Straight answer: Your prospects aren’t ignoring your emails because you used AI to write them. They’re ignoring them because you used AI to write them and then sent them without changing a single word. I ran a real test on my own list, 120 emails split into two batches, and the difference in reply rate was more than triple. The fix isn’t ditching AI, it’s using it for structure and doing the last 20% yourself.
The tell everyone’s learned to spot
I get maybe fifteen cold emails a week now that follow the exact same shape. “I hope this finds you well.” A flattering line about my “impressive work in the AI space.” Three bullet points nobody asked for. A closing question like “Would you be open to a quick chat this week?” I don’t even read past the second line anymore, and I know for a fact most business owners on my list do the same, because I asked eighty of them directly in a LinkedIn poll last month. Sixty two said they now delete anything that “smells like ChatGPT” before finishing the first sentence.
The tells are specific and by now they’re almost a genre. Overuse of “delve” and “unlock” and “handle” (words I’ve banned myself from using for this exact reason). The three-bullet-point structure that appears whether or not the content needs bullets. The closing line that’s grammatically perfect but says nothing a human would say out loud. None of this is subtle anymore. People have been reading AI text for two years, they’ve got a trained eye for it whether they realise it consciously or not.
What I found when I tested it on my own list
Back in February I ran a small experiment because I was curious rather than because I thought it would make a good blog post. I had 120 warm-ish contacts, people who’d downloaded something from me or engaged with a post but never bought anything. I split them into two batches of 60.
Batch one got a fully AI-generated outreach email. I gave ChatGPT the brief, the offer, the tone, and let it write the whole thing with one light edit for factual accuracy. Batch two got an email where I used AI to build the skeleton, the opening hook, the three-point structure, the close, but then I rewrote every sentence in my own voice, kept one deliberately clunky line, and referenced something specific about each person (a post they’d liked, a comment they’d left).
Batch one: 4 replies out of 60. That’s a 6.6% reply rate, which sounds fine until you look at batch two.
Batch two: 14 replies out of 60. A 23% reply rate. Same offer, same list, same send time, same subject line format. The only variable was how much of the final email was mine rather than the model’s.
That’s not a scientific study, it’s one consultant’s list of 120 people. But the gap was too big to be noise, and I’ve since heard similar ratios from three other people I work with who ran their own rough versions of the same test.
The fix that gets replies
The mistake most business owners make isn’t using AI for email. It’s treating the first draft as the final draft. Here’s the process I use now, step by step:
- Give the AI the facts only: who the person is, what you’re offering, one specific detail about them. Not the tone. Not the personality. Just the facts.
- Get the structural draft back and immediately delete the opening line. Every single time. “I hope this email finds you well” and its cousins need to go, no exceptions.
- Rewrite the middle two sentences in your own cadence. If you’d never say “I wanted to reach out because,” don’t send it because AI wrote it that way.
- Add one thing the model couldn’t have known. A specific comment, a shared connection, a detail from their website that shows you looked. This is the single biggest lever in the whole process.
- Cut the length by a third. AI drafts run long by default. Shorter emails with rough edges outperform long polished ones almost every time I’ve tested it.
This takes maybe four extra minutes per email. On a batch of 60 that’s four hours of extra work for roughly ten extra replies. I’ll take that trade every day.
Why the specific detail matters more than the writing
Of the 14 replies in batch two, nine of them mentioned the specific detail I’d included directly in their reply. “Thanks for noticing the post about our Q3 numbers” or similar. People aren’t replying to good writing. They’re replying to proof that a real person looked at them specifically before hitting send. AI can help you find that detail faster (I use it to summarise someone’s recent LinkedIn activity in ten seconds rather than scrolling for five minutes) but it can’t manufacture the fact that you bothered.
Where AI does earn its place
I’m not anti-AI in this. I use it every single day and I’d be lying if I said otherwise. Where it earns its place in outreach and marketing writing:
- Research and summarising, pulling the key facts about a prospect or a competitor fast
- Structure, giving you a skeleton so you’re not staring at a blank page
- Variations, generating five subject line options so you can pick the least generic one
- Data-heavy content, the kind of thing Ahrefs built its entire marketing strategy around, useful and specific rather than fluffy
Where it fails, and fails consistently: anything that’s supposed to sound like it came from one specific human with opinions. That’s most of what a small business’s email marketing and cold outreach is. Big brands can get away with more polish because the relationship is with the brand, not a person. That’s why Disney’s marketing strategy can lean on consistent, almost impersonal brand voice across every channel. A one-person consultancy or a ten-person agency doesn’t have that luxury. People are buying you.
The uncomfortable bit nobody selling AI courses wants to say out loud
Here’s the part that doesn’t make it into most “AI for small business” content, because it’s bad for anyone trying to sell you a prompt library. Using AI to write your outreach faster doesn’t save you time overall if it drops your reply rate to a third of what it was. You’ve just made yourself faster at wasting your own list’s attention. I’ve watched people proudly tell me they sent 500 cold emails in an afternoon using a tool that batch-generates them, and get four replies. I could have written 80 emails in that same afternoon and got twenty. Speed isn’t the win if the output doesn’t land.
This is the same lesson Howard Schultz built Starbucks’ early culture on, long before AI existed. In the business lessons from Howard Schultz I’ve written about before, the thing that made Starbucks different in its first decade wasn’t the coffee, it was insisting that every interaction feel like it came from someone who cared, at a time when most companies were racing to make service faster and cheaper. Faster and cheaper email outreach is exactly the trap AI makes easy to fall into now.
And Napoleon Hill’s idea of “definiteness of purpose,” the one I keep coming back to in the business lessons from Napoleon Hill I wrote a while back, applies directly here. A vague email sent to 500 people with no clear reason for that specific person to reply is the opposite of definite purpose. A shorter, rougher email sent to 60 people with one true, specific reason attached to each one does the job Hill was describing a century ago.
What to do if you’re rebuilding your outreach this month
If you run cold email or warm follow-ups for your business, audit your last 20 sent emails honestly. Read them as if you’re the recipient. If you can’t tell they came from a specific human rather than a template, that’s your answer, and it’s fixable this week, not next quarter. Pull up your open and reply rates in whatever tool you use and be honest about the number. Most small business owners I talk to haven’t looked at their actual reply rate in months, which means they’re guessing at what’s working.
Brands that get this right treat every touchpoint, email included, as part of the same voice. Away’s marketing strategy is a good example of a brand that kept its written voice consistently direct and human across every channel even as it scaled, which is exactly what most small businesses lose the moment they hand the keyboard entirely to a model.
And if writing isn’t the only place your website is losing people, look at how much your site invites interaction rather than just describing things at visitors. Pages built around interactive calculators consistently keep people on the page longer and convert better than another paragraph of polished copy, AI-written or not, because people engage with things they can do, not things they read.
If you want help building a proper system around this, rather than guessing at prompts on your own, this is exactly the kind of work I do hands-on with clients as an AI consultant for small business, setting up processes where AI handles the structure and your team handles the human part, in the right order.
Free resource: The Closing Question Swipe File.
Frequently asked questions
Is it obvious when an email was written by AI?
Yes, more often than people think. Generic openers, the three-bullet-point structure, and closing lines that are grammatically perfect but say nothing specific are the clearest tells. Most recipients now notice these patterns even without consciously trying to spot them.
Should I stop using AI for cold email entirely?
No. Use it for the structure, the research, and the subject line variations. Write or rewrite the actual sentences yourself, and always include one detail the model couldn’t have known about that specific person.
How much time does editing an AI draft really add?
In my own test it added roughly four minutes per email on average. Across 60 emails that’s four extra hours of work, which produced more than triple the reply rate compared to sending the unedited AI draft.
Does this apply to marketing emails and newsletters too, not just cold outreach?
Yes, the same pattern shows up in newsletters and follow-up sequences. Subscribers can tell when a whole email sequence was generated in one batch with no editing, and open rates on those sequences tend to drop over time even if the first send looks fine.
Related reading: Shadow AI: What Your Team Is Doing With ChatGPT Behind Your Back and How I Set Up Cold Email With AI in 2026.
