82% of AI marketing copy sounds interchangeable across brands. Yours probably does too.
The fix isn’t a better AI. It’s a better brief.
Five steps get any AI writing in your voice within an afternoon: capture the voice, build a reusable prompt, feed it examples, test on one task, refine.
Step three is the one most business owners skip, and it’s the one that changes everything.
You’ve paid for the good AI subscription. You’ve read the prompt-engineering threads. You’ve fed it briefs. And yet every piece it writes back sounds like every other business in your industry.
Punchy opener, three “compelling” bullet points, a call to action that could’ve been ripped from any LinkedIn post you’ve scrolled past this week. If your customers noticed at all, they noticed you sounded like a robot dressed up in your logo.
There’s a reason for this. And the fix is easier than most people think.
Why AI defaults to bland
AI writing tools like ChatGPT, Claude, and Gemini are trained on the entire public internet. That means the “voice” they reach for by default is the median of every marketing blog, corporate About page, and LinkedIn thought-leadership post ever written. Which, if you’ve ever read three of those in a row, sounds like a slightly overcaffeinated intern with too many synonyms and no personality.
Left to their own devices, these tools generate the safest possible version of your ask. Safe means bland. Bland means forgettable. Forgettable means your prospect scrolls on.
The good news is that these same tools are excellent mimics. Give them a clear enough picture of how YOU sound and they’ll match it convincingly. The problem is that “clear enough picture” is where most people stop. They tell the AI “write in a friendly, professional tone” and expect magic. Friendly and professional describes about 90% of the internet. It’s not a voice, it’s wallpaper.
The real cost of a generic AI voice
Here’s why this matters more in 2026 than it did a year ago.
Google’s AI Overviews now answer 48% of searches. When Google’s AI decides which two or three sites to quote in its answer, one of the strongest signals it uses is distinctiveness of voice. If your site sounds identical to three others in your category, the AI often picks whichever one has the most authority signals and ignores the rest. Sameness in your writing means invisibility in AI results.
The same is true for social. LinkedIn’s feed algorithm rewards posts that generate reactions and comments. A post that sounds like every other agency post gets a polite scroll. A post that sounds like a real person with a specific opinion gets a save, a share, or a “wait, who wrote this?” comment. That’s the difference between one lead a month and a steady flow.
For a full breakdown of why generic AI copy tanks your marketing, we covered it in Why Your AI Marketing Sounds Like Everyone Else (And How to Fix It). This post is the practical companion. The actual how.
The five-step method
Nothing about this requires a Pro subscription, plugins, or custom fine-tuning. Every step works in the free version of ChatGPT, Claude, or Gemini. Total setup time is roughly one afternoon. After that, you reuse it for every piece of copy your business ever generates.
Step 1. Capture your voice in writing
Before you touch an AI, you need to know how YOU sound.
Pull three pieces of writing you’re proud of. A blog post, an email that got a good reply, a LinkedIn post that landed. Read them out loud. Ask yourself three questions.
What words do I keep using? What words would I never use? What’s my sentence rhythm like: short and punchy, long and rolling, mix of both?
Write down what you find. Not a paragraph. A list. For example:
Words we use: get found, honest advice, straight up, sorted, spot on.
Words we avoid: leverage, synergy, in today’s landscape, cutting-edge, unlock.
Rhythm: mostly short. One long sentence per paragraph. Contractions everywhere. Full stops earn their keep.
This becomes your voice brief. Keep it in a Google Doc or Notion page. You’ll paste it into every AI conversation from now on.
Step 2. Build a reusable prompt template
Now wrap your voice brief in a prompt template that stays the same every time you use AI to write something. The pattern is simple:
“You’re helping me draft [type of copy]. My brand voice: [paste your voice brief]. My audience: [one-sentence description]. Never use these words: [list]. Always use contractions. Write in the same rhythm as this example: [paste one of your three good pieces].”
Save this template. Every time you open a new AI chat, paste this first. Then ask for the actual thing you want.
The reason this works better than “write in a friendly tone” is that you’ve replaced abstract adjectives with concrete rules and one live sample. The AI now has something to match against, not something to interpret.
Step 3. Feed it examples (the step nobody bothers with)
This is where 90% of the difference comes from, and 90% of business owners skip it.
At the bottom of your prompt template, add: “Here are three examples of writing that matches my voice.” Then paste your three pieces of good writing in full.
This technique is called few-shot learning. It’s how AI researchers have known for years that you get an AI to imitate a specific style. Instead of telling the model what you want (which it interprets through the median of the internet), you show it what you want (which it can pattern-match directly).
The moment you add three examples, the output quality jumps by roughly 40% on any subjective voice test. Your AI stops writing in generic marketing English and starts writing in something recognisably close to YOU.
If your examples are on a website already indexed by Google, you can also just link to them. Both ChatGPT (with browsing) and Claude can read live URLs. This keeps your prompt shorter and lets you update the examples any time.
Step 4. Test on one specific task before you scale
Don’t roll this out across your entire content calendar on day one. Test it on one thing first. Something small and low-stakes.
Try a LinkedIn post announcing something ordinary. A short email replying to an FAQ. A three-line intro paragraph for a blog post you’re writing anyway. Compare the AI output against how you would’ve written it manually. If it sounds recognisably like you, you’re ready to scale. If it sounds close but not quite, note where it falls short (too formal, too enthusiastic, wrong sentence rhythm) and adjust your voice brief.
Do this test loop three times before you rely on the setup for anything customer-facing. It takes maybe 30 minutes and saves you from launching a campaign that sounds like a stranger wearing your brand’s clothes.
Step 5. Refine and reuse
Once the output sounds right, save your final prompt template somewhere you can find it every time. A pinned Notion doc, a text expander snippet, a saved Custom GPT if you’re on ChatGPT Pro.
Update the voice brief every couple of months as your voice evolves or you notice new words you’re using. Add fresh examples every time you write something particularly good. Delete examples that feel dated.
The whole system stays under 500 words of prompt text. That’s small enough to paste at the start of any AI conversation and get consistent results across every tool.
Frequently asked questions
Does this work with any AI tool?
Yes. The five steps work identically in ChatGPT, Claude, Gemini, Copilot, or Perplexity. The prompt template is model-agnostic because it uses plain English rules and examples, not tool-specific features. Some tools (like ChatGPT with Custom GPTs, or Claude Projects) let you save the brief and examples once so you don’t have to paste them every conversation. Nice-to-have, not essential.
How long does the setup actually take?
About two to three hours for the first pass, mostly spent on step one (capturing your voice in writing). Steps two through five take an afternoon combined. After that, using the template is a five-second paste at the start of each new conversation. Cost per use once set up: zero.
Can I skip step three if my brief is really detailed?
No. Detailed briefs help, but examples are what actually change the output. This is a quirk of how large language models work: they pattern-match against concrete text more effectively than they interpret abstract instructions. Every AI researcher who works on style transfer will tell you the same thing. Skip the examples and you’ll always be fighting the model’s default voice.
What if I want ohKarl to build this for our team?
Voice work is one of the things we do inside our SEO, AEO and GEO services in South Africa because voice distinctiveness is now a direct SEO ranking factor for AI citation. We can capture your voice, build the template, test it against your existing copy, and hand you a system your whole team can use. Have a look at our full range of digital marketing services or get in touch for a chat.
Sounds like you, at the click of a button
The whole point of AI in your business isn’t to replace your voice. It’s to scale it. A properly briefed AI writes twenty LinkedIn posts a month that all sound like you. Twenty emails, twenty blog paragraphs, twenty product descriptions. Each one still recognisably yours. That’s the whole point.
But it only works if the voice is actually there in the first place. Steps one through three do the heavy lifting. Skip them and you’ll be back to bland by the end of the week.
If you want a shortcut, book a free 15-minute AI visibility audit. Part of what we check is whether the copy on your site reads as distinctly YOU or as generic AI-generated wallpaper. It takes 15 minutes, costs nothing, and tells you exactly where you stand before AI-driven search leaves you behind.
For the research angle on why “voice” now beats “SEO tricks” in the AI search era, Anthropic’s own writing on Claude’s design principles is a solid starting point on how the modern models make style-matching decisions.