The task
You read the landing page draft Monday morning, the launch email Monday afternoon, and the help-center FAQ on Tuesday — and they sound like they came from three different companies. Three people wrote them and you have no document to point them to. The 27-page brand book your previous agency wrote is collecting dust because nobody has time to read it before writing a tweet. You want a 1-page tone guide tight enough that a new copywriter writes “on brand” by the end of their first week — and that the existing team will actually look at before publishing.
Where AI helps — and where it does not
AI is excellent at distillation: take 10 of your real sentences and 3 that you flagged as “not us,” and the model can reverse-engineer a voice profile that explains the difference. It is also good at producing the do/do-not vocabulary lists that catch 80% of off-brand drift. Where AI fails: deciding where your brand should sit on the tone spectrum if you have not anchored it. Without 5 sample sentences you genuinely love, the model defaults to “professional but approachable” — the most useless tone guide ever written. Anchor with real samples from your own writing; the model finds the pattern, you make the judgment call.
A common failure mode: the model outputs a guide full of “X but also Y” balances — “casual but professional,” “confident but humble,” “playful but trustworthy.” These read smart and produce zero behavior change because writers can justify any sentence as “the balance point.” Tell the model: every slider must pick a side, and every “but” must die.
What to feed the AI
- 5 sample sentences you wrote that feel perfectly on-brand — emails, landing pages, social posts, FAQ answers; mix sources
- 3 sample sentences from your own past content that feel not us (the ones that make you wince); these are gold
- 3 competitor sentences in your category that you actively want to sound different from
- The audience you write for in one sentence — who, what they care about, what they do for a living
- The action you want them to take after reading you — sign up, click, share, buy, refer
- The 3 strongest emotions you want a reader to feel — calm, sharp, witty, reassured, energized
- The platforms you publish on — Twitter and the help center demand different lengths, your voice slider can shift slightly between them
- The 2-3 phrases you are tired of seeing in your category (“revolutionary,” “next-gen,” “delight”)
Copy-ready prompt
Distill a 1-page brand tone guide from the samples below.
On-brand samples (the voice we want): {paste 5 sentences}
Off-brand samples (the voice we want to avoid, from our own past): {paste 3 sentences}
Competitor sentences we want to sound *different* from: {paste 3 sentences}
Audience: {one sentence}
Action we want readers to take: {one specific action}
Top 3 emotions we want readers to feel: {list}
Platforms we publish on: {list}
Category clichés we are tired of: {list}
Return:
1) Tone position on 4 sliders, with a number from 0 to 10 each and a one-sentence rationale per slider. The 4 sliders: formal/casual, serious/playful, warm/distant, simple/sophisticated. Pick a side; do not write "balance."
2) 3 voice principles. Each principle gets a one-sentence statement + 1 do-example + 1 do-NOT example pulled from my samples.
3) 10-word "use these" vocabulary list — verbs and nouns we should prefer.
4) 10-word "do-not use" list — including the category clichés I gave you.
5) Edit-this-sentence exercise: take this generic line — "We're excited to announce our newest features that empower users to do more." — and rewrite it twice, on-brand. Add one-sentence justification per rewrite.
Format: 1 page. No "but," no "balance," no "professional yet approachable." If a slider position has a "but" in the rationale, you have not finished thinking.Shorter variant — voice audit on existing copy
Audit the copy below against this brand tone guide: {paste 1-page guide}.
Copy: {paste 5-10 sentences}.
For each sentence: rate 1-10 on each of the 4 voice sliders, and flag any phrase that violates the "do not use" list.
Then rewrite the 3 weakest sentences on-brand. Briefly explain what made them off-brand.Sample output
A useful slider rationale (no “but”): “Casual: 7/10. We use contractions (‘we’ll,’ ‘don’t’), short sentences, and verbs that do real work (‘ship,’ ‘pull,’ ‘drop’ over ‘release,’ ‘extract,’ ‘announce’). We never use slang we’d be embarrassed to read in 3 years.”
A useful voice principle: “Specific over impressive. Do: ‘cut your onboarding email queue by 80% in week one.’ Do NOT: ‘transform your customer engagement with AI-powered automation.’ We earn trust by saying what actually happens, not by sounding sophisticated.”
A useful do-not list: “Avoid: revolutionary, next-gen, robust, leverage (as verb), delight, empower, supercharge, unlock, seamlessly, world-class, innovative.” Each of these gets replaced by the underlying specific thing it was trying to claim.
A useful rewrite of the generic line: Original — ‘We’re excited to announce our newest features that empower users to do more.’ Rewrite 1: ‘Three new things shipped today. Here’s the one we’re most excited about.’ Rewrite 2: ‘You can now [specific feature]. The full changelog is here.’ Both rewrites refuse the meaningless verbs (‘excited to announce,’ ‘empower’) and lead with the specific.
How to refine
- If the guide reads as “casual but professional” (useless): “Kill every ‘but.’ Each slider picks a side. If two sliders feel contradictory in practice, write 5 test sentences that satisfy both and rate them.”
- If the do-not list misses obvious offenders: “Add: every verb I never want to see (‘leverage,’ ‘empower,’ ‘unlock,’ ‘supercharge,’ ‘delight’). These are the words my team writes when they have nothing specific to say.”
- If the voice principles are too abstract: “Each principle needs a do-example AND a do-NOT example pulled from sentences I gave you. If the example does not come from my samples, the principle is too generic.”
- If new hires still sound off-brand: “Add a 5-sentence ‘baseline drift test’ — 5 representative sentences a new writer rewrites in week 1; I evaluate against the guide and flag where they drift.”
- If the guide grows beyond one page: “Cut to one page. Anything that survives is signal. Vocabulary lists belong; meta-commentary about how to use the guide does not.”
Common mistakes
- Adopting a voice that does not match how you actually speak: the guide gets ignored after week 2 because writing in a borrowed voice is exhausting; pick the voice that is a sharpened version of how the founder or strongest writer actually writes.
- Skipping the off-brand list: half the value of a tone guide is what NOT to do; without examples of off-brand, writers cannot self-correct.
- Letting AI pick the voice for you: the model finds patterns in your samples; it does not decide the brand. If you have not made the call, the guide will be generic.
- Writing a 27-page brand book: nobody reads it; nobody references it; the document might as well not exist. One page or it does not get used.
- No specific do-not vocabulary: abstract principles like “be specific” are unenforceable; “never use these 10 words” is.
- Updating quarterly: voice drift is real but quarterly updates train your team to assume the rules will change; annual updates with clear changelog notes are better.
- No example of an edited sentence: without before/after, the guide is theory; the rewrite exercise is what makes the principles concrete.
- One slider position for all platforms: your help-center FAQ and your launch tweet can sit at different points; the guide should call out the platform-specific shift.
FAQ
- How long should the tone guide be?: One page. Two pages and your team skips it. If your principles cannot fit on one page, they are not tight enough; sharpen, do not expand.
- How often should I update it?: Annually. Voice does shift — usually following audience growth or product evolution — but quarterly updates train the team that rules are unstable, and stability is what makes the guide usable.
- Should each platform have its own variant?: One master guide + a small “platform notes” appendix is the right shape. The voice principles stay; the slider position can shift slightly (casual a notch higher on Twitter, simple a notch higher in product copy).
- What about voice for technical content (API docs, error messages)?: Voice principles apply; vocabulary changes. Tighter, more precise, fewer adjectives. Have a separate vocabulary list for technical writing but the same 4 sliders.
- How do I get the team to actually use it?: Two moves: (1) include the do-not list in your weekly editorial review so off-brand words get flagged in PRs / drafts, and (2) every new hire’s first week ends with a 5-sentence rewrite exercise against the guide. The guide enters muscle memory through that exercise, not through being “read.”