When we talk about brand tone of voice, most people nod along — and then write a brief that describes it as “friendly but professional.” That’s not tone of voice. That’s a contradiction with a comma in the middle. The confusion exists because the terms brand voice and tone of voice are used interchangeably, when they actually describe two different things — and conflating them is exactly why AI tools produce copy that sounds plausible but never quite sounds like you. At Choco Media, we’ve found that getting this distinction right is one of the most practical things a brand can do before handing any writing task to an AI model.
This post is for brand managers, content leads, and founders who are about to build or refine a set of brand guidelines — and want those guidelines to actually work inside a prompt, not just look good in a PDF. By the end, you’ll have clear definitions, a structure for each, and the specific fields that translate brand voice decisions into AI-readable instructions.
We’ll also look at where most brands go wrong when they try to document these things, and how to avoid the two most common traps.
Brand voice is what you always sound like
Brand voice is the stable, consistent personality of your brand in language. It doesn’t change based on context. It’s not situational. It reflects who you are — your values, your perspective, and the specific way your company sees the world.
Think of brand voice as the person. A person has a character. They might be blunt, or warm, or precise. That character doesn’t disappear when they’re at a funeral versus a barbecue — it adapts, but the core personality stays recognisable. Brand voice is that core personality.
- It’s expressed in word choice: does your brand say “use” or “utilise”? “help” or “support”? “get” or “obtain”?
- It’s expressed in sentence structure: short and punchy, or measured and deliberate?
- It’s expressed in perspective: first-person (“we think”), second-person imperative (“do this”), or neutral observer?
- It’s expressed in what you don’t say: do you avoid corporate jargon? Do you never use superlatives? Do you never make a claim you can’t back up?
These traits are documented once and applied everywhere — ads, onboarding emails, error messages, blog posts, sales decks. They’re non-negotiable. When a new team member starts writing for your brand, they learn the voice. When you brief an AI tool, you give it the voice parameters first.
Tone of voice is how that personality adapts to context
Tone of voice is the modulation of brand voice across different situations. Same personality, different register. The underlying character doesn’t shift — but the energy, warmth, or formality does, depending on who you’re talking to and why.
A brand with a direct, confident voice will write a sales email differently than they’ll write a payment failure notification. The voice is the same. The tone adapts.
- In a social caption promoting a new feature: energetic, concise, light
- In a terms of service update email: direct, clear, no jargon, slightly warmer than legal boilerplate
- In a customer complaint response: calm, owning accountability, not defensive
- In a long-form blog post: deliberate, evidence-based, direct without being cold
Tone is the how; voice is the who. And both need to be documented separately, or your guidelines become noise — a list of adjectives that nobody agrees on when it comes to execution.
Where most brand guidelines fail
Most brand voice documents list three to five adjectives — “bold, human, smart” — and leave it there. These are useless without context. Bold compared to what? Human as opposed to robotic? Smart as opposed to shallow? Without examples, without contrast, without the “we do / we don’t” structure, adjectives produce inconsistency. Every writer interprets them differently. Every AI tool interpolates between them and returns something generic.
Why this distinction matters more now than it did five years ago
Five years ago, brand guidelines primarily served human writers. A skilled copywriter could read between the lines of a vague brief, absorb the style from existing copy, and produce something that felt right. They had context. They understood nuance implicitly.
AI tools don’t have that implicit understanding. They have what you put in the prompt. If you give an AI a vague adjective like “approachable,” it will produce something that is the statistical average of approachable content across all the text it was trained on. That’s not your brand. That’s everyone’s brand averaged out.
This means brand guidelines written for human writers need to be reformatted before they’re useful for AI. Not because the underlying content changes — but because the delivery mechanism is different. Humans read for feel; AI reads for pattern. You need to give patterns.
- Instead of “we sound confident” → “we make declarative statements; we avoid hedging phrases like ‘might’, ‘could potentially’, ‘it’s possible that'”
- Instead of “we’re direct” → “sentences under 20 words when possible; no passive voice; active verb in every sentence”
- Instead of “we’re warm but professional” → “we use first-person plural ‘we’; we address the reader as ‘you’; we don’t use first names in headers; no exclamation marks”
A brand voice document that can’t be copy-pasted into a system prompt and produce consistent output is a brand voice document that hasn’t been finished yet.
The metadata fields that make brand voice AI-readable
When we build brand guidelines for clients, we include a set of machine-readable fields alongside the prose explanation. These are the parameters that go into a prompt and produce consistent output. They function as structured metadata for the brand’s voice.
Core voice metadata
- Person: first-person plural (“we”), second person (“you”), third person narrator
- Register: conversational / semi-formal / formal
- Sentence complexity: short (under 18 words average) / medium / long form with subordinate clauses
- Punctuation rules: no em-dash in body copy; Oxford comma required; no rhetorical questions in headlines
- Vocabulary tier: plain English (8th grade reading level) / industry-standard terminology allowed / technical terms with definitions
- Avoid list: specific words or patterns that break voice (“leverage”, “seamless”, “game-changer”, “unlock”, “delve”)
- Claim policy: no superlatives without proof; no statistics without source; no case study claims without approval
Tone matrix — context-specific parameters
For each key content type, add a tone modifier row. This is the tone-of-voice layer, mapped to contexts:
- Error messages: calm, owning, solution-first. No “oops.” No “whoops.”
- Promotional emails: direct, benefit-led, one clear CTA. No countdown urgency language.
- Social captions: punchy, one idea per post, 10% more energy than blog register
- Blog posts: deliberate, evidence-based, structured argument. Cite sources. No padding.
- Customer support: calm, first-person accountability, no deflecting to “our team” as a passive entity
When this matrix exists, briefing an AI becomes mechanical: specify content type, paste the corresponding tone parameters, paste the core voice metadata, write the prompt. The output requires far less editing — in our experience, often just a single pass for factual accuracy.
How to build this if you’re starting from scratch
If your brand guidelines are currently a list of adjectives in a slide deck, here’s a practical sequence for rebuilding them in a format that works for both humans and AI tools. Our AI content creation service involves this process for every client onboarding.
- Collect 10-15 pieces of existing copy you’re proud of. Blog posts, emails, ads — things that felt right when you wrote them. These are your ground truth.
- Find the patterns. What sentence structures repeat? What words appear often? What words are conspicuously absent? What claims do you make, and how do you back them up?
- Draft the avoid list first. It’s faster to say what the brand doesn’t do than to articulate what it does. Start negative, then flip each item into a positive rule.
- Write 3 example pairs per content type: “this, not that.” Short examples are worth more than long explanations.
- Test in a prompt. Paste your metadata + one example pair + a writing task into an AI tool. Read the output critically. What still sounds off? That’s a gap in your documentation.
- Iterate once. Most brands need one round of prompt-testing before the guidelines are genuinely useful for AI-assisted production.
Common mistakes when documenting tone of voice
Beyond the adjective trap, there are two other patterns we see regularly that undermine otherwise solid brand work.
Mistake 1: Not separating audience segments
Many brands have more than one audience — and those audiences often need meaningfully different tone approaches even if the voice stays constant. A B2B SaaS brand talking to a CFO in a case study and talking to a developer in a product blog is the same voice, different register. If your guidelines treat all audiences as one, the tone matrix will either be too conservative (for your energetic channels) or too loose (for your formal ones).
The fix: add an audience dimension to your tone matrix. Two or three rows is usually enough. Define the register shift, not the full brand voice — that’s already constant.
Mistake 2: Conflating brand voice with content strategy
Brand voice governs how you say things. Content strategy governs what you say and when. These are often written in the same document and then confused. The result is guidelines that include strategic directives (“we focus on education over promotion”) inside the voice section, and then voice instructions bleed into the content calendar. Neither document ends up doing its job cleanly.
If you’re building a social strategy alongside your brand voice work, keep the documents separate and reference each other explicitly. Voice guidelines link out to content strategy for topic decisions. Content strategy links to voice guidelines for execution. Each document does one job.
Applying this inside an AI workflow
Once the two documents — brand voice metadata and tone matrix — exist, the workflow for AI-assisted content production becomes structured rather than experimental.
For each content piece:
- Identify the content type (blog / email / caption / etc.)
- Pull the relevant tone-matrix row
- Combine with the core voice metadata
- Add the specific brief (topic, angle, target keyword, word count, internal links needed)
- Run the prompt
- Edit for facts, not for voice — the voice should already be right
This is what we mean when we talk about building an AI content system rather than using AI reactively. The system does the repetitive work; the human focuses on editorial quality and strategic judgement. Our AI automation work builds these structured workflows for clients who want consistent output without constant oversight.
The upfront investment in clean documentation is maybe 4-6 hours for a mid-sized brand. The time it saves across every piece of content produced afterward compounds quickly.
What to do right now
If you don’t have separate brand voice and tone of voice documentation, start with one hour of collection: pull your 10 best pieces of past copy, open a blank document, and list every pattern you notice. Don’t write guidelines yet — just observe. The rules emerge from the patterns, not the other way around.
If you already have guidelines but they’re adjective-heavy, run a test: copy them into a prompt, ask an AI to write a short product description in your brand voice, and read what comes back. If it sounds generic, your guidelines are generic. That’s the diagnostic.
If you want to work through this with someone who does it regularly, we’re easy to reach — send us a note and we’ll take a look at what you’ve got.