“Write in our brand voice” is a useless instruction to an AI — and it’s why your output sounds like everyone else’s.
The model has no idea what your voice is. Generic output is almost always a documentation problem, not a model problem. Give an agent a real, concrete voice spec and the difference is night and day. Here’s exactly what goes in one — and the trick that makes it click.
What a usable voice doc contains
The “but not” trick
Adjectives alone are too vague — every brand claims “friendly and professional.” The contrast is what carries the signal: “Warm, but not cute. Direct, but not cold. Smart, but not academic.” Those boundaries tell the model where the edges are, which is the information it’s actually missing.
Test it in two minutes
Paste the doc, give the agent a deliberately boring sentence, and ask it to rewrite in voice. If the result sounds like you, the doc works. If not, it’s too abstract — add more concrete samples until it clicks. A good voice doc is reusable across every agent and prompt you’ll ever run.
A worked example: vague doc vs usable doc
The gap between a voice doc that works and one that doesn’t is easiest to see side by side. A vague doc says: “We’re professional, friendly and approachable. We speak to busy professionals.” Every brand on earth could have written that, and an agent given it produces the same beige output it always would.
A usable doc says: “Confident, but not arrogant. Warm, but not cute. Plainspoken, but not dumbed-down. We write short sentences. We use ‘you’ constantly and ‘we’ rarely. We never say ‘leverage,’ ‘unlock,’ ‘game-changer,’ or ‘in today’s fast-paced world.’ When in doubt, we sound like a smart friend explaining something over coffee, not a brand talking at a customer.” Hand that to an agent and the output shifts immediately, because every line gives the model a concrete boundary instead of a feeling.
The four inputs that move the needle most
If you only have ten minutes to build a voice doc, spend it on these four, in order. First, the three “but not” pairs — they carry more signal than anything else. Second, one real writing sample you’re proud of, because the model learns rhythm from example better than from description. Third, the do-not-use list — the specific words and phrases that instantly read as “not us.” Fourth, a single before/after rewrite showing a generic line fixed into your voice. Those four, even roughly done, beat a beautifully formatted ten-page brand bible the model will never internalize.
Keeping the voice consistent as you scale
A voice doc isn’t just for you — it’s the mechanism that keeps your brand sounding like one person even when it isn’t. The moment a second writer, a contractor, or a new agent touches your content, the doc is what prevents drift. This is exactly why documenting voice matters more, not less, as you grow: an undocumented voice that lives only in the founder’s head can’t survive delegation. Revisit the doc every few months, add any new “never say” phrases you’ve started noticing, and refresh the writing sample as your style sharpens. A living voice doc is one of the few brand assets that quietly compounds in value the bigger you get.
Common ways voice docs fail
How to extract a voice you’ve never documented
Most people freeze at this step because they’ve never put their voice into words — they just write and it sounds like them. There’s a fast way through. Take three pieces you’ve written that you genuinely like, paste them into a model, and ask it to describe the voice: the tone, the sentence rhythm, the recurring moves, the words that show up often. The model becomes a mirror, naming patterns you use unconsciously. You won’t agree with all of it — and that disagreement is gold. Crossing out “it says I’m formal, but I’m actually blunt” is exactly how you find your “but not” boundaries. Within twenty minutes of this back-and-forth you’ll have a first draft of a voice doc built from how you actually write, not from how you imagine you write. Then test it the usual way: hand it back, ask for a rewrite of a boring sentence, and refine until the output sounds unmistakably like you.
Where the voice doc pays off beyond AI
The doc earns its keep with agents, but its value spreads further than that. The same one page onboards a freelance writer in an afternoon instead of a month of awkward revisions. It settles internal arguments about whether a piece “sounds like us” by replacing opinion with a written standard. It keeps your social posts, your emails, and your landing pages sounding like one coherent brand instead of three different companies. And it forces a clarity about who you are that sharpens your positioning as much as your prose. Plenty of brands discover that the act of writing the voice doc — being forced to choose three adjectives and their boundaries — is the moment their brand identity actually crystallizes. The AI use case is the excuse; the clarity is the real prize.
Voice versus tone — and why the distinction matters
One nuance trips people up and quietly weakens their voice docs: confusing voice with tone. Your voice is constant — it’s who you are, the same whether you’re announcing good news or apologizing for an outage. Your tone shifts with context — warmer in a welcome email, more serious in a security notice, lighter in a casual social post. A voice doc that only captures voice produces output that’s recognizably you but tone-deaf to the moment; a doc that confuses the two ends up contradicting itself. The fix is to document voice as the fixed core — your “but not” adjectives, your do-not-use list, your rhythm — and then add a short note on how that voice flexes across a few common situations: “In apologies we stay direct and drop the humor; in launches we let the confidence show; in support we slow down and over-explain.” With that distinction written down, an agent can hold your identity steady while still reading the room — which is exactly what a skilled human writer does without thinking.