The short version: Your AI-written marketing doesn't sound bad because it's AI-written. It sounds bad because you're feeding the tool the same generic prompt as every other small business owner, and it's handing all of you the same 200 phrases back. The fix isn't writing everything yourself again. It's feeding the machine specifics only you have.
The week I read 40 client LinkedIn posts and wanted to lie down
Last autumn I did an audit for a group coaching client, 40 small business owners, all posting on LinkedIn at least three times a week. I read every post from a two week window. Twenty seven of them used the phrase "game changer." Nineteen started with "In today's fast paced world." Eleven had a sentence that began "As a business owner, I know how important it is to..." and then finished with something so vague it could apply to a dentist, a dog groomer, or a divorce lawyer.
None of these people knew each other. They weren't copying one another. They were all typing some version of "write me a LinkedIn post about the importance of customer service" into ChatGPT and posting the first thing it gave back. The model wasn't the problem. The prompt was.
Here's the uncomfortable bit nobody wants to say out loud at marketing conferences: banning AI or slapping an "AI detector" badge of honour on your content strategy isn't going to save you. Detectors are unreliable, easy to fool, and increasingly beside the point, because your customers were never checking for AI in the first place. What they're checking for, without even realising it, is whether you sound like you or like everyone else. That's the actual battle, and it has nothing to do with whether a large language model touched the sentence.
I've watched clients spend hours agonising over whether to disclose that ChatGPT helped draft an email, while sending out a newsletter so bland it could have come from a template store. Wrong worry, wrong energy.
Why the sameness happens (it's not what most people think)
AI models generate the statistically most likely next word. That means left to its own devices, with a vague prompt, it reaches for the safest, most common phrasing available, because that's what "likely" means in training data terms. Ask a thousand business owners to describe their customer service and you'll get a thousand versions of "we go above and beyond," because that phrase appears in millions of training examples and almost no specific ones.
The fix is blunt: stop asking AI to invent your business. Start feeding it facts about your business that only you have, and make the tool arrange those facts. Specificity is the only thing a language model can't hallucinate its way around convincingly, because specificity has to come from you.
What differentiates brands (and it's rarely the writing style)
Look at any brand people can't stop talking about and you'll find the differentiation sits in the offer or the mechanism, not the prose. I've picked apart the Airbnb marketing strategy more times than I can count with clients, and the thing that made it work wasn't clever copywriting, it was "live like a local," a specific, ownable idea that every piece of content could hang off. Same with the Away marketing strategy: a suitcase brand built around one repeatable story (the founder couldn't find a decent case for a work trip, so she made one), told the same way, in every channel, for years. And when I break down the Ahrefs marketing strategy for clients, the lesson is almost always the same one: they publish their own data, their own numbers, their own screenshots. Nobody else has those. AI can't invent them either.
That's the actual unlock (sorry, wrong word for this brand, let's say "the actual point"): differentiation lives in facts, numbers, and stories that are yours. AI is brilliant at arranging those into clean sentences. It is useless at inventing them from nothing, and every attempt to make it try produces exactly the sludge those 40 LinkedIn posts were drowning in.
The five-step process I use with clients now
This is the process I now run through with every small business client who's frustrated that their AI content feels flat. It takes about 45 minutes the first time, then five minutes per piece after that.
- Step 1: Build a facts file, not a brand voice document. Skip the "our tone is friendly yet professional" nonsense. Write down 20 real facts: your actual prices, a client's actual result with a number attached, the thing a customer said to you last week, the mistake you made in year one, the number of hours a job takes. Facts, not adjectives.
- Step 2: Feed AI three of your own past pieces as examples before you ask for anything new. Paste in three emails, posts, or pages you wrote yourself, even rough ones, and tell the model "match this rhythm and word choice, not a generic professional tone." Few-shot examples work far better than tone instructions.
- Step 3: Ban the sludge words in your custom instructions. In ChatGPT or Claude's custom instructions field, add a permanent line: "Never use game changer, in today's fast paced world, unlock, delve, or." You'll be shocked how much cleaner every output gets immediately.
- Step 4: Ask for a first draft, not a final one. Tell it explicitly: "This is a rough first draft for me to rewrite, not a finished piece, so don't try to sound polished." Oddly, asking for rough gets you something more usable than asking for polished.
- Step 5: Add one true, weird, specific detail by hand before you publish. Not a stat you Googled, something only you know: the client's name (with permission), the exact time of day it happened, the number that's slightly embarrassing. This is the single edit that moves a post from "written by a robot" to "written by a person," and it takes 90 seconds.
None of this requires you to stop using AI, write everything longhand, or feel guilty about a tool that saves you hours. It requires you to stop treating the tool as the author and start treating it as the typist.
A small example that made the difference obvious
One client, a bookkeeper working with 30 small business clients, had ChatGPT writing her weekly email tips. Generic version: "Cash flow is the lifeblood of your business. Make sure you're tracking it regularly to avoid surprises." Fine. Forgettable. Same email 10,000 other bookkeepers could have sent.
We ran it through the process above. Her facts file had this line: "Three clients this year missed their VAT payment because they forgot it was quarterly, not monthly." New version: "Three of my clients got caught out by VAT this year, all for the same reason: they thought it was monthly. It's quarterly. If you don't know your next VAT date off the top of your head right now, go check. I'll wait." Same core message. Completely different email. Open rates on that specific send went up noticeably compared to her usual average, and two people replied asking if she could check their VAT dates for them, which turned into paid work. That's the difference specificity makes, and AI wrote most of the sentence structure. She just refused to let it invent the substance.
If you'd rather have someone build that whole system with you instead of piecing it together from blog posts, that's the kind of work I do inside a short AI consultant engagement for small business, mapping your actual facts, offers and past writing into prompts and templates you can run yourself afterward.
Where this connects back to your offer, not just your writing
Alex Hormozi's whole body of work, and I go through the business lessons from Alex Hormozi regularly with clients, comes back to one idea: make the offer so specific and so good that the marketing barely has to work hard. A generic offer needs generic-sounding, AI-flattened copy to disguise how ordinary it is. A specific, sharp offer, a specific number, a specific guarantee, a specific timeframe, gives AI something real to write about, and suddenly the sameness problem mostly disappears on its own.
This also applies beyond written copy. If you're building anything interactive on your site, a pricing calculator, a quiz, a quote tool, the same rule holds: the tool needs your real numbers and your real logic, not a generic template. I've written before about how interactive calculators convert more visitors precisely because they force specificity, a visitor puts in their exact numbers and gets an exact answer, and that's the opposite of the AI-sludge problem. It's specific by design.
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 bit people don't want to hear
If your marketing has felt flat for the last year, it is very unlikely to be because you used AI. It's much more likely because you never had anything specific to say in the first place, and AI simply made that gap visible faster and at higher volume. A vague business owner with AI produces vague content at scale. A specific business owner with AI produces specific content at scale. The tool amplifies whatever was already true about the input. That's not a flaw in the technology. It's the whole point of it, and it's worth sitting with rather than blaming the software.
I'll be honest about my own habits here too, because it feels dishonest not to. I use AI for first drafts most days now. I don't announce it, and I don't apologise for it, because the sentences you're reading went through several rounds of me cutting, adding facts, and killing phrases I'd banned in my own custom instructions. The work is in the specifics I put in and the sludge I take out, not in typing every word myself for the sake of it. That's the standard worth holding yourself to, not "did a machine touch this."
Free resource: The Marketing Automation Map Template.
Frequently asked questions
Does it matter if customers can tell my content was written with AI?
Almost never, in my experience. What customers notice and react to is whether the content is specific to your business or could belong to any competitor. Generic content written entirely by hand reads just as badly as generic content written entirely by AI. Specificity, not authorship, is what customers respond to.
What's the fastest way to make my existing AI content less generic?
Open your last five AI-written posts and search for phrases like "game changer," "in today's world," or " experience." Delete the sentence around each one and replace it with one real fact, number, or story from your own business. That single edit does more than any prompt rewrite.
Should I stop using AI for marketing content altogether?
No. That throws away real time savings for no real benefit. The problem was never the tool, it was vague prompts producing vague output. Feed it your own facts, past writing, and banned phrases, and the sameness problem largely disappears while you keep the speed.
How do I know if my brand voice sounds different from competitors, or if I'm just assuming it does?
Print out your last three pieces of marketing content and three from a direct competitor, take the names off all six, and read them cold. If you can't tell which is which within a sentence or two, you have a sameness problem regardless of who or what wrote them.
Want this done for you? See AI automation for coaches and consultants.
Related reading: Using AI to Chase Unpaid Invoices Without Sounding Like a Debt Collector and AI Tool Costs Explained for Business Owners in 2026.
I go much deeper on this in the AI marketing guide.
