The short version: Feed a private AI project (a custom GPT, a Claude Project, or a Gemini Gem) your own past emails, proposals, blog posts, and a list of words you never use, and it will write like you rather than like every other business owner using the same three generic prompts. This takes about an afternoon to set up and it's the single biggest difference between AI content that sounds like slop and AI content that sounds like a person.
Worth reading next: I Let AI Answer My Business Calls for Six Weeks: What It Really Cost M.
Why your AI content sounds like everyone else's
I can spot AI-written LinkedIn posts from three words in. So can you, if you're honest. The dash-heavy rhythm, the "in today's fast-paced world" opener, the three-item list that ends in a semi-colon and a shrug. Everyone using ChatGPT with a bare prompt ends up with the same voice because they're pulling from the same training data and the same handful of viral prompt templates that got shared 40,000 times on LinkedIn last year.
Here's the bit nobody quite says out loud. The problem isn't the AI. The problem is that most business owners type "write me a LinkedIn post about productivity" and hit send, then wonder why it sounds like a stranger wrote it. Of course it sounds like a stranger. You gave it nothing to work with except the instruction. You didn't give it you.
I made this exact mistake for about eight months. I'd ask ChatGPT for a blog draft, get something perfectly competent and completely characterless, then spend forty minutes rewriting it to sound like a human being who's run a business for twenty years. Eventually I realised I was doing the AI's job for it every single time, and there was a better way.
What a "brand brain" is
A brand brain is a private AI project, inside ChatGPT, Claude, or Gemini, that has permanent access to a folder of your own material. You're not prompting from scratch every time. You've already told it, once, who you are, how you write, what you've said before, and what you'd never say. Then every new request builds on that foundation instead of starting cold.
Practically, this means one of three things depending on which tool you use:
- A Custom GPT in ChatGPT Plus or Team (£20 a month per seat), with a knowledge base of uploaded files
- A Claude Project (available on the £18 a month Pro plan), with project knowledge documents attached
- A Gemini Gem inside Google's Gemini Advanced plan, with the same idea, uploaded reference files it checks before answering
They all work the same way underneath. You upload documents, you write instructions once, and every conversation inside that project reads from the same well.
The exact documents I fed mine
When I built mine, I didn't overthink it. I gave it twelve documents on the first pass:
- Five blog posts I'd written entirely myself, no AI involved, so it had unfiltered examples of my actual sentence rhythm
- Three client emails where I'd said no to something, because how you decline work reveals more voice than how you pitch it
- A one-page "banned words" list, mine includes, use, delve, and no em-dashes, ever
- A short bio with real facts: 53, based in the UK, Forbes-listed, worked with brands you'd recognise, rebuilding publicly after five rough years
- Two case studies I'd written about brands I admire, including one on how Away built its brand voice around community and travel storytelling, because studying someone else's distinct voice sharpens your sense of your own
That's it. Twelve documents, roughly 9,000 words total, uploaded in one sitting. I didn't write a novel of instructions either. My custom instructions were four short paragraphs: who I am, who I'm writing for, what I never do (no fluff openers, no em-dashes, no corporate softening), and what "good" looks like with two examples pasted in.
The step-by-step build
If you want to do this this week, here's the actual sequence, not the vague version:
- Pick one tool and commit to it. Don't split your material across three platforms. Pick whichever you already pay for.
- Gather ten to fifteen documents that are unmistakably yours. Old emails, your best blog posts, a transcript of a talk you gave, your actual proposals. Not AI-generated content from before, that just teaches it to sound like AI.
- Write your banned words list first. Five minutes, one page. Mine has about fifteen words on it now, built up from every time I've winced reading back an AI draft.
- Write four short paragraphs of instructions. Who you are, who you're writing for, what you never do, one pasted example of your voice done right.
- Upload everything and test it with something small. Ask it to write one LinkedIn post before you trust it with a client deliverable.
- Correct it out loud. When it gets something wrong, tell it exactly why, "too formal here" or "I'd never say unlock", and it holds that correction for future chats in that same project.
- Revisit the knowledge base every couple of months. Swap out two or three older documents for fresher ones so it doesn't calcify around an old version of you.
Total time for the first build: about 45 minutes if your documents are already sitting in a folder somewhere, closer to two hours if you have to go digging for old emails and past posts.
What changed once I did this
A client of mine, a bookkeeper running a five-person practice in Leeds, built her version of this in one lunch break after we worked through it together. She fed it eight client newsletters she'd written over two years and a page of her actual objections to jargon ("I will never use the word synergy, ever"). Within a week she told me the drafts had gone from "something I have to rewrite completely" to "something I tweak for ten minutes." That's the real return here. Not magic, just a lot less rewriting.
For me, the change showed up in a boring but telling number: my average editing time on an AI-assisted first draft dropped from about 40 minutes to under 10. Not because the AI got smarter overnight, but because it finally had something to copy that wasn't a stranger's LinkedIn thread.
The uncomfortable truth about AI sameness
Here's the part most people writing about AI and small business skip past, because it's a bit awkward. The sameness epidemic in AI content isn't really AI's fault. It's laziness dressed up as efficiency. Business owners want the speed of AI without spending the one afternoon it takes to feed it their own voice, and then they're baffled that their content reads exactly like their competitor's, because it was generated the exact same way, from the exact same blank prompt.
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.
I've sat in enough workshops to know this isn't a knowledge gap. Most people I train know they could do this. They just haven't, because it feels like admin rather than marketing, and admin always loses to the urgent thing. But the businesses that will stand out over the next few years won't be the ones with access to better AI models, everyone will have access to roughly the same models. They'll be the ones who bothered to teach their AI who they are.
There's a useful parallel in how strong brands build distinct voices in the first place. When you look at how Ahrefs built a brand voice around blunt, data-first honesty or how Disney has protected a consistent emotional tone across nearly a century of output, the common thread is documentation. Someone, somewhere, wrote down the rules of the voice so it could be repeated by other people. A brand brain is that same document, just fed to a machine instead of a new hire.
Where this goes wrong
I've watched three mistakes repeat themselves with clients trying this:
- Uploading too much, too vaguely. Fifty documents with no clear instructions confuses the model more than it helps. Ten sharp examples beat fifty average ones.
- Never correcting it. If you get a bad draft and just fix it yourself without telling the AI what was wrong, it never learns the correction for next time. Say the correction out loud, in the chat, every time.
- Treating it as finished after one build. Your voice shifts as your business shifts. I've rebuilt my own brand brain twice in the past year because the business itself changed, more consulting, less influencer-style content, and the old documents no longer matched who I am now.
There's also a useful lesson in restraint here, the kind of long-view thinking you find in older business writing that's aged surprisingly well. Some of the business lessons from Napoleon Hill and the lessons from Howard Schultz on building something with a consistent identity apply here almost word for word: define the thing clearly once, then protect it fiercely, rather than reinventing it every single time you sit down to write.
How much this costs
The honest number is somewhere between £18 and £20 a month, whichever paid AI plan you already have, plus your own time. There's no extra tool to buy, no developer needed, no six-week onboarding. If you'd rather have someone build the whole system alongside you, including the knowledge base, the correction workflow, and training your team to use it consistently, that's the kind of work an AI consultant for small business would set up with you directly, usually inside a single working session rather than a long project.
One more practical note before you try this yourself. If your business involves interactive content, quote forms, or anything where a visitor answers questions and gets a personalised result, that same brand brain approach works beautifully for the copy inside those tools too. I've used it to write the microcopy for interactive calculators that convert visitors, because the tone of a calculator's results page matters just as much as the tone of a blog post, and it's exactly the kind of small, repetitive writing task a trained AI handles well once it knows your voice.
Frequently asked questions
Do I need to know how to code to build a custom GPT or Claude Project?
No. Custom GPTs, Claude Projects, and Gemini Gems are all built through a simple upload-and-instruction interface, no code involved. If you can attach a file to an email, you can build one.
How many documents do I need to upload?
Ten to fifteen is a solid working set. Five pieces of your own unedited writing, a few real emails, a short bio, and a banned words list will outperform a folder of fifty random files every time.
Will this stop AI content sounding like AI content entirely?
It gets you most of the way there. You'll still want to read every draft and correct it out loud in the same chat so the project learns the fix. Skipping that step is the most common reason people say "it still sounds robotic" after building one of these.
Is this worth doing if I only post on social media occasionally?
Yes, if inconsistently occasionally still matters to you. A brand brain saves the most time for people who write often, but even someone posting twice a month benefits, because the drafts arrive closer to finished and the voice stays consistent between posts written weeks apart.
Related reading: Business Lessons from Ramit Sethi and Rolex Marketing Strategy: How They Built a Brand That Wins.
Want this done for you? See write for us about AI.
