The short version: AI didn’t ruin your brand voice, it just made it obvious you never had one written down anywhere. Fix that with a one-page brand voice document built from things you’ve said to real customers, not from a branding workshop. It takes about ninety minutes to build and it changes everything you paste into ChatGPT afterwards.
The About page that made a room go quiet
I ran a small workshop in February 2026 for fourteen business owners, a mix of florists, bookkeepers, a physiotherapist and a man who sells reclaimed scaffolding boards. I asked everyone to read their About page out loud, the one they’d written with ChatGPT the week before.
By the third reader, people started laughing. Not unkindly. They’d noticed it too. Eleven of the fourteen pages opened with some version of “in today’s fast-paced world” or “in an ever-changing industry.” Nine used the word “passionate.” Eight used “dedicated.” Six ended with a call to action shaped exactly the same way: short sentence, exclamation mark, “let’s talk.”
None of these people know each other. They were using different prompts, different accounts, some on the free version and some paying for the higher tier. And they’d all landed in the same place, because they’d all given the model the same information: a job title and a list of services. Give a system the same thin input and it will hand you back the same average output. That’s not a flaw in the tool. That’s the tool working exactly as designed.
The bit nobody wants to say out loud
Here’s the uncomfortable part. Most small businesses never had a distinct voice before AI turned up. They had a logo, a font from Canva, and a tagline someone’s cousin suggested at a barbecue. There was no document anywhere describing how the business talks, so when a new tool arrived and asked “what’s your tone,” the honest answer was “I don’t know, professional but friendly I suppose,” which is the same answer 90 percent of small businesses give. AI didn’t flatten anyone’s personality. It just exposed that the personality was never written down, so it had nothing specific to copy.
Big brands don’t have this problem because someone was paid, sometimes for years, to write it down. Look at how deliberately plain and gentle the Headspace marketing strategy reads, every sentence short, every metaphor calm, on purpose, in a document their writers are trained against. Or how consistent the tone stays across every touchpoint in the Disney marketing strategy, whether it’s a park sign or an email footer. That consistency isn’t instinct. It’s a written rulebook that gets enforced.
Why “just add more personality” doesn’t work
The usual advice is to tell ChatGPT to “write in a fun, quirky, human tone.” I’ve tested this with client accounts more times than I can count and the result is nearly always worse, because “fun and quirky” is still an instruction with no anchor. The model has to guess what fun and quirky means for you specifically, and it guesses using the internet’s average idea of fun, which is exclamation marks and the word “folks.”
There’s a useful idea here from behavioural economist Richard Thaler, whose work on defaults shows that most people stick with whatever option is already set for them, because changing it takes effort they’d rather not spend. I’ve written before about the business lessons from Richard Thaler, and the same logic applies inside a chat window. ChatGPT’s default tone is the enrolment box that’s already ticked. Most business owners never untick it, because unticking it requires giving the model something specific to replace it with, and nobody hands over specifics if they’ve never written them down.
The one-page document that changes the output
What works isn’t more instruction, it’s more evidence. Feed the model real sentences you’ve said, not adjectives about how you’d like to sound. I built this with a client last spring, a bookkeeper in Manchester with nineteen years in business and zero written brand guidelines. We spent ninety minutes on a video call and built this:
- Twenty real sentences, pulled from her actual emails, voicemail transcripts and WhatsApp replies to clients. Not written fresh, copied word for word from things she’d already said.
- Three words she’d never use, hers were “synergy,” “solutions,” and “streamline.” Small but this alone stops a huge chunk of generic output.
- One real client story, kept word for word, used as the anchor example whenever she needs a case study tone.
- Five firm rules, things like “never use an exclamation mark in the first sentence” and “always explain the number before the benefit, not after.”
- Two sample paragraphs she was proud of, from an old newsletter, kept exactly as written, no editing.
That’s the whole document. One page, saved as a text file, pasted into the start of every single ChatGPT session before any actual prompt. Not a fifty-page brand bible nobody reads. One page, built from real words she’d already said out loud, not from adjectives she thought she should aspire to.
Six weeks later her open rate on the client newsletter went from around 21 percent to 34 percent, and three separate clients mentioned they could “hear her” in the emails for the first time. That’s the entire fix. Not a better prompt. Better raw material for the prompt to work from.
What this looks like in practice
Before the document, a prompt like “write a LinkedIn post about late tax return penalties” produced this kind of opener: “Are you worried about missing the tax deadline? You’re not alone!” Generic, could be any accountant in the country.
After pasting in her one-page document first, the same prompt produced: “I’ve had four calls this week from people convinced HMRC is about to fine them into oblivion. Here’s what happens if you miss 31 January, and it’s rarely what you think.” Same request, same model, completely different starting material. The words ” happens” and the specific detail about four calls came straight from her real sentences, not from the prompt itself.
Editing out the average
Coco Chanel famously told people to look in the mirror before leaving the house and remove one thing. That instinct, cutting rather than adding, applies directly here. I’ve written about the business lessons from Coco Chanel before, and the discipline of taking something away is more useful for AI copy than any instruction to add more flair.
Once you’ve got a draft from ChatGPT, don’t ask it to make it punchier or more engaging, that just adds more average words on top. Instead, go through and remove every sentence that could have been written by any other business in your industry. If a competitor could paste your paragraph into their own website and it would still fit, delete it. What’s left is usually short, sometimes a bit rough, and always more specific than what you started with.
Beyond the wording, into the format
There’s a second layer to this that most people skip entirely. Even when the wording is fixed, the format is still generic, because everyone is publishing the same blog post shape: intro, three subheadings, conclusion. If you want to stand out on format rather than just tone, build something people have to use rather than read. I’ve covered this before with interactive calculators as an engagement tool, and the same logic works here. A simple calculator, a quiz, a quick assessment tool, these can’t be produced by pasting a prompt into ChatGPT the way a blog post can, so they instantly separate you from the flood of AI-written pages competing for the same search terms.
Keeping the document alive
The one-page document isn’t a one-off task. Update it every time something in your business changes, a new offer, a new type of customer, a phrase that starts working in real conversations. Review it monthly, not yearly. Five minutes, read it back, add anything new you’ve said that week that sounded like you.
If you run a team, this becomes more important, not less. Every person on your team pasting prompts into ChatGPT without a shared document produces five different versions of your brand within a fortnight. That’s the point where it’s worth getting someone to sit down with your team for an afternoon and build this across every department, which is exactly the kind of hands-on session an AI implementation coach would run, rather than trying to fix it through a company-wide memo nobody reads twice.
The part that surprises people
The thing clients are always surprised by is how short their finished document ends up being. They expect a brand voice guide to be a proper strategy deck with a colour palette and a mission statement. It isn’t. It’s twenty sentences you already said, three words you hate, one story, five rules. That’s all the model needs, because the model was never short on ability, it was short on material. Give it your actual words instead of your idea of professional words, and the sameness disappears within the first paragraph.
Frequently asked questions
Why does all AI-generated marketing copy sound so similar?
Because most people give ChatGPT the same thin input, a job title and a list of services, with no real examples of how they talk. The model fills the gap with the internet’s average tone, which produces the same phrases, “in today’s fast-paced world,” “passionate and dedicated,” across thousands of unrelated businesses.
Do I need a full brand style guide to fix this?
No. A single page works, built from twenty real sentences you’ve already said to customers, three words you’d never use, one real client story and five firm rules. Paste that page into every ChatGPT session before your actual prompt.
How long does it take to build a brand voice document?
Around ninety minutes for a one-person business, longer if you’re pulling examples from a team. The document itself is short, one page, but gathering real sentences from old emails and messages takes most of the time.
Will this work for a team, not just a solo business owner?
Yes, but it needs to be shared and actively used, not just written once. Without it, every team member’s prompts drift into a different generic tone within weeks. Larger teams often get more out of running this as a proper session with outside help than trying to enforce it through a memo.
Related reading: I Asked AI to Audit My Own Blog. It Told Me to Delete Half of It. and I Write Client Proposals With AI in 20 Minutes. Here’s the Bit Nobody Tells You.
