The short version: AI content sounds the same across every small business because everyone is feeding ChatGPT the same generic prompt with no real information about how they talk. Build a one-page "voice document" with your banned words, your real sentence patterns, and three examples of you writing badly in your own words, and the sameness mostly disappears. It takes about two hours to build once and saves hours on every single piece of content after that.
Worth reading next: Why "Write Like Me" AI Prompts Fail for Small Business Content (And Wh.
The email that made me stop and check something
Back in March, a client sent me a LinkedIn post she was proud of. I read the first two lines and knew, before she even told me, that it had come out of ChatGPT with no editing. Not because it was bad. Because I'd read the same sentence structure, the same "In today's fast-paced world" opener, the same three-point list with a bolded takeaway at the end, on about eleven other business pages that same week.
I went and checked. I pulled ten LinkedIn posts from ten different small business owners in completely different industries, a plumber, a divorce coach, a SaaS founder, a bakery owner. Seven of the ten had the exact same rhythm: short hook, rhetorical question, three bullet points, one-line call to action. Different words, identical skeleton. That's when it clicked that this isn't a writing problem. It's a default settings problem.
Why this happens, and it isn't the AI's fault
Here's the bit most people writing about AI content skip, because it makes the tool look worse than they want to admit: ChatGPT and Claude and Gemini are trained to produce the statistically most likely next sentence. If you give them a thin prompt, "write a LinkedIn post about time management for small business owners," they have almost nothing to work with except the average of every time-management post ever written. So they give you the average. That's not a flaw. That is exactly what the tool is built to do when you don't feed it anything specific.
The uncomfortable part is that most business owners have never written down what makes their own voice different. Not their brand colours, not their mission statement, their actual writing habits. Do you use short sentences? Do you swear a little? Do you start paragraphs with "and" or "but"? Do you hate the word "use" the way I do? Nobody sits down and documents that, so when they ask an AI to sound like them, the AI has nothing to go on and defaults to LinkedIn-generic.
What I built instead: the one-page voice document
After that LinkedIn moment, I built a single-page document for my own content and started asking clients to build one for theirs. It's not a brand guideline. It's shorter and much more specific. Here's exactly what goes in it, in order:
- Ten banned words or phrases. Mine includes use, delve, handle, unlock, and the phrase "in today's world." I never want to see them in a first draft.
- Three real sentences you wrote, pulled from an old email or post, pasted in exactly as they were. Not polished. This gives the model something real to pattern-match against instead of a description of your style.
- Your sentence length habit. I write short. Full stops instead of commas. I tell the tool that directly: "short sentences, one idea per sentence, no more than one semicolon per 500 words."
- One thing you disagree with in your industry. This is the one that changes everything. Generic content has no opinion. Mine is that most "AI for business" advice treats the tool like magic instead of a slightly clever intern who needs supervision. Give the model your actual opinion and it stops writing neutral mush.
- A story you're allowed to reuse. One real example from your business, with a number in it if you have one. AI can't invent your lived experience, so give it one to draw on.
That's it. Five sections, one page, built once. I keep mine in a text file and paste the whole thing at the top of every content prompt now.
What changed when I used it
I tracked this across 40 blog posts and LinkedIn posts over about ten weeks, because I wanted a real number, not a feeling. Before the voice document, my average editing time per AI-assisted draft was around 45 minutes, mostly stripping out the generic openers, deleting bullet points I hadn't asked for, and rewriting the ending because it always tried to sound inspirational. After the voice document, that dropped to roughly 10 to 15 minutes per draft. Same tool, same subscription, same me. The only variable was giving it real material to copy instead of a vague instruction.
The bigger shift was qualitative. A repeat client told me, unprompted, that a recent post " sounded like you were annoyed about something," which was the exact reaction I wanted, because I was. That's the test I use now. If a piece of AI-assisted writing could have been posted by any business in any industry, the voice document step got skipped.
The contrarian bit nobody selling AI courses wants to say
Most AI training for small businesses focuses on prompts. Better prompts, longer prompts, "expert-level" prompt frameworks with 15 variables. I've bought several of these. Most of it is a distraction from the actual bottleneck, which is that the business owner has never sat down and been honest about what they think, what annoys them, and what they'd say if nobody was watching. A perfect prompt fed no real opinion still produces generic output, just more confidently generic. You can't prompt your way around having nothing specific to say. The voice document forces the specificity to happen before the prompt, not inside it.
This is also why brands that clearly know who they are barely have this problem. Look at how deliberately Away built its marketing strategy around one consistent, opinionated voice about travel, or how Disney's marketing has stayed recognisable for a century because the tone was documented and protected long before AI existed. Those brands never sound like a template because someone, somewhere, wrote down exactly what they do and don't sound like. Small businesses can copy that discipline in an afternoon instead of a century.
How to build yours in one sitting
Block out two hours, not more. Here's the order that worked for me and for the six clients I've since run through this:
- Step 1: Pull three real pieces you wrote. An email, a LinkedIn comment, a text to a client. Anything unedited and yours.
- Step 2: Highlight the sentences you'd never delete. The ones that sound like you, not like a brochure. Paste them into your voice document verbatim.
- Step 3: Write your ten banned words. If you're stuck, open any AI-written draft you've published and circle every word that made you wince slightly. Those are your ten.
- Step 4: Write one opinion you hold that most people in your field would disagree with. Not a hot take for the sake of it, something you'd defend at a dinner table.
- Step 5: Save one real story with a number in it. A client result, a mistake that cost you money, a specific week that went badly. Numbers make it usable later.
- Step 6: Paste the whole document at the top of every content prompt, every time, before the actual request. Not as a one-off "remember this," as a permanent header.
If you use a tool that lets you save custom instructions or a project-level system prompt, put the voice document there permanently so you're not re-pasting it each time. If you don't, a saved text file you copy in takes about eight seconds and is still faster than the editing time you'll save.
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.
Where this fits with the rest of your marketing
None of this replaces having an actual point of view about your business, which is the harder work. It's the same discipline that made Ahrefs' marketing recognisable in a crowded SEO space, they picked a specific, slightly blunt tone and kept it everywhere. If you're building content strategy from scratch and want the tool doing more than drafting, a proper look at working with an AI implementation coach is worth it once you've hit the limit of what a voice document alone can fix, particularly if you're trying to roll this out across a team rather than just for yourself.
There's a Napoleon Hill line about definiteness of purpose that I think about more than I expected to when I first read the business lessons pulled from his work, the idea that vague intentions produce vague results. AI content is the fastest possible proof of that principle. A vague prompt produces a vague post. A specific voice document, built from real sentences and a real opinion, produces something a client will quote back to you.
And if you want a smaller-scale test before building the whole document, try it on something low-stakes first, an FAQ page, a set of email replies, even the copy around an interactive calculator if you've got one on your site. You'll notice the sameness problem fastest on short, functional copy because there's nowhere for useful specifics to hide.
What I'd tell you if you only read one paragraph of this
Stop tweaking prompts and start writing down what makes you sound like you, once, on one page, then feed it to the AI every single time before you ask for anything. It costs two hours upfront. It saves you 30 minutes on every piece of content after that, which pays for itself by the fourth post. The businesses whose AI content still sounds distinct in 2026 aren't using a smarter model. They're the ones who did the boring documentation work everyone else skipped because it felt like homework instead of a hack.
Frequently asked questions
Does a voice document work with any AI tool, or only ChatGPT?
It works with any large language model, ChatGPT, Claude, Gemini, or Copilot, because the underlying issue is the same across all of them: thin prompts produce generic output. Paste the same voice document at the start of a prompt in any of these tools and you'll see a similar improvement, though Claude in particular tends to hold onto tone instructions more consistently across a long conversation.
How long should a voice document be?
One page. If it runs longer than that, you're writing brand guidelines, not a voice document, and most models will start ignoring parts of it. Keep it to the five sections: banned words, three real sentences, sentence length habit, one opinion, one reusable story.
What's the single biggest mistake people make when they try this?
They write a description of their tone instead of pasting real writing. "I sound friendly but professional" tells the AI nothing useful. Three actual sentences you wrote, pasted exactly as they were, teach it far more than any adjective could.
Do I need to update the voice document over time?
Refresh it every few months or whenever you notice your drafts drifting generic again. Swap in newer real sentences, update the banned words list if a phrase has crept back in, and add a fresh story once your old one stops feeling current.
For the bigger picture, see my full guide to AI marketing.
Want this done for you? See AI automation for coaches and consultants.
Related reading: Shadow AI: What Your Team Is Doing With ChatGPT Behind Your Back and How to Build a Personal Brand as a Consultant in 2026.
