The short version: AI is brilliant for grammar checks, research and knocking out a first draft fast, but a fully AI-written email to a real prospect reads differently to a human eye than it does to yours, and people notice more than you think. I tested this on my own outreach for three months and my reply rate roughly halved. What worked better was a five-step hybrid process, and I’ll give you the exact one I use with clients now.
The three months I handed my outreach over to AI
Last spring, mid way through rebuilding my client list from scratch, I did what a lot of small business owners were doing: I let ChatGPT write my first-touch emails. Not the whole strategy, just the actual messages I was sending to warm leads and old contacts I hadn’t spoken to in a year or two. I gave it my tone of voice, some sample emails, a clear brief each time. It saved me maybe twenty minutes a day.
In that first month I sent 62 emails. Nine replies. A 14.5% open-to-reply conversion that looked fine on paper until I compared it to my own numbers from the year before, which sat closer to 11% replies on a cold list and nearer 30% on a warm one.
The second month I got lazier, not more careful, which is the honest bit most people leave out of these stories. I stopped editing the AI drafts line by line and started sending them almost as they came out, because I was busy and the emails read fine to me. 58 sent. Four replies. Under 7%.
Third month, out of frustration more than strategy, I went back to writing the first line myself every time and using AI only for the middle section and a grammar pass. 54 sent. Fourteen replies. Just under 26%.
Same list quality, same offer, same time of year. The only variable was how much of the email was mine.
Where the AI voice gives you away
Here’s the part that made me uncomfortable when I sat down and read a stack of my own AI-drafted emails side by side. They were fine. Grammatically clean, warm enough, correctly punctuated. And they all had the same rhythm. Three-word opener. A compliment. A pivot sentence starting with “I noticed” or “I wanted to reach out because.” A tidy close with a soft call to action.
People who get thirty emails a day have started to feel that rhythm the way you feel a cold caller’s script two words in. They might not consciously clock “this was written by AI,” but they clock “this was written for anyone,” and they treat it accordingly. Filed, or deleted, rarely answered.
This is the bit nobody wants to say out loud when they’re trying to sell you an AI writing tool: polish and connection are not the same thing, and past a certain point more polish actively works against you. A slightly clumsy, slightly specific human sentence outperforms a smooth generic one, every time, in my own numbers and in every client account I’ve looked at this year. If you want the deeper version of how brands that do get this balance right operate, the Grammarly marketing strategy is worth reading, because even the company that built its entire business on “cleaner writing” is careful never to sell “identical writing.”
The five-step process I use now
I still use AI every single day. I just changed where it sits in the process. This is the version I now teach clients, word for word:
- Write the opening line yourself, always. One sentence, referencing something specific and true about this exact person or business. No AI tool knows that their office moved last month or that they mentioned a book at a conference.
- Brief the AI for the middle, not the whole email. Give it the point you want to make and the outcome you want, and let it draft the body. This is where AI earns its keep, structuring an argument, tightening a paragraph, cutting waffle.
- Cut every sentence that could go to anyone else on your list. If a line would work unchanged in an email to a completely different prospect, delete it or make it specific.
- Read it out loud before sending. If you wouldn’t say a sentence in a phone call, don’t send it in an email. This single step catches most of the “AI tell.”
- Send from a smaller batch, more often. Twenty edited emails beat sixty templated ones. I now send in batches of fifteen to twenty and spend real time on each one, rather than mass producing.
It takes longer than pure AI drafting. It takes less time than writing from a blank page. And the numbers hold up.
Where AI email tools do earn their keep
I don’t want this to read as anti-AI, because it isn’t. There are three jobs I now happily hand over completely.
Subject line testing is one. I’ll ask for ten variations on a subject line and pick the one that sounds least like a subject line, because the most “email marketing” sounding one is usually the one people scroll past. Research is another. Before I email anyone now, I ask AI to summarise their recent public posts, their company’s last funding round or product launch, anything I can turn into that first line I insist on writing myself. And the internal side, meeting notes, follow-up summaries, drafting the boring administrative emails nobody reads for tone, that’s a total gift. If your team collaborates on any of this, a shared board like the one Miro built its entire marketing strategy around shows how useful visual, shared workspaces get once you stop trying to make every single piece of output sound like a person and just let the tool handle logistics.
There’s also a fourth use worth mentioning because almost nobody talks about it: using AI to build something interactive that does the selling for you, so you’re not relying on email tone at all. A well built interactive calculator on your website converts a cold visitor into a warm lead without you writing a single persuasive sentence, because the tool does the persuading through a number the visitor generated themselves. I’ve had clients see calculator-based lead capture convert at three or four times the rate of a generic contact form, simply because the visitor feels like they got something rather than being asked for something.
When DIY stops being the cheap option
I’ll say the quiet part here too. There’s a point where trying to fix this yourself, tweaking prompts, testing subject lines, rewriting openers one lead at a time, starts costing you more in hours than it would cost to get someone to set up the system once. I see this constantly with business owners who are three years into “just Googling it,” still sending emails that read like everyone else’s, still wondering why replies have dried up.
If you’re at that point, an hour with someone who does AI implementation for small businesses for a living tends to save far more time than it costs, because they’ll spot the rhythm problem in your last twenty emails in about five minutes flat. That’s not a plug for spending money you don’t have, it’s a plug for spending it once instead of bleeding it out slowly over a year of low reply rates.
What this looks like when a whole business rebuilds its voice
I watched this play out at brand level too, not just in one-to-one emails. Businesses that rebuild in public, the way I’ve had to over the last five years, live or die on whether their content still sounds like a person once the team scales and the tools multiply. It’s a big part of why the Wix marketing strategy stayed recognisable through years of product expansion, the tone of voice was treated as a fixed asset, not something that drifted every time a new tool got introduced. Small business owners can steal that exact discipline. Decide what your voice sounds like on a page, write it down in three bullet points, and check every AI draft against those three bullets before it goes out, whether it’s an email, a social caption, or a landing page.
None of this is about distrusting AI. I use it more now than I did a year ago, not less. It’s about being honest that “AI wrote it” and “it worked” are two completely separate questions, and small business owners who conflate them end up with clean, correct, forgettable outreach that nobody replies to.
Frequently asked questions
Should I stop using AI to write emails to prospects completely?
No. Use it for structure, research and grammar, but write your own opening line and read the final version out loud before sending. That single habit fixes most of the problem without costing you the time savings.
How can I tell if my AI-written content sounds like everyone else’s?
Read three of your recent emails or posts back to back. If any sentence could be dropped unchanged into a message to a completely different person or audience, it’s generic and needs cutting or rewriting with a specific detail.
What’s the quickest fix if I don’t have time to rewrite everything?
Write the first sentence of every email yourself, referencing something true and specific about that person or business, and let AI handle the rest. This one change recovered most of my reply rate in testing.
Is it worth paying for help with this rather than fixing it myself?
If you’ve been stuck at a low reply rate for months and can’t work out why, yes. A short session with someone who does AI implementation professionally usually costs far less than the hours you’ll lose tweaking prompts on your own without knowing what to test for.
Related reading: The Best AI Marketing Tools for Small Business in 2026 (And the Ones I’d Skip) and AI Is Reading Your Marketing Emails Before Your Customers Do, And Your Open Rate Doesn’t Mean What You Think.
If you want the full breakdown, here is everything I know about AI marketing.
