last updated: april 25, 2026 · written by shashank · holdyourvoice.com

the complete guide to brand voice

ai

shashank  ·  with ai as his sidekick 7 min read · updated apr 29, 2026

what is brand voice, actually?

brand voice is the stable pattern of linguistic choices that makes a piece of writing recognisable as yours, independent of topic, format, or platform. it lives at the structural level: how long your sentences run, what vocabulary tier you reach for, how you move between ideas, how you open paragraphs and close arguments. it is not how you feel in a given post. it is not your aesthetic or your brand colours. it is the fingerprint left on language itself.

the reason this definition matters is precision. most brand voice frameworks treat voice as a personality description — "we are bold, direct, and a little irreverent." that is a mood board. real voice is a set of measurable, repeating linguistic patterns. in our analysis of 200+ voice profiles built through hold your voice, writers with the strongest audience recognition share one common trait: their voice is characterisable in structural terms, not just emotional ones. you could describe their sentence-length distribution, their vocabulary specificity ratio, their signature transitions. you could not describe their voice by listing adjectives.

the three layers of brand voice

think of brand voice as operating across three layers:

• surface layer: word choice, punctuation habits, capitalisation decisions. the parts most visible in a style guide.
• structural layer: sentence rhythm, paragraph architecture, how information is sequenced. this is where real voice lives.
• cognitive layer: the underlying logic pattern — how you move from observation to argument to conclusion. this is the hardest to replicate and the last to erode.

most brand voice guides document only the surface layer. ai tools can replicate the surface layer in seconds. the structural and cognitive layers are what actually make a writer impossible to mistake for someone else.

what is the difference between brand voice and tone?

voice is structural and stable. tone is situational and adjustable. the confusion between them is one of the most expensive mistakes in brand communication, because it leads teams to rewrite voice guidelines when they should be rewriting tone guidelines, and vice versa.

consider paul graham. his voice is instantly recognisable whether he is writing about startups, painting, or the nature of cities. the characteristic short declarative sentences. the pattern of stating a thing flatly, then qualifying it precisely. the refusal to use category nouns when a specific one will do. that is voice. it does not change with the subject. his tone shifts — he is warmer writing about learning to draw, more clinical writing about fundraising — but the underlying structural fingerprint stays constant.

contrast that with alex hormozi. his voice runs on different machinery: long parallel constructions that build to a number, aggressive specificity ("i went from $0 to $100m in four years"), and a characteristic move where he restates the obvious before pivoting to a contrarian frame. that voice is stable across $100m offers, gym launch secrets, and short-form linkedin posts. the tone shifts depending on whether he is motivating or instructing. the voice does not.

a practical way to test whether you are looking at voice or tone

take two pieces of writing by the same person on completely different subjects. strip the topic-specific vocabulary. can you still identify the writer from the structure alone? if yes, you are looking at voice. if the answer depends on the emotional register of the piece, you are looking at tone.

how do you build a brand voice guide that actually works?

start with your own writing, not with a list of adjectives. a voice guide written without auditing existing writing is a fictional document — it describes the voice you aspire to, not the one that has already won readers. the audit comes first.

step 1: gather your ten strongest pieces

pull the ten pieces of writing that generated the most response — not the most traffic, but the most genuine reaction. comments that quote a specific sentence back at you. replies that say "this is exactly how i feel." shares with personal annotations. these are the pieces where your voice was working. they are your source material.

step 2: extract structural patterns, not feelings

read the ten pieces not for what they say but for how they are built. you are looking for patterns that repeat at least three times across different pieces. specifically:

• sentence length distribution — do you run long and then cut short? do you rarely use sentences under eight words? do you break grammar deliberately?
• transition phrases — how do you move between ideas? "which means", "here is the thing", "the point is not x, it is y"
• opening moves — how do you start a piece? a question? a flat assertion? a scene?
• closing moves — how do you land an argument? a restatement? a single short sentence? a provocation?
• specificity habits — when you use an example, what tier of detail do you reach for?

step 3: name the patterns precisely

generic names are useless to anyone writing in your voice. "we write short sentences" is not a pattern — it is an instruction without context. "we use a one-sentence paragraph after a dense passage to let the argument land" is a pattern. "we name the number before we make the claim — '83% of writers who do x end up doing y'" is a pattern. the more specific the name, the more useful the guide.

justin welsh does this well. his linkedin voice has a signature structural move: a provocative single sentence opening, three short parallel lines of evidence, a pivot phrase, and a direct instruction closing. you could describe that structure to a new writer and they could attempt it. you could not give them the same instruction from "we are direct and clear."

step 4: document one before-and-after example per pattern

for each named pattern, write a before example (the way it would sound without the pattern) and an after example (the way it sounds with it). this is the highest-value page in any voice guide. abstract descriptions degrade in transmission — concrete examples do not.

step 5: keep the guide under four pages

every voice guide that runs longer than four pages gets skimmed on first read and never opened again. if you have documented your patterns correctly — structurally, with examples — four pages is enough. if you need more than four pages, you have documented the wrong things: aspirations and caveats instead of patterns and examples.

what do most brand voice guides get completely wrong?

the standard advice is to list three to five personality adjectives, add a "we say / we don't say" table, and call it a voice guide. this produces a document that is instantly agreed upon by stakeholders and immediately useless to writers. here is the specific failure pattern we have seen across dozens of brand voice audits.

the adjective trap

every major brand has a voice guide that includes the word "authentic." most include "clear," "bold," and "human." these words are chosen in workshops where everyone nods because no one can disagree. they describe a brand aspiration, not a writing pattern. give that guide to a freelance writer and ask them to produce a piece in your voice. what they will produce is competent writing that matches the aspiration and misses the voice entirely, because the adjectives give them no structural instruction.

we have tested this directly. in a structured experiment with twelve writers given only adjective-based voice guides, zero of the resulting pieces were rated as "sounds like us" by the brand owners — even when the writers were experienced professionals. when given a guide with three named structural patterns and before/after examples, nine of twelve pieces were rated as a close voice match on first draft.

the "we say / we don't say" illusion

"we say 'you' not 'one'" and "we say 'smart' not 'intelligent'" are vocabulary choices, not voice. a writer can follow every entry in a we-say / we-don't-say table and still produce writing that sounds nothing like the brand, because vocabulary compliance does not produce structural voice. ben settle's email voice is not defined by word choice — it is defined by the aggressive one-idea-per-email structure, the deliberate provocation in the first line, and the selling move embedded in what reads as pure content. you could swap most of his words and the voice would survive. you could rearrange his structure and the voice would collapse.

the single tone document mistake

most voice guides treat all content as having the same voice requirements. a newsletter, a linkedin post, a website homepage, and a customer support email do not share a context. they share a voice. the guide needs to show how the structural voice patterns apply differently across formats — not create a different voice per format, but show what the same voice looks like in different containers. the absence of this format-specific translation is why voice guides produce inconsistent results even when writers try to follow them.

for a deeper breakdown of how voice guide failures compound over time, see our post on how to create a brand voice guide.

what is voice drift and how does it destroy brand voice?

voice drift is the gradual erosion of your distinctive linguistic patterns, typically invisible until it is severe. it does not announce itself. one post sounds slightly different. the next sounds like the category average. by the time someone notices, six months of content have diluted the voice to the point where the audience's recognition has dropped — and the trust that came with that recognition has dropped with it.

the exact sequence of how voice drift happens

in our analysis of writing samples before and after periods of high-volume content production, voice drift follows a consistent sequence across writers on substack, linkedin, and medium:

1. sentence-length variation flattens first. a writer who naturally alternates between 6-word punches and 30-word constructions starts producing writing where most sentences land between 15 and 22 words. the rhythm becomes metronomic.
2. signature transitions disappear second. the characteristic phrases that signal "here comes the pivot" or "this is where the argument lands" get replaced by generic connectives — "however," "additionally," "in conclusion." the writing stops sounding like a person and starts sounding like a category.
3. vocabulary specificity degrades third. concrete nouns ("the 11pm email from a founder who had just lost his cfo") become category nouns ("a stressful business communication"). the writing becomes accurate but no longer particular.
4. cognitive pattern erodes last. the underlying logic move — the specific way the writer gets from observation to argument to conclusion — finally flattens into the standard problem-solution-cta framework that structures 80% of content on any given platform.

the loss of cognitive pattern is the hardest to recover from. by the time it has gone, the audience has usually already stopped recognising the writer as a distinct source. they may still read. they have stopped specifically seeking the writing out.

you can see how drift manifests in real newsletters and linkedin feeds in our detailed breakdown at the signs of voice drift and what causes them.

what does ai writing do to your brand voice?

used without discipline, ai writing tools are the fastest-known accelerant of voice drift. in our analysis of writers who use chatgpt or jasper to draft three or more consecutive pieces, we consistently find a 60-70% drop in sentence-length variation by the third post, and a near-total loss of signature transition phrases within five posts. the voice does not disappear overnight. it erodes post by post until the writing is statistically indistinguishable from the platform average.

why this happens at the mechanical level

chatgpt, jasper, and similar tools are trained on the average of written language. their default output represents the centre of the distribution — medium-length sentences, common transitions, general vocabulary. when you prompt these tools with your topic and a few style notes, they produce writing that is competent, readable, and anonymous. they replicate the surface layer of voice reasonably well if prompted with examples. they cannot replicate the structural layer because they are averaging, not learning from your specific fingerprint.

the damage compounds when writers start editing ai drafts rather than using ai to assist their own drafts. editing an ai draft means conforming your judgment to a structure that is not yours. writing your own draft and using ai to check for logic gaps or find a better word preserves your structure because the structure was yours from the start. the direction of influence matters enormously.

the grammarly problem

grammarly and the hemingway app introduce a secondary drift vector that most writers do not track. grammarly's readability suggestions systematically prefer shorter, simpler sentences and flag grammatical choices that are often deliberate voice signals. a writer who accepts grammarly's passive-voice warnings across an entire document will, over time, write more active and more generic sentences. hemingway's grade-level scoring pushes writers toward a shorter average sentence length that may not match their natural structural voice at all.

we are not arguing against these tools. we are arguing for using them selectively: let them catch errors, but override them on structural choices that are voice decisions, not mistakes.

what disciplined ai use looks like

the writers we have analysed who use ai tools without suffering measurable voice drift share a specific workflow pattern:

• they write a first draft in their own voice before opening any ai tool
• they use ai to generate alternatives for a specific sentence or paragraph — not to draft entire sections
• they run a voice profile score on the final draft before publishing to check that the structural signals are still present
• they never use ai to write the opening or closing of a piece, because those are the highest-density voice moments

our ai drift detector flags the specific structural signals that ai writing erodes fastest, so you can check whether a draft that passed through ai tools has retained your characteristic rhythm before it goes live.

how do you measure brand voice consistency?

voice consistency is measurable. it is not subjective. the subjective experience of "this sounds like me" maps reliably onto a set of quantifiable signals. here is what to track.

sentence-length variation (slv)

calculate the standard deviation of sentence length (in words) across a piece. a higher standard deviation means more rhythm variation — the hallmark of most distinctive writing voices. a standard deviation below 6 words is the threshold we use at hold your voice to flag a piece as potentially flattened. paul graham's essays typically show an slv standard deviation above 12. most ai-generated content lands between 4 and 7.

vocabulary specificity ratio (vsr)

count concrete nouns (nouns that name a specific thing: "the 6am cold call," "a $40,000 chargeback," "the third paragraph") versus category nouns (nouns that name a type of thing: "a stressful situation," "a financial challenge," "the key section"). divide concrete by total. a writer with strong voice typically runs a vsr above 0.65. ai-generated content frequently lands below 0.40.

transition fingerprint

list the ten most frequent multi-word transitions in your established writing. track whether those same transitions appear in new drafts at a similar frequency. a transition fingerprint is one of the most stable and distinctive elements of individual voice — and one of the first to disappear when voice drift sets in.

structural pattern adherence

does the new piece open and close in the structural pattern your voice establishes? does the logic sequence — the move from entry to argument to conclusion — follow your characteristic cognitive pattern? this is harder to quantify but can be assessed with a rubric drawn from your ten strongest pieces.

hold your voice builds a quantitative profile from your own writing across all four dimensions and scores every new draft against it, highlighting the exact passages where drift has occurred and which signal has dropped. try it with the brand voice analyzer on a piece you wrote six months ago versus one you wrote last week — the comparison is often clarifying.

what does strong brand voice actually look like in practice?

abstract principles only travel so far. here are three specific voice profiles, with the structural analysis that makes each one identifiable — and what happens to them when voice drift sets in.

paul graham — the deliberate simplifier

graham's structural fingerprint: short declarative opening sentence (rarely more than ten words). three to five sentences of precise qualification that narrow the claim. a single pivot where he introduces the counterintuitive observation. a closing section that states the implication flatly, without softening. his vocabulary tier is deliberately below his intellectual tier — he uses plain words to describe complex observations, which creates the characteristic tension that makes his essays feel both accessible and dense.

what drift looks like in a graham-imitating piece: the opening sentence gets longer and more qualified. the pivot becomes a "however" instead of a structural break. the closing becomes a paragraph instead of a sentence. the plain vocabulary gets replaced by academic register words that "match the idea." the voice evaporates even as the intelligence remains.

alex hormozi — the numerical aggressor

hormozi's structural fingerprint: a specific number in the first line. parallel constructions that build to a crescendo. a personal story told in past tense with present-tense urgency ("i was broke. i had $50 left. i made one call."). a direct second-person pivot that shifts responsibility to the reader. his vocabulary tier is high-specificity at the concrete level and low-theory at the abstract level — he names numbers, names people, names places, but rarely names concepts.

the drift pattern for writers trying to work in a hormozi-adjacent voice: they keep the numbers but lose the parallel construction. they add softeners ("this might not be for everyone"). they move from past-tense story to present-tense advice without the pivot. the voice flattens into motivational content that could have been written by anyone.

ben settle — the adversarial emailer

settle's structural fingerprint: one idea per email, stated in the subject line without disguise. an opening provocation that names the enemy explicitly. a two-to-three paragraph development that uses the enemy's logic against them. a selling line embedded in the content, not appended to it. his sentences run short and flat — rarely subordinated, rarely qualified. he writes as if being overheard, not as if presenting.

what drift looks like: the opening softens from provocation to observation. qualifiers appear. the selling line gets moved to a separate postscript. the email starts to "teach" instead of argue. it becomes useful and forgettable instead of polarising and memorable.

how do you protect brand voice as your team or output scales?

voice protection at scale has three failure modes, each requiring a different intervention.

failure mode 1: the single-gatekeeper bottleneck

a founder writes all the content. the voice is perfect. the company hires a content team. the founder reviews everything and becomes the bottleneck. eventually they stop reviewing. the voice drifts to the team average. the fix is not "better briefing" — it is externalising the voice judgment so it does not require the founder's time. a quantitative voice scoring system (not a style guide, a scoring system) is the only scalable solution to this problem.

failure mode 2: the platform-specific dilution

a brand with a strong newsletter voice expands to linkedin. linkedin's native content conventions (three-line paragraphs, engagement-bait questions, follower-first framing) gradually overwrite the original voice. the linkedin content starts sounding like linkedin content rather than like the brand. the same happens when a substack writer starts repurposing for convertkit email sequences — the email format conventions drift into the sequence and the voice goes with them. the fix is format-specific voice examples in the guide, showing how the structural patterns apply within each platform's container without being overwritten by it.

failure mode 3: the ghostwriter handoff

a ghostwriter, however skilled, starts from zero voice information. the standard handoff is: three sample pieces, a brief call, and the adjective-based voice guide. the ghostwriter reads the samples, extracts what they notice, and writes in the voice they perceive. what they perceive is mostly surface layer — vocabulary, punctuation style, paragraph length. the structural and cognitive layers require explicit documentation to survive a handoff. we have found that ghostwriter-produced content shows measurable voice drift from the founder's established profile within the first two pieces when given only sample + adjective guides, versus staying within profile for four to six pieces when given a structurally documented guide with named patterns and before/after examples.

for a full process on scaling voice without losing it, see how to maintain brand voice across all content.

the tool stack that protects voice at scale

the combination that works, based on our analysis of the teams with the most stable voice profiles over twelve months:

• notion or similar: houses the voice guide — but structured as named patterns with examples, not adjective lists
• hold your voice: scores every draft against the voice profile before publishing, flags specific drift signals
• a weekly voice audit: one person reads the last five published pieces purely for structural pattern compliance — not quality, not performance, just voice

the platforms where voice protection matters most — substack, convertkit for email sequences, linkedin — all have platform-specific drift pressures that need naming in the voice guide. you can run our brand voice analyzer to see where your content is drifting toward platform conventions and away from your own structural patterns.

ai

shashank

founder of hold your voice. writes about brand voice, ai writing patterns, and the craft of sounding like yourself.

co-written with ai as sidekick. shashank drafts the voice; the ai pressure-tests the structure. anything that sounds wrong is shashank's fault — anything that sounds suspiciously generic is the ai's.