Straight answer: Your AI content sounds like everyone else's because you're feeding the tool the same three prompts every other small business owner is using. The fix isn't a better prompt, it's a voice bank, a document made of your real words that the AI reads before it writes anything for you. Build one and the difference shows up in the comments section within a fortnight.
The sameness problem is worse than anyone's saying out loud
Scroll LinkedIn for ten minutes right now. Count how many posts start with "I used to think X, then I learned Y" or end with a one-line paragraph that reads like a fortune cookie. That's not a coincidence and it's not you imagining it. It's thousands of small businesses typing "write me a LinkedIn post about [topic], make it engaging and professional" into the same handful of tools and getting back the statistical average of every confident-sounding business post ever written.
I went through this myself, and it stung because I used to be good at this. Between 2015 and 2019 I built a personal brand that got me onto the Forbes list, a following built almost entirely on writing the way I talk, opinionated, a bit blunt, British. My average LinkedIn post pulled 60 to 80 comments. By 2023, after I'd leaned hard on AI to speed up my content because I was rebuilding the business after a rough few years, that number had dropped to 3 or 4 comments a post. Same topics. Same expertise. Flat results.
I didn't clock it straight away because the posts read fine on their own. It was only when I put six months of them side by side that I saw it: every post had the same rhythm, the same three-word punchy closing line, the same "here's the thing" transition. I sounded like a business coach template. Not like me.
Why "just add personality to the prompt" doesn't work
The obvious fix everyone reaches for is prompting harder. Add "write in a warm, conversational British tone" or "be blunt and opinionated" to the prompt. I tried this for weeks. It helps a little, the same way telling someone to "be funnier" helps a little. What it can't do is give the model information it doesn't have, which is what you sound like when you're annoyed, or excited, or explaining something to a mate down the pub rather than to a client.
Here's the uncomfortable bit most AI content advice skips over: a prompt describes a voice, it doesn't demonstrate one. You can tell a model to be "direct and specific" until you're blue in the face and it will still default to hedging and generic phrasing, because "direct and specific" is an instruction, and instructions get diluted the moment the model has to invent examples of what that means. What the model needs is your actual raw material, real sentences you've written or said, so it has something to copy the pattern of rather than something to guess at.
What a voice bank is
A voice bank is a single document, mine runs to about 4,000 words, made up entirely of things you've written or said, pasted in raw with no editing. Not a description of your tone. Not a brand style guide with adjectives like "friendly but authoritative." The real stuff itself. You paste this document (or a trimmed version of it) into the AI tool before you ask it to write anything, or you build it into a custom instructions field if your tool has one, and the model uses it as its reference point instead of the internet's average business voice.
I built mine over one Saturday afternoon. It took about three hours and I pulled from:
- 15 old LinkedIn posts from my 2016 to 2019 period, back when I wasn't leaning on any tool
- 40 client emails where I was explaining something, not selling anything
- Transcripts from six recorded client calls (I use a transcription tool for calls anyway, so this was just copying and pasting)
- A ten minute voice memo I recorded rambling about why most AI content advice is rubbish, transcribed and pasted in exactly as spoken, including the "and another thing" bits
- Three complaint emails I'd sent to suppliers, because apparently I'm at my most "me" when I'm annoyed about something
That last one surprised me the most. My sharpest, most distinctive writing wasn't in my polished content at all, it was in the emails I fired off without thinking about how I sounded. That's usually where your real voice lives, in the writing you did fast, for one person, with no audience in mind.
How to build your own, step by step
This works whether you're a one-person consultancy or you've got a small marketing team writing on behalf of the business.
- Step 1: Gather 15 to 20 pieces of unedited writing. Old emails, Slack messages to your team, WhatsApp voice notes transcribed, a rant you posted in a Facebook group. The rougher and more unfiltered, the better. Skip anything that was already written with heavy help from AI, that defeats the point.
- Step 2: Add one piece of spoken content. Record yourself explaining your business to a friend who knows nothing about it, five minutes is enough. Transcribe it with any transcription tool and paste it in exactly as spoken, filler words and all. Written and spoken voice differ, and the gap between them is often where your most natural phrasing hides.
- Step 3: Paste it all into one document with no cleanup. Don't summarise it, don't tidy the grammar, don't add headings that say "here is an example of my tone." Raw text only.
- Step 4: Write one paragraph of instructions at the top, but keep it short: "The following is a collection of my own writing and speech. When you write anything for me, match the sentence length, the level of directness, and the vocabulary you see here. Do not use any phrasing that doesn't appear in this document or something close to it."
- Step 5: Test it on something low stakes first. An internal email, not your homepage. Compare the output to what the same prompt gives you without the voice bank attached. The difference is usually obvious within one paragraph.
- Step 6: Update it every few months. Add new emails, new posts, new calls. A voice bank from 2024 will start to feel slightly off by late 2026 because your business and your language move on, so treat it as a living document, not a one-off project.
What changed when I used mine
My comment counts didn't jump back to 2019 numbers overnight, that would be a lie and I'm not going to sell you one. But within about six weeks of writing every LinkedIn post with my voice bank pasted in first, my average comments went from the low single figures back up to 15 to 20 per post, and two posts crossed 40 comments, the first time that had happened since 2020. The content wasn't objectively "better" in a way I could measure with any metric other than comments and shares, but it read like me again, and readers noticed even when they couldn't have told you why.
The other thing that changed, and this is the part nobody wants to hear, is that it didn't save me time. Building and maintaining a voice bank is extra work on top of the AI writing, not instead of it. If you're chasing AI content purely to cut hours, a voice bank will feel like a step backwards at first, because you're adding a documentation task to a process that was supposed to make your life easier. I still think it's worth it, but go in knowing that the "AI will save you time" promise and the "AI will sound like you" promise pull in slightly different directions, and most people selling AI content tools won't tell you that because it complicates the pitch.
Where this matters beyond LinkedIn posts
A voice bank isn't just for social content. I use the same document as the reference point for client proposal emails, for the "about" section on my site, and for any written case study I put together, the kind of narrative-driven brand storytelling I dig into when I write about how brands like Away built theirs around a consistent, recognisable voice, or how Ahrefs built their content strategy around a very specific, opinionated tone that readers can spot without seeing the logo. Consistency of voice is a competitive advantage precisely because most businesses using AI right now are producing content with no voice bank at all, which means their output blends into the background. Yours doesn't have to.
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.
If you're building an actual content team or scaling this across a small marketing department, the same principle applies at a bigger scale, which is where a lot of my consultancy work happens now. I've sat with small business owners who've spent months trying to prompt-engineer their way to a consistent brand voice across five team members using AI tools, when the real fix was one shared voice bank document that every team member's tool references before writing anything. It's the kind of unglamorous groundwork an AI implementation coach should be setting up in week one, not something left until the content already sounds like a committee wrote it.
The parts people get wrong
Three mistakes I see constantly when people try this after hearing about it:
- Using content you already ran through AI as the source material. This creates a feedback loop where the AI is learning to sound like its own previous output, which is how you end up further from your real voice, not closer to it.
- Making it too polished. If every example in your voice bank is a finished, edited blog post, the model learns your edited voice, not your natural one. Include drafts, rough emails, the version you sent before your assistant tidied it up.
- Treating it as a one-time fix. I've had clients build a beautiful voice bank, use it for two weeks, then quietly drift back to bare prompts because it's an extra step. The businesses that keep the edge are the ones who make attaching the voice bank a habit, the same way you'd never send a client proposal without checking the figures first.
There's a broader lesson in here too, one that echoes something I come back to often when I write about persistence and specificity in Napoleon Hill's business writing, or the deliberate, unglamorous consistency behind how Howard Schultz built Starbucks' culture one repeated detail at a time. The businesses that stand out rarely do one dramatic thing differently. They do one boring thing consistently that everyone else can't be bothered to keep up. A voice bank is boring. That's exactly why it works.
What I'd tell a small business owner starting this week
Don't wait for the perfect document. Pull 15 pieces of your own writing today, however messy, paste them into one file, add the one-paragraph instruction, and use it on your next email newsletter. You'll know within one draft whether it's working, because the output will stop sounding like it could have been written by your biggest competitor. That's the whole test. Not "does this sound professional," professional is easy and everyone's AI output is professional now. The test is whether a client who's spoken to you on the phone would recognise your writing without seeing your name on it.
Frequently asked questions
How long should a voice bank document be?
Mine runs to roughly 4,000 words, but there's no fixed rule. Aim for 15 to 20 separate pieces of your real writing or speech, enough variety that the model sees your voice across different situations, an email, a social post, a rant, a proposal. Quality and range matter more than total word count.
Can I use content my team has already written with AI help as source material?
Avoid it where you can. If the source material was already generated or heavily edited by AI, you're teaching the model to imitate its own average output rather than a real human voice. Use writing from before you started using AI tools, or writing done quickly under time pressure, which tends to be more natural anyway.
Does a voice bank work for a whole team, not just one person?
Yes, though the goal shifts slightly. Instead of capturing one individual's exact voice, you're capturing the shared tone of the brand, pulled from your best writers' work across emails, proposals, and social posts. Build it once, store it somewhere every team member can access, and treat updating it as part of your regular content process rather than a one-off task.
Will a voice bank save me time on content, since that's usually the point of using AI?
Not at first, and it's worth being honest about that. Building and maintaining the document is extra work layered on top of your AI writing process. What you gain isn't speed, it's content that sounds like you instead of a generic average, which tends to matter more for engagement, trust and client relationships than shaving a few minutes off a draft.
Related reading: How to Build an AI Content Calendar That Gets Used (Not Just Built) and How to prepare your startup business for growth.
