Voice DNA: defining the 5-8 traits that make your writing recognizable

Voice DNA is the section of the Persona Brief that defines who you sound like. Most teams get this wrong by writing aspirational descriptors ("engaging," "premium," "authentic") that have no operational meaning to the AI.

This post is the methodology for extracting voice DNA from your actual writing and codifying it in a way that produces consistent AI output.

The diagnostic: what voice DNA actually is

Voice DNA is the set of behaviors that make your writing recognizable across multiple posts. Not your personality. Not your brand values. The specific writing-level patterns:

• Sentence length distribution (do you write short? long? mixed deliberately?)
• Vocabulary register (formal vs colloquial, technical vs accessible)
• Hook patterns (do you open with claims, stories, questions, numbers?)
• Transition style (do you use bridges, jump cuts, or implicit transitions?)
• Closing style (callbacks, provocations, CTAs, no close?)
• Specific quirks (capitalization, em-dashes, double periods, ellipses)

Step 1: Collect your 10 best voice-accurate posts

Same criteria as reference-post selection: voice-accurate first, performance second. Pick 10 posts that sound the most like you and performed well. Span platforms if possible.

Step 2: Identify recurring patterns

Read all 10 posts back-to-back. Look for what repeats:

• Do your sentences trend short, long, or deliberately mixed?
• Do you tend to open with the punchline or build up to it?
• Do you cite specific numbers or use general claims?
• Do you use first-person, second-person, or third-person framing?
• Are your transitions explicit ("now, here is the second point") or implicit (jump cut to the next idea)?
• How do you close? Callback to the opener? CTA? Provocation? Drop the mic?
• What words or phrases recur? What words do you NEVER use?

Step 3: Codify each pattern as a one-line trait

Each voice DNA trait should:

• Produce a specific, observable behavior in output
• Be unambiguous — the AI can validate compliance
• Be actionable — the AI knows what to do

Examples that work:

• "Direct, no hedge words" — produces sentences without "may," "might," "could potentially"
• "Short sentences, max 18 words" — caps sentence length
• "Specific numbers always — cite hours, dollars, percentages, never 'many' or 'a lot'" — forces concreteness
• "Contrarian openers — lead with a claim that surprises" — sets hook pattern
• "Closing is a callback to the opening, not a summary" — defines closing behavior
• "Writes like Naval Ravikant if Naval ran a B2B SaaS" — references a known author for pattern-matching

Examples that do NOT work:

• "Engaging and thought-provoking" — vague, no operational behavior
• "Premium tone" — meaningless to the AI
• "Voice of a thought leader" — no specific pattern
• "Professional but friendly" — contradictory pair, model averages to nothing

Reference creator pattern

Reference creators by name when their voice resonates with what you want. Base models (GPT-5, Claude 4, Gemini 2) know most well-known writers. "In the style of [name]" is the single highest-signal voice instruction.

How to use it well:

• Pick creators in your industry or adjacent. Naval Ravikant for B2B founder voice, Paul Graham for SaaS startup voice, Morgan Housel for finance writing, Mark Manson for personal-development voice.
• Pair with a modifier. "Like Naval if Naval ran a B2B SaaS" beats "like Naval" because it adds your specific context.
• Avoid over-referencing. 1-2 creator references in your voice DNA section is enough. More than that produces averaging.

Common voice DNA failures

• Aspirational descriptors with no operational meaning ("authentic," "premium," "thought-provoking")
• Contradictory pairs ("professional but friendly," "concise yet thorough")
• Industry cliches ("data-driven," "value-focused," "results-oriented")
• Personality essays that span paragraphs (voice DNA is one-line traits)
• Too many traits (above 8, signal dilutes)
• Too few traits (below 5, voice is under-defined)

Industry-specific voice DNA patterns

B2B SaaS founder

• Direct, no hedge words
• Specific numbers always
• Engineering-first framing — cites technical specifics
• Anti-thought-leader — no motivational filler
• Writes like Paul Graham if Paul Graham still ran companies

Creator / personal brand

• Story-led openers — open with a specific moment
• Short sentences mixed with long ones — deliberate rhythm
• Specific timestamps and dollar amounts
• Conversational vocabulary — uses contractions
• Closes with a provocation or callback, never a summary

Real estate investor

• Numbers-first — every claim has a dollar amount or percentage
• Operator voice, not influencer voice
• Cold-emailer mentality — direct, anti-fluff
• Specific deal references over general principles
• Closes with a tactical takeaway

Coach / consultant

• Framework names always explicit
• Client-language not industry jargon
• Story-first openers tied to a specific session or moment
• Authority + warmth balance
• Closes with an actionable next step

How voice DNA interacts with other sections

Voice DNA is the most general layer. It is overridden by:

• Required structures (section 4) — for format-specific rules
• Reference posts (section 5) — for concrete examples that pattern-match
• Platform overrides — for platform-specific voice adaptations

Think of voice DNA as the base personality, required structures as the format rules, and reference posts as the live examples. All three reinforce each other.

Iterating voice DNA over time

Voice DNA is not static. Audit it quarterly:

1. Read 10 of your most recent best-performing posts.
2. Identify any new patterns that have emerged.
3. Add new traits if needed (or refine existing ones).
4. Remove traits that no longer match your current voice.

A mature voice DNA section stabilizes after 6-12 months. Once stable, quarterly audits surface minor refinements rather than major shifts.

Frequently asked questions

Related guides in AI Brand Voice & Persona

• The 5-section Persona Brief template that captures your voice — Who you are, voice DNA, banned words, required structures, reference posts — the exact template that powers Kompozy outputs.
• How to use reference posts to fine-tune AI voice (without actually fine-tuning) — Why 3-5 well-chosen reference posts beat any amount of voice description in a prompt.
• The complete AI banned-word library (kill every AI tell) — 120+ phrases that flag content as AI-written: hedge words, tricolons, "not just X but Y," vague authority claims. With drop-in copy-paste lists.

Adjacent clusters

• Autonomous Content Creation — Most "autonomous" AI content is slop. Here is how 4 quality gates make autopilot output indistinguishable from manually-approved content — and the exact 14-day ramp to flip the switch safely.

← Back to AI Brand Voice & Persona overview · Start a free trial → · See pricing