If you have used AI for marketing for more than five minutes, you have probably seen this happen.

You ask it to write an email.

The grammar is fine.

The structure is fine.

Nothing is technically wrong.

And yet it sounds like it came from a company you have never heard of.

That is the real problem with a lot of AI writing.

It isn’t necessarily bad.

It is interchangeable.

The same opening could belong to a software company, a consultant, a fitness coach or someone selling kitchen knives.

When everyone has access to the same writing models, generic competence becomes cheap.

Your brand voice becomes more valuable.

The solution is not to keep asking:

“Make this sound more human.”

That usually creates another flavour of generic.

A better approach is to teach the AI what your version of human sounds like before it starts.

What Does “Brand Voice” Actually Mean?

A brand voice is not:

friendly, professional and engaging.

Thousands of companies could say that.

AI needs something more concrete.

A useful brand voice can be broken into things the model can actually observe.

For example:

• words you use often;
• words you avoid;
• sentence length;
• level of formality;
• whether you use contractions;
• how direct you are;
• whether you use humour;
• how you explain difficult ideas;
• how you speak to customers;
• phrases that recur across your content;
• how strongly you state opinions.

Recent brand-voice guidance increasingly treats this as a specification rather than a vague personality description.

That distinction matters.

“Sound casual” is subjective.

“Use contractions, avoid corporate jargon, write mostly in short paragraphs and explain ideas with concrete examples” is much easier for a model to follow.

1. Start With Writing You Already Like

The easiest way to teach AI how you write is to show it.

Collect five to ten pieces of content that already sound right.

They might include:

• emails;
• blog posts;
• landing pages;
• social posts;
• sales copy;
• customer replies;
• scripts.

Don’t choose pieces just because they performed well.

Choose them because they actually sound like your brand.

Your source article makes this same point indirectly with before-and-after examples: showing the model what you changed teaches it far more than abstract instructions alone.

You can begin with a prompt such as:

Analyse these writing samples.

Identify recurring patterns in:

• sentence length
• vocabulary
• tone
• use of humour
• directness
• paragraph length
• level of formality
• repeated phrases

Do not rewrite anything yet. Build a concise brand voice profile from the examples.

The important part is that the AI observes first.

Writing comes later.

2. Give AI Real Customer Language

This may be even more useful than giving it your own marketing copy.

Your customers already have a vocabulary for the problem you solve.

Use it.

The source article recommends pulling language from reviews, support messages, testimonials and similar sources rather than asking AI to guess how customers speak.

That is good advice because brands often describe products differently from customers.

A software company might say:

streamlined project visibility

A customer may say:

I finally know what everyone is working on.

The second line is more useful.

Collect language from:

• product reviews;
• testimonials;
• support tickets;
• sales calls;
• surveys;
• customer emails;
• Reddit discussions;
• social comments.

Then create a separate section in your brand voice file:

Customer language

Customers often say:

“easy to set up”
“finally made sense”
“saved me hours”
“I didn’t need to watch ten tutorials”

Prefer this vocabulary when relevant.

Now the AI is working from language that already exists in the market.

It has less room to invent marketing fluff.

3. Create a List of Words You Never Want to See

Knowing what your brand does not sound like is surprisingly useful.

The original article suggests explicitly banning words such as “synergy,” “revolutionary” and “game-changing.”

That can be turned into a permanent rule.

For example:

Avoid

• leverage
• game-changing
• revolutionary
• seamless
• unlock
• elevate
• cutting-edge
• transform your journey

Your list will be different.

The point is not that these words are forbidden English.

It is that repeated AI writing tends to drift toward safe, familiar marketing vocabulary.

Your anti-AI writing material makes a similar observation: AI output often leans heavily on generic positive adjectives, jargon and words such as “elevate” and “delve.”

A banned-word list gives the model fewer hiding places.

4. Define How Your Brand Builds Sentences

Voice isn’t only vocabulary.

Rhythm matters.

Consider these two versions.

Version A

Our platform provides comprehensive solutions that enable businesses to improve their operational efficiency and achieve better outcomes.

Version B

The tool does one job.
It gets the busywork out of the way.

Same general idea.

Different voice.

The source article recommends using contractions, sentence fragments and deliberate sentence-length variation.

Rather than adding those as random prompt tricks, turn them into permanent style rules.

For example:

Sentence style

• Prefer short and medium-length sentences.
• Use contractions.
• Sentence fragments are allowed sparingly.
• Break long thoughts into separate sentences.
• Avoid long chains of clauses.
• Use one-sentence paragraphs occasionally for emphasis.
• Don’t make every paragraph follow the same rhythm.

That gives AI a repeatable writing pattern.

5. Give It Signature Language

Brands often develop small phrases without deliberately trying to.

A company might call customers:

builders

instead of:

users.

Or refer to small experiments as:

small bets.

These phrases matter because they make the brand recognisable.

Your source article specifically recommends preserving these linguistic fingerprints and feeding them directly to the model.

Create a simple section:

Preferred phrases

• “small bets” = low-risk tests
• “working system” = repeatable process
• “remove friction” = simplify unnecessary effort

Then tell AI:

Use these phrases only when they fit naturally. Do not force them into every piece.

That last sentence matters.

A signature phrase used occasionally feels like a voice.

Used six times in one article, it feels like branding software malfunctioning.

6. Give AI Examples of What You Changed

This is probably the most useful technique in the original article.

Don’t only show finished content.

Show the edit.

For example:

AI draft

Our innovative platform enables users to leverage powerful automation capabilities to maximize productivity.

Edited version

Automate the repetitive work and spend your time on the part that needs you.

Then tell the model why.

Changes made:

• removed vague praise;
• replaced jargon with plain language;
• focused on the result;
• shortened the sentence;
• removed “innovative,” “leverage” and “maximize.”

Do this several times.

The source describes this as teaching through before-and-after examples rather than relying solely on instructions.

Current brand-voice advice also recommends contrastive examples because models can infer style differences more reliably from examples than from vague labels.

This gives AI something extremely useful:

your editing behaviour.

That is closer to your real voice than any list of adjectives.

7. Stop Asking AI to “Sound Human”

That instruction is too vague.

So are:

make it engaging.

make it natural.

make it punchier.

make it conversational.

They can help a little.

But the model still has to guess what you mean.

Replace them with observable instructions.

Instead of:

Make this more human.

try:

Rewrite this using:

• short paragraphs;
• contractions;
• plain language;
• no corporate jargon;
• concrete examples where possible;
• varied sentence lengths;
• no generic introductory paragraph;
• no summary that merely repeats the article.

Now the model knows what behaviour to change.

This also fits your anti-AI source, which points out that generic AI writing frequently restates itself, overexplains and uses unnecessary preamble.

8. Don’t Force Personality Where None Exists

This is where many “make AI sound human” tutorials go wrong.

They overcorrect.

Everything becomes snarky.

Every paragraph contains a joke.

Every headline tries to provoke somebody.

That is not necessarily a brand voice either.

The original article suggests forcing a “controversial or unexpected opinion.”

I would change that rule.

A better instruction is:

When the subject warrants it, state a clear point of view instead of defaulting to a neutral consensus. Do not manufacture controversy.

That is much safer.

Sometimes the strongest brand voice is simply clear.

A tax software company doesn’t need to sound like a stand-up comedian.

A funeral business probably doesn’t need “hot takes.”

Distinctive does not mean eccentric.

9. Use Examples Instead of Arbitrary Writing Tricks

Some of the techniques in the source are useful exercises but weak permanent rules.

For example:

use only one-syllable words.

That can help reveal simpler alternatives.

But you probably don’t want an entire brand writing that way.

Likewise:

kill all adverbs.

That may tighten some copy, but not every adverb is a mistake.

The better rule is:

Prefer precise verbs and nouns. Remove modifiers that don’t add meaning.

That preserves the useful principle without turning your writing into a grammar game.

The aim is not to follow rigid rules.

The aim is to make the model behave more like your editor.

10. Build a Reusable Brand Voice File

Once you have all of this, stop rebuilding it inside every prompt.

Create one reusable document.

Call it something simple:

Brand Voice Guide for AI

Include:

Brand identity

Who you are and who you write for.

Tone

Direct, warm, dry, technical, casual or whatever actually fits.

Vocabulary

Words you prefer and words you avoid.

Customer language

Real phrases collected from customers.

Sentence rules

Length, rhythm, paragraphs and formality.

Signature language

Recurring phrases unique to your brand.

Good examples

Three to five pieces that sound right.

Bad examples

A few examples of what you never want.

Before-and-after edits

Examples showing how you personally improve AI drafts.

Now you have a reusable context layer.

That is much more powerful than trying to remember thirteen clever instructions every time you open a new chat.

11. Give the AI a Separate Writing Task

Once the voice profile is established, the actual prompt can remain relatively simple.

For example:

Use the attached brand voice guide.

Write a product announcement email for our new reporting feature.

Audience: existing customers.

Goal: encourage them to try the feature.

Length: 150–200 words.

Important facts:

• reporting is now automated;
• reports can be scheduled weekly;
• the feature is included in existing plans.

Follow the brand voice rules exactly. Do not invent customer claims or product benefits.

Notice what is missing.

You don’t need to explain your entire writing philosophy again.

The system already contains it.

12. Edit the First Draft Yourself

Even with a good brand voice system, the first output shouldn’t automatically become the final output.

Read it.

Preferably out loud.

Your anti-AI reference makes this point very clearly: if AI smooths away too much of your individual style, some of the mess may need to go back in. It recommends using AI for pruning and criticism rather than allowing it to erase the writer from the process.

Ask:

Does this sound like something we would actually say?

Is the opinion ours?

Are the examples real?

Did AI invent certainty?

Would I send this to a customer without feeling slightly embarrassed?

If not, edit it.

Those edits become useful data.

Save the best ones and add them back to the voice guide.

Now the system improves.

13. Update the Voice Guide as Your Brand Changes

Brand voice isn’t fixed forever.

Your business changes.

Your audience changes.

You discover phrases customers respond to.

Some wording gets overused.

Certain jokes stop being funny.

Once every few months, compare your latest good content with the current voice guide.

Ask AI:

Compare these recent examples against our current brand voice rules.

Identify:

• new patterns;
• rules we no longer appear to follow;
• phrases we’re overusing;
• vocabulary that has entered the brand naturally.

Do not change the guide yet. Show me the differences first.

Then decide what deserves updating.

That final decision should stay human.

A Simple Brand Voice Prompt Template

You can reduce the entire process to this:

Brand: [NAME]

Audience: [AUDIENCE]

Voice: [SHORT DESCRIPTION]

Writing rules:
[RULES]

Words to prefer:
[WORDS]

Words to avoid:
[WORDS]

Signature phrases:
[PHRASES]

Customer language:
[EXAMPLES]

Reference samples:
[PASTE 3–5 EXAMPLES]

Before/after edits:
[PASTE EXAMPLES]

Task:
[WHAT YOU WANT WRITTEN]

Output requirements:
[FORMAT, LENGTH, PURPOSE]

Preserve the meaning and brand point of view. Do not invent facts, claims, anecdotes or customer experiences.

That is far more useful than:

Make this sound human.

How to Know Whether the Voice Is Working

Try a simple test.

Take five recent pieces generated with AI.

Remove the company name.

Read them together.

Do they sound as though the same organisation wrote all five?

If one sounds like a consultant, another like a SaaS startup and another like a motivational LinkedIn creator, you don’t have a stable AI brand voice yet.

Another useful test is simpler.

Read a paragraph aloud.

Would somebody on your team actually say it?

If the answer is no, the AI has probably taken over too much.

Current discussion around AI writing is increasingly circling back to this point. Generic AI content has become common enough that distinctly human, recognisable communication is becoming more valuable, not less.

Frequently Asked Questions

How do I train ChatGPT on my brand voice?

Give ChatGPT a reusable brand voice guide containing your tone rules, preferred and banned vocabulary, real writing samples, customer language and examples of edits you’ve made. Then reference that guide whenever you ask it to create content.

How many writing samples should I give AI?

A small set of strong, representative examples is usually more useful than dumping everything you’ve ever written into the prompt. Several current guides recommend roughly three to ten carefully chosen examples depending on the task.

Can AI completely copy my writing style?

It can imitate observable patterns, but it still does not possess your experiences, opinions or judgment. You should review the final content, particularly where the piece contains claims, personal experiences or strong opinions.

How do I make AI writing sound less generic?

Give it more specific inputs: real customer language, examples of your writing, banned words, sentence rules, signature phrases and examples of edits you have made.

Should I use AI humanizer tools?

They may change surface-level wording, but they do not automatically understand your brand. A documented voice system with real examples gives you more control over the final result.

Should I use the same brand voice for every type of content?

Not exactly. The underlying voice should remain recognisable, but the level of formality and detail can change depending on whether you’re writing a support message, landing page, technical article or social post.

Tools & Resources

You can build this system using almost any major general-purpose AI assistant.

You don’t necessarily need specialised software.

The important assets are the ones you own:

• your best writing;
• your customer language;
• your editing preferences;
• your vocabulary;
• your brand point of view.

Specialised brand-voice tools can help store and reuse this material, and some platforms now build persistent voice profiles from writing samples.

But the underlying principle is the same.

The AI needs examples of what your brand sounds like.

And somebody still needs to decide what the brand believes.

That part remains yours.