The short version: You can train an AI tool on your own old emails, blog posts and social captions so it drafts in your actual voice instead of generic ChatGPT-speak, and it cuts drafting time. But it takes a proper afternoon of setup, not the “five minute custom GPT” people promise you, and you’ll still be editing every single output by hand. If nobody’s told you that part, they’ve not done it.
Why every AI-written thing sounds the same
I can spot AI-written LinkedIn posts from three scrolls away now. Same rhythm. Same “here’s the thing” opener. Same three-item list with an em-dash in the middle one. Same fake vulnerability followed by a fake lesson. It’s not that the content is wrong, it’s that it’s nobody’s voice. It’s the AI’s default voice, and millions of people are publishing in it right now.
I noticed this last spring when I asked ChatGPT to draft a newsletter for a client, a small accountancy firm in Leeds, and it came back sounding exactly like the draft I’d written for a completely different client the week before. Different industry, same cadence, same phrases. That’s when it clicked: unless you tell it whose voice to write in, it writes in the voice of everyone. Which is to say, nobody.
So I spent a weekend building what I now just call my voice file, and it’s changed how I use AI for writing more than any prompt trick ever did.
What a voice profile is
It’s not a plugin. It’s not magic. It’s a document, or a set of documents, that you feed into whatever AI tool you use (ChatGPT’s custom GPTs, Claude’s Projects, or even just a pinned prompt) that contains real examples of how you write, plus a plain-English description of your quirks.
I built mine from three sources:
- 40 old blog posts, pasted in as plain text, no formatting
- 20 email newsletters I’d sent over the previous year
- A dozen LinkedIn posts that had done well organically, not the ones I’d forced out
That’s roughly 70 pieces of content, which sounds like a lot until you realise most small business owners already have this sitting in their Mailchimp account or their website’s blog archive. If you’ve been publishing for two years, you’ve almost certainly got enough raw material already.
The step-by-step: building your own voice file
Step 1: Pull your best 30 to 50 pieces of writing
Don’t cherry-pick your most polished corporate copy. Pull the emails you dashed off at 11pm, the posts where you were annoyed about something, the ones that got replies. Voice shows up most in the unpolished stuff. I run email through Mailchimp, and if you’re using a similar setup, going back through your Mailchimp campaign history is one of the fastest ways to find writing that’s already proven it connects, because you can see open rates and click rates sitting right next to it.
Step 2: Write yourself a one-page style brief
This is the bit people skip. I wrote out, in about 400 words, how I talk: short sentences mixed with the odd long one, opinions stated flatly rather than hedged, British spellings, no corporate padding words, a habit of starting sentences with “and” or “but”, real numbers instead of vague claims. I also wrote down what I never do: no exclamation marks stacked up, no “in today’s fast-paced world” openers, no rhetorical questions as a crutch.
Step 3: Feed both into a dedicated space
In ChatGPT, this means building a custom GPT and pasting your style brief into the instructions field, then uploading your writing samples as knowledge files. In Claude, it means setting up a Project with the same material in the project knowledge. Either way, the key is that it’s a persistent space you return to, not a one-off prompt you retype every time.
Step 4: Test it against something you’ve already written
Take a blog post you wrote from scratch and ask the AI to draft the same topic from a one-line brief. Compare the two side by side. The first few times I did this, the AI version was flatter, more generic, missing my actual opinions. That’s useful information: it tells you your style brief isn’t specific enough yet, so you go back and add more detail about the exact thing it got wrong.
Step 5: Keep feeding it new work
Every month or so, I add three or four new pieces to the knowledge base, usually ones that got good engagement or that a client specifically complimented. The voice file isn’t a one-time build, it’s a living folder that gets better the longer you maintain it.
The uncomfortable bit
Here’s what most posts on this topic won’t tell you straight: even with a decent voice file, I still edit every AI draft by hand, and I’d guess I change somewhere between a quarter and a third of the words. It gets me 70% of the way there in a fraction of the time, but the last 30% is still me: swapping in a specific client story, cutting a line that’s technically fine but not quite how I’d say it, adding the blunt opinion the AI softened because it’s trained to hedge.
If someone tells you a trained AI writes exactly like you with zero editing, they’re either selling you something or they’ve never shipped the output to a real audience. The value isn’t “no editing required.” The value is going from a blank page and ninety minutes, to a rough draft and twenty five minutes of editing. That’s still a real win. I’d guess it saves me four to five hours a week across blog drafts, email drafts and social captions combined. It just isn’t the effortless thing the ads make it out to be.
Where this pays off
The biggest gains I’ve seen aren’t in blog posts, they’re in the small, repetitive writing that eats a day without you noticing: follow-up emails after a call, proposal cover notes, LinkedIn comments, the “thanks for the intro” reply. A trained voice file makes those fast without making them sound like a template, because the AI is drawing on your actual phrasing rather than a generic one.
It’s also brilliant for consistency across a team. If you’ve got a VA or a junior marketer drafting content on your behalf, giving them access to the same voice-trained GPT means the brand sounds like one person even when three people are typing. Think about how consistent a brand like Airbnb’s marketing voice stays across dozens of writers and markets, that consistency is built, not accidental, and a shared voice file is one of the cheapest ways a small business can copy that trick.
What it can’t fix
A voice file won’t rescue thin thinking. If your brief is vague, “write something about AI for small business”, the output will be vague no matter how good your style brief is, because voice is the delivery, not the substance. I still do the strategic thinking myself: what’s the actual argument, what’s the one number or example that proves it, what’s the opinion I’m willing to stand behind. The AI dresses that up in something closer to my voice. It doesn’t invent the point.
This is the same lesson Alex Hormozi hammers on constantly about content and offers: the mechanism matters less than the substance underneath it. I wrote about a few of those ideas in my piece on business lessons from Alex Hormozi, and the voice-training exercise is really just that same principle applied to writing: the tool speeds up delivery, it doesn’t manufacture the thing worth saying.
A quick note on visual and platform voice too
Voice isn’t only words. If you’re building out a brand across Instagram, Figma design files, or a website, the same principle of “feed it real examples, don’t just describe the vibe” applies. I look at how brands like the ones I broke down in my Instagram marketing strategy piece and my Figma marketing strategy piece keep a consistent look across hundreds of posts and screens: it comes from a documented system, not a one-off brief. Treat your AI voice file the same way, as a system you build once and keep feeding, not a prompt you write from memory every time.
Tools that make this easier right now
- ChatGPT Plus (custom GPTs): best for solo founders who want one central writing assistant
- Claude Projects: handles longer documents well, good if your voice file has 50+ samples in it
- Notion AI: useful if your writing samples already live in Notion docs
- A shared Google Doc as backup: whatever tool you use, keep your style brief and best samples in one plain document too, so you’re never locked into a single platform
None of these cost more than what most businesses already spend on a single freelance writer’s day rate, and the setup pays for itself within a couple of weeks if you’re publishing regularly.
How to know it’s working
You’ll know your voice file has done its job when a client or a colleague reads a draft and asks “did you write this yourself?” and you can honestly say “mostly.” Not “entirely”, that’s not realistic and I’d be suspicious of anyone who claims it. But mostly is a good outcome, and it’s a much better outcome than the flat, samey AI voice that’s currently filling up half of LinkedIn.
Frequently asked questions
How many pieces of writing do I need before this works?
Thirty is a reasonable minimum, fifty to seventy is better. Fewer than that and the AI doesn’t have enough pattern to work from, so it defaults back to its generic style. Quality and variety matter more than raw volume, so mix long blog posts with short emails and social captions.
Will this stop AI content from sounding like AI content?
It gets you much closer, but it won’t get you all the way there without editing. Think of the voice file as raising the starting point from 40% to about 70% “sounds like you”, with the remaining edit being where you add the specific opinions, stories and numbers that make writing feel human rather than templated.
Is this different from just asking ChatGPT to “write like me”?
Yes, significantly. Typing “write like Jo Smith, friendly and direct” gives the AI almost nothing to work with, so it guesses. Feeding it actual samples of your writing gives it real patterns to copy: sentence length, favourite phrases, how you open and close a piece. The difference in output quality is noticeable within the first paragraph.
Should I hire someone to set this up for me?
You can build a basic version yourself in an afternoon following the steps above. If you want it done across your whole marketing system, with your team trained to use it consistently, that’s exactly the kind of project an AI consultant for small business would normally handle in a single working session, and it tends to pay for itself within the first month of saved writing time.
Related reading: The AI Notetaker Sitting In On Your Client Calls Might Be Breaking Your NDA and Why “Write Like Me” AI Prompts Fail for Small Business Content (And What Works Instead).
