Straight answer: the reason your AI-generated posts sound like everyone else's is that you're feeding the model a generic prompt instead of your own words, and the fix is a written voice document of 800 to 1,500 words built from things you've said, not a tone-of-voice PDF someone else wrote about you. Build it once, test it against five real sentences, and update it every quarter. It takes about three hours and it's the single biggest lever I know for making AI content stop reading like a template.
Worth reading next: Why Your AI Still Writes Like Everyone Else's (And the 45-Minute Fix T.
Worth reading next: The Brand Voice Document That Stops Your AI Content Sounding Generic.
Every AI blog post reads the same now, and you can tell within a sentence
I can spot AI-written LinkedIn posts in about four words. "In today's fast-paced world." "Let's unpack this." Three exclamation marks in a row about a Tuesday. It's not that AI writes badly. It writes competently and identically, because millions of people are typing roughly the same prompt into roughly the same model and getting roughly the same output back.
I know this because I did it myself for about six weeks in 2024, when I was trying to get content out faster while rebuilding my client base after five rough years post-pandemic. Speaking gigs had dried up, my LinkedIn engagement had halved, and I was told by more than one well-meaning person that AI would "solve" my content problem. It didn't. It gave me clean, correct, forgettable posts that could have been written by any consultant with a laptop. My open rates on email dropped from around 38% to 29% in that stretch. Not a coincidence.
What a brand voice document is (and why the ones you've seen don't work)
Most "brand voice guidelines" I've seen from agencies are two pages of adjectives: "friendly, professional, confident, approachable." That's not a voice, that's a horoscope. It could describe almost any business and it gives an AI model nothing to copy from.
A working voice document is different. It's built from real material, not description of yourself. Companies that people say have "a voice" got there the same way: Airbnb's marketing works because every piece of copy sounds like it was written by one person who loves travel, not a committee describing what travel marketing should sound like. Ahrefs does the same thing in B2B SaaS, blunt and specific instead of "leading the industry in innovative solutions." Neither of them started with adjectives. They started with actual sentences someone wrote and kept.
The document I built, step by step
Here's exactly what I put together, over one afternoon, in a plain Google Doc:
- 20 real sentences. I went through eighteen months of my own emails, LinkedIn posts and client calls and pulled out sentences that sounded unmistakably like me. Not my best writing. My most me writing. Things like "that's not a strategy, that's a wish" and "I don't do vague, I do numbers."
- A banned words list. Fifteen words I never say out loud, because they're the words every AI model reaches for by default: "unlock," "," "use," "handle," "," "delve," "unpack," "elevate," "empower." If it sounds like a SaaS homepage, it's on the list.
- Three rewrites. I took three competitor blog posts, the kind that sound like every other agency post, and rewrote each one in my own voice by hand, no AI, just to prove to myself I could still do it without a tool. That rewrite becomes reference material.
- My five most-asked questions, answered in my own words. Not polished answers. The ones I give on discovery calls, verbatim, including the bits where I contradict the "safe" industry answer.
- A one-line opinion on the topic I write about most. Mine is something like: most small businesses don't have a content problem, they have a decision problem, and no amount of AI output fixes indecision.
That whole document runs about 1,100 words. I paste the entire thing into the system prompt or custom instructions before I ask any model to draft anything for me, every single time. Not a summary of it. All of it.
The test that tells you if it's working
Before I trust any new AI tool, I run the same test: I ask it to write two paragraphs on a topic with the voice document pasted in, and two paragraphs on the same topic with no voice document at all. Then I read both out loud. If I can't tell which paragraph came from which prompt, the document isn't specific enough and I go back and pull five more real sentences from old emails. This sounds fussy but it takes eleven minutes and it's saved me from publishing at least a dozen posts that would have sounded like nobody. The interactive tools and calculators I've built for lead generation get the same treatment, because the copy sitting next to a calculator still needs to sound like a person wrote it, not a template generator.
Where this pays off
The obvious place is blog posts and LinkedIn, but the bigger win is quieter: proposals, follow-up emails, sales pages. I had a client, a small accountancy firm in the Midlands, whose follow-up emails after discovery calls converted at about 18%. We built them a voice document from four years of the founder's actual sent emails, fed it into their AI drafting tool, and follow-up conversion moved to 31% within two months. Nothing else changed. Same offer, same price, same calls. Just emails that sounded like the person on the call instead of a template with their name swapped in.
Alex Hormozi built an entire following on saying things bluntly and specifically instead of hedging, and whatever you think of the style, it's the same principle scaled up. I've written before about the business lessons in his approach, and the throughline is that specificity and a clear point of view beat polish every time, whether a human or a model is doing the typing.
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 part nobody selling AI content wants you to hear
Here's the uncomfortable bit. A voice document doesn't fix weak thinking. If your actual point of view on your industry is "we care about our clients" and "quality matters," no amount of tone-matching will make that sound interesting, because there's no opinion underneath it to amplify. AI can copy your rhythm and your word choices. It cannot invent an argument you don't have. I've watched agencies sell "AI content packages" to small businesses that skip this step entirely, because building a real voice document is slow and unglamorous and doesn't demo well in a sales call. It's much easier to sell volume: fifty posts a month, all fluent, all forgettable, all sounding like they came from the same faceless account, because they did. The tool wasn't the problem. The absence of a real opinion was. If you don't have a strong point of view on your own industry yet, that's the actual work, and it comes before any prompt engineering. Away built its entire early brand on one specific opinion about travel gear rather than generic descriptions of quality, which is worth reading about in the Away marketing strategy breakdown if you want to see the same principle applied to a physical product rather than a service.
How often to update it
I revisit mine every quarter, about ninety minutes each time. I add new sentences from whatever I've written recently that felt right, remove anything that now sounds dated, and check the banned word list against whatever new AI-cliché has crept into everyone's LinkedIn posts (this quarter it's "here's the thing" as an opener, used by roughly everyone). A voice document that never changes goes stale the same way a tone-of-voice PDF from 2019 goes stale. It's a living file, not a one-off deliverable.
If you're working with someone on this rather than building it solo, it's worth being clear that this is a specific, narrow piece of work, not the whole of what an AI consultant for a small business should be doing for you. Voice documents are a good first project precisely because you can see whether the output improves within a day, which tells you a lot about whether the wider engagement is going to be worth the money.
Frequently asked questions
How long should a brand voice document be for AI to use it ?
Between 800 and 1,500 words. Shorter than that and there isn't enough real material for the model to copy from, so it defaults back to generic phrasing. Longer than that and most tools start losing the thread of it, especially in a single prompt window.
Can I just ask ChatGPT to describe my brand voice for me?
You can, but it will produce the same adjective list every agency template produces: friendly, confident, approachable. Voice documents built from your own real sentences, emails and answers to client questions work far better than anything a model invents about you from a short conversation.
Does this work with every AI tool or just ChatGPT?
It works with any model you paste it into as instructions, whether that's ChatGPT, Claude, Gemini or a drafting tool built into your CRM. The document is the asset, not the tool. Move it wherever you're writing.
What's the single biggest mistake people make with AI brand voice?
Treating tone as the whole job. Matching someone's rhythm and word choices is the easy 20%. The harder 80% is having an actual opinion worth writing in that voice in the first place, and no document fixes that part for you.
Related reading: Using AI to Chase Unpaid Invoices Without Sounding Like a Debt Collector and AI Tool Costs Explained for Business Owners in 2026.
This overlaps with my main post on the topic: ai brand voice document small business.
