The AI Sameness Problem (And How to Sound Like Yourself)

Why Everything Sounds the Same: The Default Voice Machine

Large language models are trained on internet-scale text. Billions of web pages, articles, forum posts, marketing copy. The result is a statistical average of how the internet writes -- what you might call the "median internet voice." It is competent, inoffensive, and about as distinctive as airport carpet.

When you open ChatGPT or Claude and type "write me a blog post about productivity," what you get back is not wrong. It is grammatically sound, logically structured, and thoroughly unremarkable. The AI defaults to a register that hedges everything, qualifies every claim, and pads every transition with filler. It writes the way a committee would write if that committee had read everything ever published but experienced nothing.

You have seen all of these. You have probably published some of them. So have I. The thing is, these tics are not bugs -- they are the model giving you exactly what you asked for, which was nothing in particular. A vague prompt produces a vague voice. The median of everything is the personality of nothing.

Now consider the feedback loop. AI-generated content is flooding the internet. That content gets scraped and fed back into training data for the next generation of models. The median voice converges further. The bland gets blander. It is model collapse in slow motion, except instead of the model breaking down mathematically, it breaks down aesthetically. The writing still works. It just increasingly reads like it was written by a very polite ghost with no opinions.

Readers notice. Google notices. Its Helpful Content guidelines increasingly reward content with genuine experience, expertise, and -- the E that gets the least attention -- authoritativeness. An authoritative voice is, by definition, a specific voice. Not a median one.

The Missing Piece: Your Writing DNA

Every writer has a fingerprint. Not a "tone" or a "style" in the way that those words get tossed around in marketing briefs -- "professional but approachable," "casual yet authoritative" -- but an actual pattern. A set of unconscious habits that make your writing unmistakably yours, even without a byline.

We call this your writing DNA. It is the collection of tendencies you did not choose, the patterns you fell into over years of putting words on screens: the sentence lengths you gravitate toward, the punctuation you abuse, the metaphors you reach for, the words you would never use. It is not what you write about. It is who you are when you write.

Here is the distinction most AI writing guides miss entirely. There is a difference between style and voice. Style is surface: formal versus casual, short sentences versus long ones, active voice versus passive. You can specify style in a prompt and get reasonable results. Voice is deeper. Voice is the specific way you are formal, the particular flavor of your casualness, the things you leave out as much as the things you include.

When someone tells an AI to "write in a conversational tone," they get the AI's idea of conversational, which is everybody's idea of conversational, which is nobody's. When you encode your actual writing DNA -- your sentence rhythms, your vocabulary preferences, your structural habits, your specific brand of humor or lack thereof -- you get something that reads like it came from a person. Not the person, maybe. But close enough that editing bridges the gap.

The components of writing DNA break down roughly like this: sentence rhythm (do you punch or sprawl?), word choice (do you reach for "utilize" or "use"?), structural habits (do you build to a conclusion or lead with it?), humor and tone (dry, warm, absent, weaponized?), and -- most overlooked -- what you omit. The things a writer never says reveal as much as the things they always do.

Custom Instructions That Actually Work

Both ChatGPT and Claude let you set persistent custom instructions -- rules that apply to every conversation. Most people either ignore this feature or fill it with aspirational nonsense. "Be creative and engaging." "Write like a thought leader." These instructions are the prompt equivalent of a motivational poster in a dentist's waiting room. They do nothing.

Effective custom instructions are specific, behavioral, and often negative. Telling AI what not to do is frequently more powerful than telling it what to do, because the model's defaults are the problem and you need to override them explicitly.

Notice the pattern. The weak instructions describe a vibe. The strong instructions describe behaviors. One is a wish; the other is a constraint. AI models respond far better to constraints than wishes, because constraints are falsifiable. You can look at a piece of output and immediately check: did it open with "In today's"? Did it use "delve"? Did it hedge?

Here is a starter set of custom instructions you can paste into ChatGPT or Claude right now and adapt to your own preferences:

## Writing Rules
- Never start with temporal framing ("In today's...", "In an era of...")
- Never use: delve, tapestry, landscape, leverage, nuanced, foster, multifaceted
- Never use "It's important to note" or "It's worth mentioning"
- Vary sentence length. Mix long, complex sentences with short, direct ones
- Prefer active voice. Use passive only for deliberate emphasis
- Use specific examples instead of abstract generalizations
- Do not summarize what you are about to say before saying it
- Do not end sections with generic wrap-up sentences
- When making a claim, support it with evidence or an example, not more claims

Building a Brand Voice Prompt Library

A brand voice prompt -- or what we call a "voice card" -- is a concise reference document that describes how you write, not what you write about. Think of it as a brief for a ghostwriter, except the ghostwriter is a statistical model with an excellent memory and no ego.

The voice card is different from custom instructions. Custom instructions are guardrails: things to avoid, global preferences, formatting rules. A voice card is a portrait. It captures your specific patterns, your rhythms, your quirks. You drop it into the beginning of a conversation when you want the output to sound like you specifically, not just "not like generic AI."

What Goes Into a Voice Card

To build one, you need to audit your own writing with uncomfortable honesty. Not what you think you sound like. What you actually sound like. Here are the questions to answer:

• Sentence structure: Do your sentences tend to be long and subordinated, or short and declarative? Do you use fragments intentionally? How do you handle transitions?
• Vocabulary: What words do you reach for? What words would you never use? Do you lean technical, colloquial, or somewhere between?
• Openings and closings: How do you start pieces? With a question? An anecdote? A blunt statement? How do you end them?
• Humor and tone: Is your humor dry, warm, absent, self-deprecating? Do you use irony? How do you handle serious topics?
• What you never say: This is the most revealing one. What phrases, constructions, or rhetorical moves would feel alien coming from you?
• Content type differences: Does your voice shift between blog posts, emails, and social posts? How? Capture the range, not just one mode.

An Example Voice Card

Here is a simplified voice card to illustrate the format. Yours will be different, because that is the entire point:

## Voice Card: [Your Name / Brand]

Sentence style: Mix of long, complex sentences (3-4 clauses) with
short punchy ones. Fragments OK. Average paragraph: 2-4 sentences.

Vocabulary: Plain language by default. Technical terms only when
precision demands it. No jargon for jargon's sake. Profanity: rare
but not off-limits.

Tone: Dry wit. Skeptical but not cynical. Observational. Never
earnest in the motivational-poster sense. The humor comes from
juxtaposition -- connecting high concepts to mundane realities.

Openings: Start in the middle of things. A specific observation,
a scene, a question. Never a thesis statement. Never "In today's..."

Closings: End with a forward-looking thought or a callback.
Never summarize. Never end with "In conclusion."

Never say: leverage, synergy, game-changer, deep dive (as a noun),
unpack (ideas), stakeholders, circle back, at the end of the day.

Signature moves: Parenthetical asides. Rhetorical questions that
I actually answer. Comparing complex systems to mundane objects.
Occasional one-sentence paragraphs for emphasis.

This is a living document. Update it as you write more and discover patterns you missed. The first version will be incomplete. That is fine. An incomplete voice card still produces dramatically better output than no voice card at all.

Training AI on Your Style: A Practical Framework

Custom instructions clear the weeds. A voice card plants the seeds. This section is about watering them. The most effective technique we have found for making AI output sound like a specific person is deceptively simple: show it your writing, ask it to analyze your patterns, then use those patterns as a prompt.

The Process

1. Select 3-5 writing samples - Pick pieces that sound most like you, not necessarily your "best" work. That blog post you dashed off in twenty minutes that felt effortless? That is more useful than the piece you agonized over for a week, because the effortless one reveals your natural patterns. Aim for variety in topic but consistency in voice.
2. Run the analysis prompt - Paste your samples into a conversation and ask the AI to analyze your writing patterns. Not "what is this about" -- "how does this person write." Ask for sentence structure patterns, vocabulary tendencies, tone markers, structural habits, and recurring stylistic choices.
3. Correct and refine the analysis - The AI will get some things right and some things wrong. That is the point. Correct it. "You said I use formal language -- I don't. I use plain language with occasional technical precision." This dialogue builds a more accurate model of your voice than any prompt you could write from scratch.
4. Generate a test paragraph - Ask the AI to write a short piece on a topic you care about using the voice profile it just built. Compare it to your natural writing. What feels right? What feels off? Adjust the profile based on the gaps.
5. Edit, don't rewrite - This is where most people go wrong. They see imperfect output and either accept it as-is or throw it out and start over. Neither is the right move. AI gets you 70-80% of the way there. Your editing voice -- the instinct that says "I would never phrase it that way" -- closes the remaining gap. That editing instinct is your voice asserting itself. Trust it.

A Prompt You Can Use Today

Here is the analysis prompt we use. Paste it along with your writing samples:

Analyze the writing style in the samples below. I want you to
identify my specific patterns, not generic descriptors. Focus on:

1. Sentence structure: Average length, variation, use of fragments
2. Vocabulary: Words I favor, words I avoid, register/formality
3. Paragraph structure: Length, how I build arguments
4. Tone: Specific humor style, attitude toward the reader
5. Openings and closings: How I start and end pieces
6. Structural habits: How I organize ideas, use of lists vs. prose
7. What's absent: Patterns common in similar writing that I avoid

Be specific. Don't say "conversational tone" -- tell me what makes
my conversational tone different from anyone else's. After your
analysis, generate a "voice card" I can use in future prompts.

[PASTE 3-5 WRITING SAMPLES HERE]

The Before and After: Generic vs. You

Theory is fine. Let us look at what this actually produces. Same topic -- "why most content strategies fail" -- written two ways. One is a default AI prompt with no voice guidance. The other uses a voice card and custom instructions.

The difference is not subtle. The first paragraph could have been written by anyone or anything. The second has a point of view. It makes a specific claim, illustrates it with a concrete scenario, and takes a position. It sounds like a particular person thought about this and had something to say.

That matters beyond aesthetics. Readers who recognize a distinctive voice come back. They share. They trust. Content with genuine personality converts at higher rates than competent-but-generic content, because trust is built on consistency, and consistency requires a recognizable identity.