How I Set Up Cold Email With AI in 2026

In this blog post I’m going to walk you through how I set up cold email with AI in 2026. The mailboxes, the tools, the warm-up, and the parts I handed to AI instead of doing by hand. I’ve run cold outreach before, years ago, back when it meant doing every step yourself in a spreadsheet at midnight. I hadn’t touched it in years. So I was curious about two things. How does cold email perform in 2026? And how much of it can I now hand to AI? This is what I found.

This is Week 16 of my rebuild in public, the series where I take my business apart on the internet and show you what’s underneath. Some weeks that’s my website. Some weeks it’s my newsletter. This week it’s outreach, because a business that only grows when people happen to find it isn’t a business. It’s a lucky streak. And luck runs out.

The short version. Cold email with AI in 2026 still needs the same boring plumbing it always did. Separate sending domains, warmed-up mailboxes, and strict daily limits. What’s changed is how much AI now does the heavy lifting on the list, the research and the copy. I set up 21 mailboxes across a handful of lookalike domains, let AI do the parts it’s good at, and started slow. Here’s how, and what happened when I pressed send.

The rebuild in public, week by week

If you’re new here, this is a series. Every week I fix one broken part of my business in public and publish the receipts, the good and the embarrassing. Here’s the whole road so far:

• Week 1: I’m de-indexing 1,300 pages to save my website
• Week 2: my open rate crashed to 11%
• Week 3: the unglamorous SEO work nobody talks about
• Week 4: crawled, not indexed
• Week 5: I took my open rate from 11% to 70%
• Week 6: conquering my inbox instead of my blog posts
• Week 7: the income stream that saved my business three times
• Week 8: I let Claude loose on my own website
• Week 9: I redesigned my whole newsletter
• Week 10: I built a tool to automate my WordPress publishing
• Week 11: I dug 786 de-indexed pages back out
• Week 12: AI inbox management gave me my life back
• Week 13: the month my website started paying me again
• Week 14: my rebuild in public numbers, the month traffic crossed half a million
• Week 15: how I got found by AI
• Week 16: setting up cold email with AI, which is where you are right now

Why I tested cold email again, and with AI this time

Here’s the honest reason. Most of my new business has come from people finding me. Search, referrals, the newsletter, someone reading a post and booking a call. That’s lovely when it works. It’s terrifying when it goes quiet.

I’ve written before about the feast and famine cycle, the thing where you’re either drowning in work or refreshing your inbox wondering if everyone forgot you exist. The only cure I’ve found is having a way to start conversations on purpose, instead of waiting for them to start on their own.

I did this years ago, the hard way, by hand. What I wanted to know in 2026 was simple. Does cold email still work now that inboxes are a warzone? And how much of the grind can I hand to AI so I’m not the one copying rows into a spreadsheet at midnight? I use AI every day across my business, and I wrote about a lot of it in my guide to using AI for lead generation. Cold email felt like the obvious next test.

So before I wrote anything clever, I had to build the plumbing. That’s where we start.

What a cold email campaign with AI needs

Before you send a single email, you need three things in place. AI does not change any of them. Miss one and you’ll either land in spam or get your main email account suspended.

Why you can’t send cold email from your main Gmail

This is the mistake almost everyone makes, and it’s the one that costs the most. If you load up your normal work inbox and start firing off two hundred cold emails a day from it, one of two things happens. Either the inbox providers flag you as a spammer and your deliverability collapses. Or Google decides you’re a risk and locks the account. Now you can’t email cold prospects and you can’t email the people who pay you.

Your main email is the one thing you cannot afford to lose. So you never send cold email from it. AI or no AI.

Separate sending domains, not your real one

Instead, you send from lookalike domains. If your main site is yourname.com, you buy something like getyourname.com or heyyourname.com, and you send from those. If a sending domain gets burned, your real one is untouched. Your website, your brand and your normal email carry on as if nothing happened.

Mailboxes, a sending tool, and a clean list

On each domain you create a few mailboxes, connect them to a sending tool that handles warm-up and sequencing, and point them at a clean, verified list. That last word matters. A list full of dead addresses will bounce you straight into spam. This is the first place AI earns its keep, and I’ll come to it.

How I set up cold email with AI, step by step

Here’s the whole build, and exactly where AI did the work instead of me.

Step 1: buy a few lookalike domains

I started with a few domains that look like mine but are not mine, to keep every scrap of cold-email risk away from my real website and inbox. You don’t need a clever name. Buy three or four to start. They cost about the price of a coffee each, per year.

Step 2: create the mailboxes

On each domain I created three mailboxes, so no single address does all the sending. I used Zapmail to buy and host them, which sets up a stack of separate sending addresses quickly. I ended up with 21 mailboxes in total. I didn’t start there. I started with nine, saw it working, and added twelve more once I trusted the setup.

Step 3: set up the boring technical bits

Every sending domain needs three records set up. SPF, DKIM and DMARC. In plain English, these tell Gmail and Outlook “yes, this email really is from who it says it’s from.” Without them you look like a forgery and get treated like one. Most sending tools set these up for you or hand you the exact records to paste in. It takes ten minutes per domain and it’s not optional. You can check your own domain’s health for free with Google Postmaster Tools.

Step 4: warm the mailboxes up, then wait

A brand new mailbox that immediately starts sending cold email looks exactly like a spammer. So before you send anything real, you warm it up. I used Smartlead to connect and warm every mailbox, quietly sending friendly little emails between them, opening and replying, slowly building a reputation that says “normal person, normal email.” I set every mailbox to ramp up gradually with a healthy share of replies, then left them alone. For two to three weeks. This is the step everyone wants to rush, and rushing it is how campaigns die before they start. You cannot cheat the clock here.

Step 5: let AI build and clean the list

Here’s the first big AI job. Instead of hand-collecting prospects, I used AI to build the target list, pull the right kind of businesses, and organise the details I needed. Then I ran the whole list through MillionVerifier to strip out dead and risky addresses before a single email went out. A clean list is the difference between the inbox and the spam folder, and AI turned a week of dull copy-paste into an afternoon.

Step 6: let AI research and draft, then you edit

The second big AI job is the copy. I used AI to research each type of prospect and draft the outreach, then I edited every line so it sounded like a person and not a template. One rule I never break. I don’t let AI write the final version of my first message. The first message is the most important fifty words in your whole pipeline. AI is brilliant for research and first drafts. The human edit is where the reply is won or lost. The exact prompts I used to research prospects and draft the sequences are in my free Cold Outreach Prompt Pack, so you can copy the system rather than start from a blank page.

The tools I used, and what I let AI do

For anyone who just wants the stack, here it is, plain.

• Zapmail, to buy and host the mailboxes.
• Smartlead, to warm the mailboxes and send the sequences.
• MillionVerifier, to clean the list so it doesn’t bounce.
• AI (I use Claude and ChatGPT), for building the list, researching prospects, and drafting the copy that I then edit by hand.

The pattern is the one I use everywhere in my business. AI does the volume and the grunt work. I do the judgement, the voice, and the final call.

What happened when I pressed send

I promised you receipts, so here they are, warts and all.

My first campaign went out to a list of past contacts. In the first stretch the numbers looked like this. Just under 500 emails sent. A bounce rate of 2.6%, which is well inside the safe zone, and a sign the AI-cleaned list did its job. A small handful of replies.

Here’s the honest part. Most of those replies were auto-responders. Out of office. “I’ve left the company.” The machines answering the machines. In that first week, exactly one real human wrote back with a real reply.

One.

Before you close the tab in despair, that is completely normal. Cold email reply rates sit around one to three percent when you do it well. Anyone promising you twenty percent is selling you a course, not a strategy. Cold email is a volume game played slowly. You’re not looking for everyone to say yes. You’re looking for the small percentage who were quietly in the market at the exact moment your email arrived.

The plumbing worked. Emails landed, they didn’t bounce, they didn’t get me flagged. And AI took the two most miserable jobs, the list and the first draft, off my plate entirely. That was the win I was measuring in week one. The replies are a longer game, and I’ll report those numbers as they come in, the same way I always do.

What I’d do differently

Two things. First, I’d have set up more mailboxes at the start, not fewer. I started with nine, got nervous, and only later added twelve more. Warm-up time is the bottleneck, so the earlier you start it, the better. Second, I’d trust AI sooner on the list and the research, and spend the time I saved on the human edit of the copy. That’s the half that really moves the reply rate.

Should you set up cold email with AI?

Honestly? Not everyone should.

If you sell to a small, specific group of businesses and you know who they are, cold email with AI is one of the best tools you have. You can start a conversation with someone who’s never heard of you, on purpose, on a Tuesday, and let AI do the grind. That’s powerful.

If you sell to lots of consumers, or you’re relying on volume and vibes, your time is better spent elsewhere. You might get more from using Reddit for lead generation or from building the content that gets you found in the first place. Cold email is a scalpel, not a net.

And if you do go for it, do the setup right. The number of people who skip the domains and the warm-up, blast from their main inbox, and then declare “cold email doesn’t work” is heartbreaking. It works. Their version of it doesn’t.

Frequently asked questions

Can you really automate cold email with AI?

You can automate a lot of it, but not all. AI is excellent at building and cleaning your prospect list, researching each type of prospect, and drafting the outreach. What you should not fully automate is the final version of your first message, because that’s what decides whether anyone replies. The winning setup is AI for the volume, a human for the judgement and the voice.

How much does it cost to set up cold email with AI?

Less than you’d think. A few lookalike domains cost roughly the price of a coffee each per year. Mailbox hosting, a sending tool and a list verifier run to a modest monthly fee, usually under a hundred pounds a month for a small setup. Your AI tools you likely already pay for. The expensive part isn’t money, it’s the two to three weeks of warm-up before you can send.

How long before I can start sending real emails?

Plan for two to three weeks of warm-up per mailbox before you send a single cold email. New mailboxes that start blasting straight away look like spam and get treated like it. AI can speed up the list and the copy, but it cannot speed up warm-up. That wait is the price of landing in the inbox.

What tools do I need for cold email with AI?

At a minimum, a place to buy and host lookalike-domain mailboxes, a sending tool that warms them and runs your sequences, a list verifier to strip dead addresses, and an AI assistant for the list and the copy. I used Zapmail, Smartlead and MillionVerifier alongside Claude and ChatGPT, but the categories matter more than the brands.

Is cold email legal?

In most places, business-to-business cold email is legal if you’re relevant, honest about who you are, and give people an easy way to opt out. Rules vary by country, so check the ones that apply to you. The spirit is simple. Email people who might reasonably want to hear from you, and stop the moment they say no.

The final word

I spent years telling myself I didn’t need cold email. I had search, I had referrals, I had a newsletter. And most of the time, that was true. Then it went quiet, the way it always eventually does, and I realised I’d built a business that could only grow when the world came to me. That’s not a business. That’s a wish.

Setting up cold email with AI in 2026 isn’t glamorous. It’s domains and DNS records and three weeks of waiting for mailboxes to warm up. But at the end of it I have something I didn’t have before. A way to start a conversation with someone who’s never heard of me, whenever I choose, with AI carrying the parts that used to eat my evenings.

One real reply in week one. That’s the receipt. Small, unimpressive, and exactly what week one of a real cold email campaign looks like. I’ll take the small and real over the big and fake every time. And I’ll see you next week with the next thing I’ve taken apart.

Want the prompts and templates I used?

Every Sunday morning I send a newsletter to fifteen thousand entrepreneurs, marketers and business owners. Same voice as this post, half the length, and it’s where I share the experiments before they hit the blog. Sign up here. One email a week.

If you want the exact prompts and templates behind this, grab my free Cold Outreach Prompt Pack and have a dig through the rest of my free resources library.

And if you want help building AI systems like this into your own business, without hiring an agency or drowning in tools, that’s what I do. I work with a small number of business owners at a time. Details on the work with me page here.

Related reading: How to Build a Personal Brand as a Consultant in 2026 and AI Prompts for B2B Marketing: Templates That Actually Work in 2026.

Free resource: grab The Onboarding Email Drip Template from the resource library.