an AI-powered cold email system with ideas I got from viral videos on YouTube. I rebuilt it for my own business, added advice from two other popular videos, and documented the parts the videos skip. Part 1 is the what, why, and how, and Part 2 puts my results next to the creators’. If you’re weighing a system like this for your own business, you can book a free 30-minute AI diagnostic. Every week I see another video about how AI found 1000s of leads, added tens of thousands in profit, or made this or that go viral. I’m tempted by them too. But I exist in this weird space between “Yeah, that’s possible...” and “No way, that’s too good to be true.” I know AI is powerful, and yet I know things can’t be as easy as most of these videos make them out to be. And then I remembered I have free will, a reasonably capable brain, and a powerful AI. Not to mention, I build and advise businesses, and write content as well. Some of these builds look like they’d save me hours. So I’m starting a series where I take viral AI builds and strategies, build them into my own business, and report back on the parts the videos don’t cover. I follow their advice as closely as I can. Then I put my numbers next to theirs, and you can decide what you think. The first one is an AI cold email system. This is Part 1, where I explain what I built, why, and how. Part 2 will have the results. Know that AI can help, but don’t know where to start? AI can take a lot of work off a small business. Knowing which part to hand over first is the hard bit, and so is getting it to work the way you work. I offer a free 30-minute AI diagnostic. We look at where your week goes, where you’re stuck, and what AI could realistically take off your plate. Then I tell you the first move I’d make. Afterward I send you a written plan, so you can decide in your own time. Book a free 30-minute AI diagnostic → A viral AI outreach researcher When I looked at my business, the biggest lever I could pull was more leads. I have an offer and I have some proof, but almost all my leads come from inbound, people who find me through my writing. In $100M Leads, Alex Hormozi lists four core ways to get leads, and cold outreach is the one I’ve barely used. His own benchmark for it is about 1 customer for every 100 personalized cold emails, which only works if you can send enough of them and still make each one feel personal. That’s why I pay close attention when people online say AI can do the sending, the research and the personal touch for you, thousands of emails at a time. Videos like these get a lot of views: This n8n Sales Automation Won $5,000 in Just 14 days (Nate Herk) This n8n Sales Automation Won $5,000 in Just 14 days (Nate Herk) I Let Claude AI Get Me Clients for 30 Days (240 Meetings Booked) (Jordan Platten) I Let Claude AI Get Me Clients for 30 Days (240 Meetings Booked) (Jordan Platten) I Sent 1,500,000 Cold Emails Last Month, Here’s What Works in 2026 (Aaron Shepherd) I Sent 1,500,000 Cold Emails Last Month, Here’s What Works in 2026 (Aaron Shepherd) For this experiment, I pulled from three of them. Each one covered a different part of the system. Here are my inspirations. Nick Saraev: the build The main build comes from Nick Saraev’s video, “I Deep-Personalized 1000+ Cold Emails Using THIS AI System.” It has about 119k views. He published it in May 2025, so it’s about 17 months old now. His system runs in n8n. It pulls leads from Apollo using a scraping tool called Apify, keeps the ones with an email and a website, and then reads several pages of each person’s website. AI summarizes each page and writes a few personal opening lines for the email. Those lines go into a spreadsheet, and he loads the spreadsheet into a sending tool. What he says it does for him: It “routinely” gets him 5 to 10% reply rates on his campaigns. It “routinely” gets him 5 to 10% reply rates on his campaigns. One pitch got a 4% reply rate and over 30 qualified leads who wanted calls. One pitch got a 4% reply rate and over 30 qualified leads who wanted calls. It’s one of the ways he scaled his agency to $72,000 a month. It’s one of the ways he scaled his agency to $72,000 a month. I like that it reads the whole site, but dislike that it pretends the sender actually did it themselves. Nate Herk and Suvam Khadka: the sending
Reported result
The system successfully generates drafts at a low cost (approx. $0.006-$0.014 per lead). However, actual campaign results (reply rates, conversions) are not yet available as the author is still warming up inboxes and preparing for a 300-email test. The author notes that initial AI drafts were accurate but 'flat,' requiring prompt refinement for better sales copy. No final business impact metrics are reported yet.
Human role
Writes a draft opening line that I review.
Limitations
Prerequisites include n8n, AI API access (via OpenRouter), and a sending tool like Smartlead ($39/month). Inbox warm-up takes weeks, delaying launch. Lead quality varies significantly by source; the author found Podcast Index yielded 40% usable leads vs. 6-8% from Apollo. Human review is mandatory for every email. The system does not automate sending without manual approval.
Try it
I want to build a semi-automated cold email research and drafting system using n8n. Inputs: - Lead source: [replaceable: e.g., Podcast Index CSV, Apollo export, or LinkedIn list] - Target ICP: [replaceable: e.g., Wellness Coaches in US/Canada] - My Offer: [replaceable: e.g., Free 5-min video assessment draft] - Output Location: [replaceable: Google Sheet ID or Airtable Base] Steps: 1. Ingest leads and filter for valid email and website. 2. Use AI to scrape the lead's website (homepage + about page) and extract 3 specific facts. 3. Generate a personalized opening line using these facts and my offer context. 4. Save the lead data, facts, and draft to the output sheet. 5. STOP. Do not send. Deliverables: - A working n8n workflow JSON. - A sample run with 5 leads showing the extracted facts and drafted lines. Quality Check: - Ensure drafts are not generic. - Verify all facts are sourced from the lead's site. Safeguards: - Run a dry test with 5 leads first. - I will manually review all drafts before any sending tool is connected. - Ask me for clarification if the lead source format is unclear.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.