The short version: Every small business using ChatGPT the same way, with the same prompts, is producing marketing that reads like a photocopy of a photocopy. Your subscribers aren’t unsubscribing, they’re just quietly not opening anymore, and that’s the real damage nobody talks about. Fix it by putting one unmistakably human, specific detail into everything you publish, before AI touches it.
What happened to a client’s open rate in six weeks
In autumn 2025 I worked with a small accountancy firm near Maidstone. Nice people, decent client base, a newsletter that used to sit around a 32% open rate, which is healthy for that sector. They started using AI to draft the whole thing, subject line included, because their marketing person had left and nobody had time.
Six weeks later the open rate was 17%. Not because it went to spam. Not because it looked “AI written” in any obvious way, no stray em-dashes, no “in today’s fast-paced world.” It read fine. That was the problem. It read exactly like fine. Their subscribers, most of whom are small business owners getting a dozen similar newsletters a week, had started skimming past it without registering it was even there.
We didn’t ban AI. We changed one thing: every email now opens with one real, specific sentence from the owner, a client story, a number from that week, a mistake she’d made. The AI still drafted the body. Open rate is back up to 29% as of this year.
The uncomfortable bit nobody selling AI tools will tell you
Here’s what most people writing about AI marketing skip past: your audience doesn’t need to catch AI writing to be bored by it. They don’t run it through a detector. They just feel a low-grade sameness across their inbox and stop paying attention, the same way you stop noticing a hum in a room after twenty minutes.
Researchers studying large language model output have found something similar happens at scale: when lots of people prompt the same model the same way, the writing converges toward the same phrasing, the same rhythm, the same three-point structure. It’s not that any one piece is bad. It’s that thousands of businesses are now producing marketing that sits in the same narrow band of “acceptable, competent, forgettable.”
This is the part that makes people uncomfortable, because the whole pitch for AI in marketing has been speed and cost. Nobody wants to hear that the thing making it fast is also the thing making it invisible. But if you’ve watched your own open rates or engagement slide over the last two years and blamed the algorithm, I’d gently push back. A lot of it is this.
Why “just add personality” isn’t enough advice
Every AI marketing article tells you to “add your personality.” It’s true and it’s useless, because nobody explains what that means in practice. Here’s what it looks like, in order of effort:
- One specific number in every piece. Not “many clients saw growth,” but “one client’s booking rate went from 4 a week to 11.” Specifics don’t get homogenised because AI doesn’t know your specific numbers, you have to put them in.
- One named person or place. A real client (with permission), a real town, a real supplier. AI defaults to vague because it’s trained to avoid liability, not because vague is better marketing.
- A stated opinion the AI wouldn’t volunteer. “I think most small business Instagram advice is wrong” reads nothing like AI output, because AI is built to hedge, not to pick a side.
- Sentence length variation. AI drafts tend to settle into a rhythm of medium-length sentences. Break it. Short one. Then a longer one that runs on a bit, the way people talk when they’re telling you something they care about.
- A mistake or a doubt. “I got this wrong for two years” does more for trust than any AI-generated claim of expertise ever will.
None of this is about writing worse or slower content. It’s about writing the 10% that only you can write, and letting AI handle the 90% that’s structural, the outline, the formatting, the first draft of the boring bits.
Brands that never had this problem, and why
The businesses that don’t sound like everyone else’s AI output are usually the ones with a distinct point of view before they ever touched an AI tool. Look at how GoPro built its entire brand around customer footage rather than polished ads, it’s a strategy that couldn’t be replicated by a prompt because the “content” was always someone else’s real experience, not a description of one. Or look at how a company like Figma grew its marketing through community and product-led storytelling rather than generic thought leadership, the voice came from specific product decisions, not from a content calendar.
Joe Pulizzi has been saying a version of this for years before AI existed as a marketing tool: the content that compounds is the content nobody else could have made, because it’s built on your specific experience, not a category best practice. I still think about his point about owning an audience rather than renting one, because that’s exactly what gets lost when marketing becomes interchangeable. If your newsletter could have been sent by your competitor with the logo swapped, you don’t own that audience, you’re borrowing their attention until they get bored.
A concrete way to make your emails harder to skip
If you use Mailchimp or a similar tool, you already have the data to spot this problem in your own account. Pull your last ten campaigns and check click-to-open rate, not just open rate. A dropping click-to-open rate with a stable open rate means people are opening out of habit but not finding anything worth clicking, which is the exact symptom of AI sameness setting in. It’s worth studying how a company like Mailchimp built its own marketing around plain, slightly odd, very human copy long before AI writing existed, because that’s the tone that still performs best against an inbox full of competent AI drafts.
One fix that consistently works and gets almost no attention: give people something to do rather than something to read. A short interactive tool, a quick calculator, a two-question quiz, breaks the scroll pattern completely, because AI-written paragraphs are all competing for the same passive attention and a tool asks for active attention instead. I’ve written before about how interactive calculators lift engagement and conversion, and it applies just as much inside an email as it does on a landing page. A “work out your real hourly rate” calculator embedded in a newsletter will get more genuine attention than three paragraphs of AI-polished advice, because nobody else’s newsletter is asking the reader to do anything.
The step-by-step audit I run before anything goes out
This takes about fifteen minutes per piece once you’re used to it, and I do it on everything before it’s scheduled:
- Read the opening line out loud. If it could open a hundred other companies’ emails on the same topic, rewrite it with a specific detail from this week.
- Check for at least one number that’s true only for your business, not an industry stat.
- Check for at least one sentence under six words. AI drafts rarely produce these unprompted.
- Look for the word “unlock,” “,” “game-changer,” or any variation of “in today’s world.” If it’s there, cut it, it’s the tell that the sentence came straight from a model with no editing pass.
- Ask whether a competitor’s AI tool, fed the same prompt, could have produced this exact piece. If yes, it needs another pass.
None of this means slowing your output down to a crawl. It means the human pass is now the valuable 15 minutes, not the first draft, which AI can still do in seconds.
Where hiring help pays off
I get asked a lot whether small businesses should just hire someone to sort this rather than do the audit themselves, and honestly the answer depends on volume. If you’re sending one newsletter a month, do it yourself, it’s not worth outsourcing. If you’re running content across email, social and a blog every week and you can feel the sameness creeping in but can’t see where, that’s exactly the situation where a short piece of outside AI consulting for small business marketing pays for itself, because someone outside your business will spot the copy-paste tone faster than you will, since you’re too close to your own drafts to notice the pattern.
The same logic applies to platform strategy generally, not just email. A distinct visual and tonal identity, the kind built by brands studied in things like Instagram marketing strategy breakdowns, gives AI tools something specific to draft around instead of defaulting to generic category language. AI is a much better assistant when it’s working from a clear brand voice document than when it’s asked to invent a voice from scratch, because invention from scratch is exactly where the sameness comes from.
What I’d do differently if I were starting from zero this year
I’d write the brand voice document first, before a single AI-drafted email goes out. Three pages: five words we’d never use, five phrases only we’d say, three real client stories, one strong stated opinion about our industry that we’re happy to defend in public. That document becomes the prompt context for everything after. It takes an afternoon. It’s the single highest-use thing I’ve seen small businesses skip in the rush to “get AI into the marketing,” and it’s the reason two businesses using the exact same AI tool can produce completely different quality output.
Free resource: The Context Document Template for AI.
Frequently asked questions
Can subscribers tell an email was written by AI?
Most can’t identify it consciously, but they feel it. The signal isn’t obvious robotic phrasing anymore, models write cleanly now. The signal is a lack of anything memorable, which shows up as declining click-to-open rate rather than complaints or unsubscribes.
Does using AI to write marketing copy hurt SEO?
Google has said publicly it doesn’t penalise content for being AI-assisted, it penalises content that’s low quality regardless of how it was produced. The practical risk isn’t a penalty, it’s that generic AI content simply doesn’t earn links, shares or return visits, which are the things that help rankings over time.
How much AI is too much in one newsletter?
There’s no fixed percentage that matters. What matters is whether at least one sentence in the piece could only have come from your business specifically. If every sentence could have been written about any competitor with the name swapped, that’s the real problem, not the proportion of words a model helped write.
Should I stop using AI for marketing copy altogether?
No, that’s overcorrecting. Use it for structure, first drafts, headline options and editing speed. Keep the opening line, the specific numbers, the stated opinions and the mistakes as the part you write yourself, because that’s the 10% doing almost all the work of making people stop and read.
Related reading: Why Every AI-Written Newsletter Sounds the Same (And What I Changed In Mine) and Why AI Is Making Every Small Business Sound Exactly The Same (And How To Fix It).
Want the complete version? Read where I break down AI marketing.
This goes deeper into what an AI consultant actually costs.
