How to Train AI on Your Voice (Without Sounding Like a Robot)

By GetGood · July 8, 2026 · 5-minute read

Why does AI writing sound so generic?

It helps to know what's actually happening. An AI language model has read an enormous amount of writing, and when you ask it to write with no other instruction, it produces something near the middle of all of it. The middle is grammatical, inoffensive, and completely faceless — the "As a business owner, you know that customer satisfaction is key" school of prose. It isn't broken. It's giving you the average because you never told it which voice is yours.

So the whole task of "training AI on your voice" is really one job: pull it off the average and toward you. You do that not by finding a magic prompt, but by giving it evidence of how you actually sound. This is the same principle behind giving your AI business context — the more it knows about your specific world, the less it defaults to generic.

Step 1: Feed it real samples (this does most of the work)

The single most effective move is also the simplest: collect three to five pieces of writing you're genuinely proud of and hand them to the AI as examples of your voice. Good sources:

• An email you wrote that got a great reply
• A social post that sounded like you on a good day
• A page from your website you'd never let a copywriter "improve"
• A message to a customer where you explained something clearly

Then say, plainly: "Here are examples of how I write. Study the rhythm, the word choices, and the tone. Write in this voice." You'll usually see the difference in the very first draft. Quality matters more than quantity here — five samples that are truly you teach more than fifty that were polished by committee until every fingerprint was buffed off.

Step 2: Write down your rules (the things samples don't show)

Samples capture your rhythm; a short rules list captures the decisions you make on purpose. Keep it to a page. The useful ones:

• Words you use — the phrases that are yours ("here's the honest version," "let's keep it simple").
• Words you never use — the corporate tics that make you cringe ("leverage," "synergy," "unlock," "at the end of the day").
• Sentence length — short and punchy, or longer and warm? Tell it.
• Formality — do you say "hey" or "hello," "thanks" or "kind regards"?
• What you'd never say — no hype, no fake urgency, no exclamation points. Your no-list is as defining as your yes-list.

A "never use" list is often the fastest fix of all, because most of what makes AI sound like AI is a handful of overused words. Ban them by name and half the robot vanishes.

Step 3: Give feedback that sticks

The first few drafts won't be perfect, and that's the process working, not failing. When something's off, don't just re-roll — say why: "Too formal, loosen it." "You used 'leverage' — never do that." "This second paragraph sounds like a brochure; make it sound like me talking to one person." Specific correction teaches far more than "try again."

The trick that makes this compound: capture what you learn. Every time you correct the AI, add the correction to your rules file so you never have to say it twice. Over a few weeks, that file becomes a genuine style guide — and the AI that reads it every time gets quietly, permanently better. Feedback you throw away teaches nothing; feedback you write down teaches forever.

Where to keep all this (so you set it up once)

The samples and rules only pay off if the AI reads them every time, not just when you remember to paste them. That's the difference between a one-off clever prompt and a trained assistant. In a properly set-up AI office — the kind we walk through in setting up Claude Code for your business — your voice samples and rules live in a file the AI opens automatically at the start of every task. You do the training once; it applies forever, to every email, post, and reply.

Set it up that way and "write in my voice" stops being an instruction you repeat and becomes a fact your AI already knows.

The honest limits

A few things to keep your expectations true:

• It gets you a draft, not a finished piece. A well-trained AI hands you a strong first draft that already sounds like you — but you're still the editor. It saves you the blank page, not the judgment.
• It can't invent what it hasn't seen. If you've never written about a topic, it will guess at how you'd sound. Feed it a relevant sample and the guess improves.
• You still own the truth. The AI can match your tone perfectly and still write something that isn't quite accurate. Sounding like you is not the same as being right — you check the facts, always. AI output should be reviewed by a person before it goes out.

We should disclose our bias: at GetGood we set clients up on Claude Code because it keeps this kind of standing voice file on your own machine and reads it on every task — but the method here works with any capable AI tool that lets you provide examples and instructions. The samples-plus-rules-plus-feedback approach is the point; the tool is the container.

None of this is a course you need to take. It's an afternoon: gather five samples, write one page of rules, and correct the first few drafts out loud. Do that, and the robot voice you've been fighting simply stops showing up.