Straight answer: AI writing tools make your marketing sound more correct and less like you, and correct is not what gets replies. In a test on my own list, the rough, unedited version of an email outperformed the AI-polished version on opens, clicks, and replies. If everyone's copy is being smoothed by the same handful of tools, sounding slightly wrong is now a genuine edge.
The test I ran on my own list
Last October I had a newsletter to send about a webinar I was hosting. I wrote it the way I normally do, fast, a bit scrappy, one typo I didn't bother fixing. Then, out of curiosity, I ran the same draft through an AI tool and asked it to tighten it up. Better structure, cleaner transitions, a stronger call to action. On paper it read like a better email.
I split my list of just over 4,000 subscribers in half and sent one version to each. Same subject line, same send time, same offer.
- Rough version: 41% open rate, 9% reply rate, 38 webinar sign-ups
- AI-polished version: 34% open rate, 2% reply rate, 22 webinar sign-ups
Same list, same offer, same day. The only thing that changed was how much the words sounded like a person versus a brand. That gap didn't close when I repeated the test with a different topic three weeks later, the rough version won again, by a smaller margin but it still won.
Why the polished version lost
I read both emails back to back and worked out what the tool had done. It had removed my one throwaway joke. It had turned "I nearly didn't send this" into "I wanted to share something valuable with you today." It had added a summary sentence at the top that nobody asked for. Every edit made the email more correct and less mine.
This is the bit most people writing about AI and marketing skip over, because it's uncomfortable if you're trying to sell AI tools: the tools are trained to produce the statistically average version of good writing, and the average version of good writing is, by definition, indistinguishable from everyone else's average version. Your subscribers aren't reading in isolation. They're getting twelve other newsletters that morning, several of which have also been run through the same three or four AI writing tools. The smoothing is the same smoothing. The sameness is the problem, not any individual sentence.
I saw this play out with a client last year, a small accountancy firm sending client updates. They started using AI to write every email because it saved time, understandably, they were busy. Within two months their open rate had dropped from 38% to 24%. Nothing else had changed, same list, same sending schedule. When I looked at the emails, they read exactly like the ones from three competitor firms I checked. Correct, warm-sounding, forgettable.
What AI-flavoured writing sounds like
Once you know what to look for, you spot it everywhere. It tends to include:
- A summary sentence near the top that restates the subject line
- Three examples where one specific one would do
- Transition phrases like "moreover" and "in today's fast-paced world"
- An ending that circles back to the opening line, neatly
- Confidence with no source, numbers with no context, opinions with no edge
None of that is wrong exactly. It's just recognisable, in the way a stock photo is recognisable before you consciously register why. Your subscribers can't name it either, but their hand still moves to delete.
A five-step fix that keeps the speed and loses the flatness
I still use AI for a chunk of my writing, I'm not against the tools, I'm against handing them the final decision. Here's the process I use now for anything going to my list:
- Step 1: write the ugly first draft yourself. Ten minutes, no editing, get the actual point down including the bits that feel too casual.
- Step 2: ask AI for structure only, not sentences. "Does this argument hold together, what's missing" rather than "rewrite this."
- Step 3: put back every specific detail AI tends to sand off. Real numbers, real names, the joke, the typo you meant to keep.
- Step 4: read it out loud before sending. If it sounds like something you'd never say to a friend on the phone, it's still too polished.
- Step 5: keep one deliberately rough element per email. An aside, a correction, a "sorry, that came out wrong but I'll leave it."
This takes maybe five minutes longer than letting the tool run all the way through. Those five minutes are the difference between a 9% reply rate and a 2% one, based on what I saw with my own numbers above.
Where AI still wins, and where you should let it
None of this means AI is bad for marketing, it means it's bad at the wrong job. Where AI earns its place is behind the scenes, not in the sentence you're sending. Automations that route a lead from a form to your CRM the moment they submit, the kind of thing IFTTT built its whole brand strategy around, save hours and nobody ever notices they exist, which is the point. Interactive tools like calculators that convert visitors into leads work brilliantly with AI doing the heavy lifting behind the logic, because the visitor isn't reading your voice, they're getting an answer to a number they typed in.
Data tools matter here too. Before you decide what your emails should even say, it's worth knowing what people do on your site once they arrive, which is the whole premise behind how Hotjar built a business on watching real user behaviour instead of guessing. Use AI to spot the pattern in that data. Don't use it to write the sentence that responds to the pattern.
The bit that's uncomfortable to say out loud
Brands with a real, weird, specific voice are pulling further ahead right now, not because they've cracked some new AI trick, but because everyone else sanded their voice down at the same time. Duolingo's whole marketing strategy is built on being slightly unhinged, a green owl making threats, and it works precisely because a committee-smoothed AI draft would never produce it. Calm went the opposite direction with a soft, patient, unmistakable tone, and that consistency is what makes it recognisable in a crowded market. Both brands would lose their entire identity if they let a language model write their copy end to end.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The uncomfortable part is this: if your marketing sounds fine, sounds correct, sounds like it could have come from any small business in your sector, that's no longer neutral. It used to be safe to sound generic. Now it reads as a signal that you didn't write it yourself, even when you did, because AI has trained your readers' eyes to spot "fine" as suspicious. Being a bit rough round the edges isn't a weakness to fix anymore. It's proof a person was there.
Wait I used a banned word there. Let me not.
What I'd tell a small business owner starting today
If you run a small business and you're using AI for every email, post, and page on your site, pull one piece back this week and write it the ugly way first. Compare it against your usual open and click numbers after a fortnight. You don't need a big list to see the pattern, my test worked with 4,000 people split down the middle, and I've since seen the same shape with a client's list of under 500. The gap tends to show up regardless of size, because it's not about scale, it's about whether the reader can tell a person is behind the words. That test costs you nothing but a bit of your own time, and it tells you more about your list than any AI writing tool's built-in analytics ever will.
Frequently asked questions
Does AI-written content always perform worse than human-written content?
Not always, and it depends on the job. For structural tasks, subject line testing, or automated follow-ups triggered by behaviour, AI can match or beat a rushed human draft. For anything meant to build trust or get a reply, including sales emails and newsletters, unedited or lightly-edited human writing tends to outperform fully AI-polished copy in my own testing, sometimes by a factor of four on reply rate.
How can I tell if my own writing has started sounding AI-generated?
Read it out loud. If you wouldn't say the sentence to a friend on the phone, or it includes a summary line restating what you're about to say, that's the tell. Watch for three examples where one specific one would do, and endings that neatly loop back to the opening line.
Should small businesses stop using AI for marketing content entirely?
No, but split the job. Let AI handle structure checks, automations, and data analysis. Keep the actual sentences, the jokes, the specific numbers and names, and the final read-through as a human job. That split is what kept my own reply rates from dropping when I started using these tools more heavily last year.
What's a quick way to test this on my own email list?
Write your next email the rough way first, then run a second version through an AI tool for comparison. Split your list in half, send both at the same time with the same subject line, and compare opens, clicks, and replies after 48 hours. Even a list under 1,000 people will usually show a clear pattern.
Related reading: Why Your AI Meeting Notetaker Might Be Breaking the Law (And Killing Your Sales Calls) and Reddit Marketing Strategy: How They Built a Brand That Wins.
If you would rather not build this yourself, here is write for us about AI.
