⚡ Promptolis Original · Marketing & Content

🎨 Brand Voice Builder

Paste 3-5 samples of your writing. Get a voice spec sheet so precise that anyone on your team — or any AI — can write as you without sounding like a parody of you.

⏱️ 12 min to try 🤖 ~90 seconds in Claude 🗓️ Updated 2026-04-19

Why this is epic

Most 'brand voice guides' are vague poetry ('warm but direct!') that nobody can execute. This one extracts surgical patterns: your actual sentence-length distribution, your three most-used transitions, the words you never use, and the rhetorical moves that make your writing recognizable.

It separates the signature moves (do more of these) from the tics (cut these before they become a caricature). Most brand voice work conflates the two and turns teams into bad impressionists of the founder.

The output is structured so an AI can actually use it. No hand-wavy 'be authentic' advice — it's a spec sheet with rules, examples, anti-examples, and a do-not-do list that a model can follow verbatim.

The prompt

Promptolis Original · Copy-ready
<role> You are a voice analyst who has spent 15 years studying how distinctive writers actually sound on the page. You reverse-engineer voice the way a sommelier reverse-engineers wine: you name specific compounds, not vibes. You are allergic to phrases like "warm but professional" and "authentic yet approachable" because they mean nothing a writer can act on. </role> <principles> 1. Specificity over vibes. Never write "conversational tone." Instead: "Average sentence length 11 words, uses sentence fragments 2-3x per paragraph, opens ~30% of paragraphs with a one-word sentence." 2. Separate signature moves from tics. Signature moves are what makes the writer recognizable and should be amplified. Tics are unconscious habits that, when imitated by others, turn the voice into a caricature. Flag both, explicitly. 3. Name the anti-voice. What would this writer NEVER say? What words, constructions, and postures are off-limits? The negative space defines voice as much as the positive. 4. Quote the evidence. Every claim about the voice must be backed by a direct quote from the samples. No claim without citation. 5. Write rules an AI can execute. The final spec should be usable as a system prompt. If a rule can't be followed by a model reading it fresh, rewrite it. 6. Don't flatten. If the writer contradicts themselves (formal in one piece, casual in another), don't average them — identify the contexts that trigger each mode. </principles> <input> Writing samples (3-5 pieces, ideally a mix of long and short form): {PASTE 3-5 SAMPLES HERE, separated by "---"} Context about the writer/brand (optional but helpful): {WHO THEY ARE, WHO THEY WRITE FOR, WHAT MEDIUMS} </input> <auto-intake> If the samples are missing, fewer than 3, or the placeholders are unfilled, do NOT proceed. Instead, ask conversationally: 1. "Can you paste 3-5 samples of your writing? I need a mix — ideally one long-form piece (essay, about page, or 500+ word post) and a couple of short-form pieces (emails, tweets, captions)." 2. "Who is the writer and who do they write for? One sentence is enough." 3. "Any pieces you want me to EXCLUDE from the voice (e.g., co-written, heavily edited, or writing you don't want more of)?" Wait for all three before proceeding. </auto-intake> <output-format> # Voice Spec Sheet: [Writer Name] ## The One-Paragraph Summary [A single paragraph that captures the voice so precisely that someone who has never read the writer could, after reading this, recognize their next piece. No vibes — specifics.] ## Voice Fingerprint (The Measurable Stuff) | Dimension | Pattern | Evidence | |---|---|---| | Avg sentence length | [X words] | [quote] | | Paragraph length | [pattern] | [quote] | | Sentence fragments | [frequency] | [quote] | | Questions to reader | [frequency] | [quote] | | First-person usage | [pattern] | [quote] | | Formatting tics | [e.g., em-dashes, bolding, lists] | [quote] | ## The 5 Signature Moves (Do More of These) For each: name the move, give a direct quote, explain why it works, give a template. ## The 4 Tics (Do Less of These Before They Become a Caricature) For each: name the tic, quote it, explain why imitating it breaks the voice. ## What This Writer Would Never Say A bulleted list of specific words, phrases, and postures that are off-limits. Each with a one-line reason. ## Mode Switching: When Voice Changes [If the samples show different modes — e.g., louder on social, quieter in long-form — describe each mode and its triggers.] ## Which rules matter most for writing in this voice? Ranked list, top 3. If someone only remembers three things, these are the three. ## How should a new writer practice this voice? A 20-30 minute exercise using the spec. ## Ready-to-Use AI System Prompt A paste-ready block that encodes the spec as instructions for Claude or GPT. ## Key Takeaways 3-5 bullets summarizing the core of the voice. </output-format> Now, produce the voice spec sheet:

Example: input → output

Here's how this prompt actually performs. Real input below, real output from Claude Opus 4.

📝 Input
Writing samples:

---
Sample 1 (About page, ~180 words):
I started Linear because I was tired of project management tools that felt like filing taxes. Every tool I tried — Jira, Asana, the graveyard of others — treated software teams like they were tracking inventory in a warehouse. Ship dates as deadlines. Tickets as line items. It's no wonder engineers hate these tools. They weren't built for the work.

Linear is built on a different belief: that software is made by small teams who care, and the tool should disappear into the work. No bureaucracy. No status meetings held hostage by a dashboard. Just the shortest path from idea to shipped.

We're a small team. We ship weekly. We use Linear to build Linear, which is the only honest way to build a tool for builders.

---
Sample 2 (Tweet):
Most software is built for the person buying it, not the person using it. That's why enterprise tools are beautiful in demos and miserable in practice.

---
Sample 3 (Product launch email, ~200 words):
Three years ago, we said Linear would be the tool engineers actually wanted to use. Today we're shipping the thing that makes that promise real: Cycles.

A Cycle is a week or two of focused work. Not a sprint — sprints are a metaphor that broke the moment Scrum consultants got hold of it. A Cycle is just: here's what we're doing, here's when it's done, here's what slipped. No ceremony.

We've been using Cycles internally for eight months. Our shipped-per-week count is up 40%. Meetings are down. The thing we kept saying in retros — "this actually feels like we're building" — is what we want you to feel too.

Turn on Cycles in Settings. It takes 30 seconds. If you hate it, turn it off. We won't be offended.

—Karri

Context: Karri Saarinen, co-founder and CEO of Linear. Writes for software engineers and startup founders. Voice appears on the Linear blog, product announcements, and Twitter.
🤖 Output

Voice Spec Sheet: Karri Saarinen / Linear

The One-Paragraph Summary

A voice spec is a reverse-engineered set of rules that captures how a specific writer actually sounds on the page. Karri writes like a builder who has already had the argument in his head and is now telling you the conclusion. His sentences are short (avg 11.4 words across samples), often fragments, and structured as a thesis followed by the evidence that earned it — not a warm-up followed by a reveal. He quietly positions Linear against incumbents without naming them as villains, uses the first-person plural ("we ship weekly") to signal discipline rather than togetherness, and reaches for one concrete number or comparison in nearly every paragraph. He almost never uses marketing adjectives. When he's moved, he gets shorter, not louder.

Voice Fingerprint (The Measurable Stuff)

DimensionPatternEvidence
Avg sentence length11.4 words; range 3-26"No bureaucracy." (2 words) vs. "Every tool I tried — Jira, Asana, the graveyard of others — treated software teams like they were tracking inventory in a warehouse." (24 words)
Paragraph lengthShort: 2-4 sentences. Long paragraphs get broken by a one-line punch."We're a small team. We ship weekly."
Sentence fragments~2 per 100 words. Used as emphasis, not laziness."Ship dates as deadlines. Tickets as line items."
Questions to readerZero. He never asks; he asserts.
First-person usage"We" dominates (13x) over "I" (3x). "I" appears only in origin stories."I started Linear because…"
Formatting ticsEm-dashes for asides (4x). No bold, no bullets, no emojis in long-form."— Jira, Asana, the graveyard of others —"
NumbersEvery email/post contains at least one specific number. 100% of samples."shipped-per-week count is up 40%", "eight months", "30 seconds"

The 5 Signature Moves (Do More of These)

1. The Quiet Villain. Names the competitor or category dismissively, then moves on without dwelling. Quote: "Jira, Asana, the graveyard of others." Why it works: credibility through specificity without turning into a grudge post. Template: "[Category] — [named incumbent], [named incumbent], [dismissive plural] — [what they all get wrong]."

2. The Earned Conclusion. States the belief first, then the evidence that earned it. Quote: "software is made by small teams who care, and the tool should disappear into the work." Why it works: reads as conviction, not pitch. Template: "[Thesis]. [Concrete evidence]. [Consequence]."

3. The One Number Rule. Drops a specific metric or duration to anchor abstract claims. Quotes: "We ship weekly," "eight months," "up 40%." Why it works: each number makes the surrounding sentences land. Rule: never publish a paragraph about a product without at least one number.

4. The Dogfood Receipt. Casually mentions that Linear uses Linear, or the team does the thing they're selling. Quote: "We use Linear to build Linear." Why it works: proof without bragging.

5. The Permission to Leave. Closes with an exit, not a CTA. Quote: "If you hate it, turn it off. We won't be offended." Why it works: inverts the anxious marketer's posture. Template: "[Try it]. [How to undo]. [Low-stakes sign-off]."

The 4 Tics (Do Less of These Before They Become a Caricature)

1. The em-dash aside. Karri uses em-dashes well, but at scale, writers imitating him stack 3-4 per paragraph and the voice collapses into a tic. Cap: one em-dash aside per 150 words.

2. The "we" that flexes. "We ship weekly" is earned; "We believe in craft" is LinkedIn sludge. Don't copy the grammar without copying the specificity.

3. "Not X — Y" construction. Quote: "Not a sprint — sprints are…" Powerful once, performative at three uses in a piece. Limit: one per post.

4. The punchy fragment. "No bureaucracy. No status meetings." When imitators do this, it reads like a bad trailer. Only use fragments when the full sentence would dilute a conclusion you've already earned.

What This Writer Would Never Say

  • *"Delighted to announce"* — he announces by describing the thing, not his feelings about it.
  • *"Game-changing," "revolutionary," "best-in-class"* — no marketing adjectives, ever.
  • *"What do you think?"* — he never asks the reader to validate him.
  • *"Our amazing team"* — team praise is implied through "we," never stated.
  • *"At [Company], we believe…"* — the corporate throat-clear. He starts with the belief.
  • Emoji in long-form. Exclamation points outside of a rare genuine moment.
  • Any sentence longer than 30 words.

Mode Switching: When Voice Changes

Two modes detected across samples:

  • Long-form (about page, product launch): More narrative, uses origin-story "I," em-dashes, 3-4 sentence paragraphs. Still no adjectives.
  • Short-form (Twitter): Single-idea, thesis-plus-evidence, often one compound sentence. No "we" — short-form Karri is observational, not institutional. Trigger: platform length.

Which rules matter most for writing in this voice?

1. Short sentences, specific numbers. If you nail nothing else, nail this. Avg sentence under 13 words; at least one number per 150 words.

2. Thesis first, evidence second. Never warm up. State the belief, then earn it.

3. No marketing adjectives, ever. Strip every "powerful," "seamless," "intuitive" before hitting publish. If the sentence collapses without the adjective, rewrite the sentence.

How should a new writer practice this voice?

The 25-minute exercise:

1. (5 min) Pick any Linear feature. Write a 100-word launch post the way you'd normally write it.

2. (10 min) Edit using only three rules: (a) cut every adjective, (b) add at least one specific number, (c) break any sentence over 18 words into two.

3. (5 min) Add a Signature Move: either a Quiet Villain line or a Permission to Leave closer.

4. (5 min) Read it aloud. If any sentence sounds like a press release, delete it.

Ready-to-Use AI System Prompt

You are writing as Karri Saarinen, co-founder of Linear, for an audience of engineers and startup founders. Follow these rules without exception:

1. Average sentence length under 13 words. Hard cap at 30.
2. Every 150 words must contain at least one specific number, duration, or comparison.
3. No marketing adjectives. Banned: powerful, seamless, intuitive, delightful, amazing, game-changing, revolutionary, best-in-class, world-class.
4. Structure: thesis first, evidence second. Never warm up.
5. Use "we" for institutional claims, "I" only for origin stories. Never "I" for opinions.
6. Em-dashes allowed for asides, max one per 150 words.
7. No questions to the reader. No exclamation points. No emoji in long-form.
8. When closing a product post, offer an exit ("if you hate it, turn it off"), not a CTA.
9. If naming competitors, do it once, dismissively, then move on.
10. If a sentence could appear in any company's blog post, delete it.

Before returning the draft, audit it against these rules and list any you broke.

Key Takeaways

  • The voice is conviction-plus-evidence, not warmth. Karri sounds like someone who already won the internal argument.
  • Numbers do the emotional work that adjectives would do in a weaker voice. "40%" and "eight months" are load-bearing.
  • Negative space defines the voice. What he refuses to say (marketing adjectives, questions, exclamation points) matters more than any single signature move.
  • The most imitable move is the Permission to Leave. It's the one rule that instantly makes any marketing copy sound more like Linear.
  • Don't stack the tics. Three em-dashes, two "not X — Y" constructions, and four fragments in one post is not "writing like Karri" — it's wearing a Karri costume.

Common use cases

  • Onboarding a new writer, marketer, or ghostwriter to your voice
  • Building a custom GPT or Claude project that drafts in your voice
  • Founders scaling content without every post having to route through them
  • Agencies documenting a client's voice after a discovery phase
  • Solo creators who want consistency across newsletter, LinkedIn, and podcast show notes
  • Rewriting AI-generated drafts to sound like you instead of ChatGPT
  • Detecting voice drift when your content starts feeling 'off'

Best AI model for this

Claude Sonnet 4.5 or GPT-5. Claude has a slight edge because voice analysis rewards careful reading of rhythm and subtext; GPT-5 is great if you want punchier rule summaries. Avoid smaller models — they flatten distinctive voices into generic 'professional friendly.'

Pro tips

  • Feed it your BEST writing, not your average. The spec gets calibrated to whatever you paste, so paste the stuff you're proud of.
  • Include at least one long-form piece (essay, about page) and one short-form (tweet, email subject line). Voice shows up differently at different lengths.
  • Don't include co-written or heavily edited pieces. The model will pick up the editor's patterns and credit them to you.
  • After you get the spec sheet, test it: ask Claude to write something new using only the spec, then compare. If it feels like a bad cover band, tell the model which rules to downweight.
  • Re-run it every 6-12 months. Voices evolve, and an old spec will slowly turn your content into a museum of your past self.
  • Paste samples in plain text. Formatting like bold, bullets, and emojis is part of your voice — keep it intact.

Customization tips

  • Replace the sample writing with your own. 3-5 pieces minimum. If you only have 1-2 pieces, the spec will overfit — go write a couple more first.
  • If you have a co-founder or team member whose voice is blending with yours in the samples, exclude co-written pieces. The model can't tell whose patterns are whose and will credit both to you.
  • After you get the spec, paste it back into Claude with a new writing task and say 'draft this using only the spec.' Compare the draft to your real writing. Whatever feels off is usually a rule that needs softening or a tic the model overweighted — tell it which one to tune down.
  • If you maintain multiple voices (personal vs. company account, newsletter vs. LinkedIn), run this separately for each. Don't try to merge them into one spec — you'll get mush.
  • Save the 'Ready-to-Use AI System Prompt' block as a Claude Project or custom GPT. That's where the compounding value lives: every future draft starts 80% in your voice by default.

Variants

Team Onboarding Mode

Output is formatted as a Notion-style internal doc with a 30-minute exercise for new hires to practice the voice.

AI System Prompt Mode

Output is a ready-to-paste system prompt for Claude/GPT projects, optimized for drafting in your voice.

Voice Drift Audit

Paste old samples AND recent samples. Get a diff showing how your voice has changed — intentional evolution or unintentional drift.

Frequently asked questions

How do I use the Brand Voice Builder prompt?

Open the prompt page, click 'Copy prompt', paste it into ChatGPT, Claude, or Gemini, and replace the placeholders in curly braces with your real input. The prompt is also launchable directly in each model with one click.

Which AI model works best with Brand Voice Builder?

Claude Sonnet 4.5 or GPT-5. Claude has a slight edge because voice analysis rewards careful reading of rhythm and subtext; GPT-5 is great if you want punchier rule summaries. Avoid smaller models — they flatten distinctive voices into generic 'professional friendly.'

Can I customize the Brand Voice Builder prompt for my use case?

Yes — every Promptolis Original is designed to be customized. Key levers: Feed it your BEST writing, not your average. The spec gets calibrated to whatever you paste, so paste the stuff you're proud of.; Include at least one long-form piece (essay, about page) and one short-form (tweet, email subject line). Voice shows up differently at different lengths.

Explore more Originals

Hand-crafted 2026-grade prompts that actually change how you work.

← All Promptolis Originals