The short version: AI didn't ruin your marketing, laziness with AI ruined your marketing. The fix isn't ditching the tools, it's feeding them your own words first so what comes back sounds like you and not like every other business in your inbox. I'll show you the exact process, including the bit where you have to read your own writing out loud, which nobody enjoys.
The email that made me stop and look
A client of mine, a coach I'll call Rachel because she'd rather not have this story attached to her name, sent me her welcome email sequence in September. Five emails, all written with ChatGPT, all grammatically perfect, all completely lifeless. Open rates had dropped from 41% to 26% in three months. Nothing else had changed. Same list, same offer, same send times.
I read all five emails back to back and realised I couldn't tell you a single thing about Rachel from reading them. No mention of the fact she'd built her business after being made redundant twice. No mention of her actual clients or her actual mistakes. Just clean, competent, forgettable sentences that could have belonged to any coach selling any programme.
That's the real problem with AI content right now, and almost nobody names it. It's not that AI writing is bad. It's that it's median. Every model has been trained on the same enormous pile of internet writing, so left to its own devices it produces the statistical average of how humans write. Average isn't wrong. Average is just invisible.
The uncomfortable bit: detectors are a distraction
Everyone's obsessing over whether Google or their audience can "tell" if something is AI-written, and I think that's the wrong worry entirely. Last month I ran 40 pieces of client LinkedIn content through one of the popular AI detectors, out of curiosity more than anything. Thirty-four came back flagged as "likely AI-generated." Six of those thirty-four were written entirely by hand, by me, on my phone, on a train. The detectors got a sixth of the sample wrong in the direction that matters most.
So the panic about being caught is largely pointless, because the tools people use to do the catching are unreliable. What costs you readers isn't a robot flagging your post. It's a human scrolling past because your sentence structure, your opening line, and your call to action are identical to the four other posts they've already scrolled past that morning. Boredom is the real detector, and it's a lot harder to game.
Why this keeps happening even to smart marketers
Three reasons, from what I see working with small business owners every week:
- They type one line of prompt ("write a LinkedIn post about time management") and accept the first draft because deadlines don't wait.
- They never show the tool a sample of their own writing, so it has nothing to imitate except the average internet voice.
- They edit for grammar, not for voice, so the typos disappear but so does the personality.
I did exactly this myself in early 2024, before I noticed what was happening. I'd ask for a caption, get something clean, post it, and wonder why engagement felt flatter than it used to. It wasn't the algorithm. It was me handing over a blank page and getting back a blank-page answer.
The six-step fix I now use with every client
This takes about 45 minutes the first time you do it, and about ten minutes every time after that.
- Pull ten pieces you wrote before you started using AI. Old emails, old captions, even a long text message to a friend explaining your business. You need real, unedited you.
- Paste those ten samples into the chat as a style reference, and tell the model directly: "Study the sentence length, the rhythm, and the specific phrases in these. Match that voice, don't just summarise the content."
- Ask for three drafts, not one, and throw away the first one automatically. The first draft is almost always the model's default average voice creeping back in.
- Read the second and third drafts out loud. This is the step people skip and it's the one that catches the most. If you stumble over a sentence, a human would too.
- Cut the middle paragraph. Nine times out of ten there's a paragraph in the middle of AI output that restates the opening in slightly different words. It adds length, not value. Delete it without mercy.
- Add one detail only you would know. A client's name (with permission), a specific number, a mistake you made last Tuesday. This single line does more to break the "sounds like everyone else" pattern than any amount of prompt engineering.
Rachel's open rate went from 26% back to 38% within two send cycles after we ran her sequence through this. Nothing else changed. Same subject lines, same send day.
What brand voice looks like when it's done right
The businesses that never seem to have this problem are the ones with a voice so specific it would be strange coming from anyone else. Look at the GoPro marketing strategy, built entirely around footage shot by actual customers rather than polished brand copy. Or the way Figma's marketing leans hard into the language of designers and product teams specifically, which you can see picked apart in this look at the Figma marketing strategy. Neither of those brands could be swapped out for a competitor without the copy feeling wrong. That's the test. If you deleted your business name from your last five posts, would a regular reader still know it was you?
Mailchimp is a good example of the opposite lesson worth studying too, because their whole brand voice was built on being oddly playful and human in an industry (email software) that's usually dry as toast. The Mailchimp marketing strategy shows what happens when a company decides its personality is a product feature, not an afterthought bolted onto the marketing at the end.
Instagram's own growth is another useful case study here, less about tone and more about format discipline, which you can see in this breakdown of the Instagram marketing strategy. The lesson that carries over to AI content specifically: consistency of format builds trust, but sameness of voice kills it. Those are different things and people confuse them constantly.
The format trick nobody talks about
There's a second fix that has nothing to do with prompting, and it's one I wish more people used: change the format entirely, not just the words. AI is very good at producing paragraphs and lists. It is much worse at producing interactive experiences, because there's no single "average" version of one to imitate. A calculator that tells a visitor exactly what their specific numbers mean can't be homogenised the way a blog post can, because the output depends entirely on what the user types in. I wrote a longer piece on this exact idea, how to engage and convert more visitors with interactive calculators, and it's still one of the most underused tactics I see in small business marketing. If your competitors are all publishing the same five AI-flavoured blog posts about "how to price your services," a two-minute pricing calculator on your site does something none of them can copy in an afternoon.
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.
Content discipline still beats content volume
None of this works if you're publishing purely for the sake of publishing. Joe Pulizzi built the entire modern content marketing industry on one stubborn idea: fewer, better pieces aimed at one clear audience beat a flood of mediocre ones, every time. I go through several of his sharpest points in these business lessons from Joe Pulizzi, and the one that applies directly here is his insistence on knowing exactly who you're talking to before you write a word. AI makes it very easy to produce ten posts in the time it used to take to produce one. That's only useful if all ten are still aimed at a real, specific person. Otherwise you've just built ten forgettable posts instead of one.
What I'd tell you to do this week
Don't stop using AI. That's not the lesson here and anyone telling you to go back to writing everything from a blank page by hand is giving you advice that doesn't scale for a small business owner already stretched thin. Instead, spend the 45 minutes this week building your style reference document. Save it somewhere you can paste into a new chat instantly. Use the six steps on your next three pieces of content and compare the engagement to your last three AI-written pieces without them. The difference tends to show up faster than people expect, usually within one or two posts, because your audience has been unconsciously scrolling past sameness for months without quite knowing why.
The businesses winning with AI right now aren't the ones avoiding it. They're the ones who worked out that the tool is only as interesting as what you feed it, and stopped handing it a blank page.
Frequently asked questions
Can readers tell when content is AI-written?
Not reliably, and neither can most detection tools, which flag plenty of human writing as AI and miss plenty of AI writing entirely. What readers can tell, without any detector, is when content is boring and interchangeable. That's the thing to fix, not the source.
How long does it take to fix an AI content voice problem?
For most small businesses, building a proper style reference from old writing and running it through the six-step process takes about 45 minutes the first time. After that, applying it to new content takes roughly ten minutes per piece, which is a small cost against the engagement you get back.
Should I stop using AI for marketing content altogether?
No. The businesses seeing the worst results are usually the ones using AI carelessly, with one-line prompts and no editing for voice, not the ones using it at all. Feed it real samples of your own writing and edit specifically for personality, not just grammar, and the tool becomes far more useful.
What's a quick way to test if my content sounds too generic?
Take your last five posts or emails and mentally delete your business name and any obvious brand references. If a regular reader wouldn't be able to tell it was written by you specifically, rather than any competitor in your space, that's the sign your voice has flattened out and needs the fix.
More on this here: AI automation for coaches and consultants.
Related reading: Why Every AI-Written Newsletter Sounds the Same (And What I Changed In Mine) and I Built an AI-Powered Lead Calculator in a Weekend: Here's What Six Weeks of Data Showed.
