The short version: AI hasn’t made cold email worse because AI is bad at writing, it’s made cold email worse because everyone is using it the same lazy way, which means every inbox now gets five nearly identical messages a day and people delete on pattern recognition alone. I tested this on my own outreach list, cut my reply rate almost in half using a popular AI email tool, then rebuilt the process by hand and got it back above where I started. The fix isn’t “write more like a human,” it’s a specific set of rules about what AI is allowed to touch and what it isn’t.
What I did (and what it cost me)
Last spring I ran a small experiment on 200 warm-ish leads, people who’d downloaded a guide from my site or engaged with a LinkedIn post, so not cold cold, but not customers either. I split them into two batches of 100. Batch one got emails drafted almost entirely by an AI writing tool, lightly proofed by me for typos. Batch two I wrote by hand, using the AI only for research on the company beforehand.
Batch one: 6 replies. Batch two: 14 replies. Same offer, same subject line length, same send time, same list quality. The only variable was who wrote the words.
That’s not a scientific study, I know that, I run a marketing business not a lab. But I’ve since compared notes with three other consultants doing similar tests on their own lists, and the gap was always somewhere between two and three times in favour of the human-written batch. One of them, a business coach in Manchester, told me her AI-drafted sequence got a 1.8 percent reply rate. Her old hand-written one, sent to a colder list, got 4.9 percent.
The pattern recipients have learned to spot
Here’s the uncomfortable bit nobody selling AI email tools wants to say out loud: your prospects have seen thousands of these emails now. They don’t need to read past the first line to know it’s a template, because the templates all share the same fingerprints.
- The opener that name-checks something specific but generic, like “I saw you recently expanded your team”
- The three-sentence flattery paragraph before the ask
- “Quick question” or “curious if” as the pivot line
- A soft close that offers “15 minutes” with zero friction, which somehow makes it feel more like spam, not less
- Perfect grammar with no actual personality, no opinion, nothing you could disagree with
I get four or five of these a week myself, from people trying to sell me things I already use. I can spot them in the subject line half the time. If I can spot them running my own business at 53, a busy operations manager scanning 40 emails before 9am can spot them in about two seconds.
The tools aren’t the problem. I use AI for research, first drafts, subject line variants, all of it. The problem is that thousands of small business owners are using the same three or four tools with the same default prompts, which means the output converges. It’s the same reason a lot of AI-written blog posts read like the same person wrote them, which I’ve written about before when looking at brands that built something distinct instead, like the way Duolingo built a voice nobody else could copy or how Calm built a tone so specific it became the brand. Sameness is the actual cost of convenience, and cold email is where that cost shows up fastest because there’s a measurable number attached to it: the reply rate.
What I changed, step by step
I didn’t drop AI from the process. I changed what it’s allowed to do.
1. AI does the research, not the writing
Before I write a single line, I ask AI to pull together what it can find on the company: recent news, funding, hires, a product launch, a review they left somewhere, anything specific. Not “tell me about their industry,” which gives you generic filler. I want three facts I couldn’t have guessed.
2. I write the first two lines myself, every time
This is the part that costs time and it’s the part I won’t hand over. The opening line has to reference something true and specific enough that it couldn’t apply to any other company on my list. Not “I noticed your company is growing,” but something like “saw you just moved to the new unit off the A26, that’s a big jump from the old Tunbridge Wells site.” That took me forty seconds to write once I had the fact. It’s the forty seconds that gets the reply.
3. I let AI draft the middle, then I cut it in half
AI is useful for turning a rough pitch into a clean paragraph. The problem is it always writes too much. My rule is I take the AI draft of the offer paragraph and cut it by at least 40 percent. If it wrote four sentences, I want two.
4. No AI on the close
The call to action is where AI tools get most generic, “would you be open to a quick chat” is the cold email equivalent of a form letter. I write my own close, usually a direct yes/no question rather than an open invitation, because a specific question is easier to answer than a vague one. “Worth a 10 minute call Thursday or would it make more sense if I just sent over the pricing?” gets more replies than “let me know if you’re interested.”
5. Send from a real inbox, at real send limits
This is separate from the writing but it matters just as much. Sending 500 emails a day from a bulk tool through a warmed-up domain tells spam filters and increasingly tells recipients (through delivery patterns and formatting quirks) that this is mass outreach. I cap my own sends at around 25 to 30 a day from my actual Gmail-connected address, which is roughly what one person can personalise in an hour.
The trade-off nobody selling AI outreach tools mentions
Here’s the part that’s inconvenient: this process is slower than the fully automated version, not faster. AI outreach was sold to small business owners as a time saver, send a thousand emails a night, wake up to booked calls. What’s happened is the market got flooded, spam filters got smarter, and recipients got faster at pattern matching, so the fully automated version now needs a much bigger list to get the same number of replies it used to get from a small one.
I’ve had small business owners tell me they need 3,000 emails a month now to get the meetings they used to get from 400. That’s not AI failing, that’s the market adjusting to AI being used badly at scale. The honest trade is this: AI can compress your research time and your drafting time, which is real and worth having, but it can’t compress the personalisation time without costing you the reply rate, because the personalisation is the entire point of a cold email in the first place. If you’re not willing to spend the forty seconds per prospect, you’re better off putting your energy into an interactive tool people opt into themselves, like a calculator or quiz on your site, because that flips the whole dynamic: they come to you instead of you interrupting them.
A simple system if you’re doing this alone
If you’re a solo founder or a two-person team without a dedicated SDR, here’s the version of this that’s sustainable:
- Build a list of no more than 30 to 50 companies a week, quality over volume
- Spend 10 minutes per company on research (AI-assisted), noting one specific fact
- Write your own opener and close, every single time, no exceptions
- Use AI only for the middle paragraph, and cut whatever it gives you by half
- Follow up twice, spaced five and twelve days apart, each one shorter than the last
- Track replies weekly by hand in a spreadsheet, not just what the tool reports, because open rate tracking through images is increasingly unreliable with privacy changes in most email clients
This is closer to how Dale Carnegie taught people to build relationships nearly a century ago than it is to anything in a modern automation stack, which is a strange thing to say about email software, but the principle hasn’t moved: people respond to being seen as specific individuals, not as rows in a spreadsheet. I wrote about this when I went through what Dale Carnegie still gets right about business relationships, and cold email is maybe the clearest modern test of it. The tool changed. The reason people reply hasn’t.
If you’re trying to build this into an actual repeatable process for a small team rather than doing it one email at a time yourself, this is the sort of workflow I build with clients directly as an AI implementation coach, because the tools are easy to buy and hard to use well, and most of the failure I see isn’t the software, it’s the process wrapped around it.
One more thing worth saying plainly: brands that get cold outreach and cold growth right almost never look like they’re doing outreach at all. Revolut’s early growth and Hotjar’s founder-led approach both leaned on specificity and directness rather than volume, which is the same lesson at a bigger scale. Small business owners don’t need Revolut’s budget to copy that instinct. You just need to write your own opening line.
Frequently asked questions
Does AI-written cold email get lower reply rates than human-written email?
In my own test of 200 leads split evenly, the AI-drafted batch got roughly half the replies of the hand-written batch, and other consultants I’ve compared notes with saw a similar two to three times gap. The tool isn’t the issue, using it to write the whole email with no personal input is.
How much should I still use AI for cold outreach?
Use it for research, for drafting the middle section, and for generating subject line variants to test. Write your own opening line and closing question by hand every time, since those two lines carry most of the weight in whether someone replies.
Why do AI cold emails all sound the same?
Because most small businesses are using the same handful of tools with default prompts, so the sentence structures, openers and closing lines converge. Recipients now see the same patterns dozens of times a week and delete on recognition before reading the offer.
Is cold email still worth doing in 2026?
Yes, but the volume that used to work no longer does. Smaller, better-researched lists with genuine personalisation on the first and last line still get replies. Fully automated high-volume sending mostly gets ignored or filtered now.
Related reading: Miro Marketing Strategy: How They Built a Brand That Wins and Grammarly Marketing Strategy: How They Built a Brand That Wins.
