Ask most marketing teams for their brand's HEX codes and they'll have them memorized. Ask for the exact stability and similarity settings on their cloned brand voice, and you'll get a blank look. That gap is the whole problem. Visual identity gets treated like law. Voice gets treated like a vibe.
It doesn't have to be this asymmetric, and honestly, it shouldn't be. A style guide, whatever form it takes, documents how your brand looks, sounds, and behaves, and is meant to reduce the cognitive burden on anyone producing content so they're not reinventing the voice from scratch every time. Visual identity has had decades to build that discipline — logo clear space, type scales, color codes down to CMYK. Voice is catching up, but the tools now exist to document it with the same precision. Here's how we build that document, and why it matters just as much as the one sitting in your design team's Figma file. For practical advice on saving specific preset values, read our guide on why we save every voice setting as a named preset.
Why Voice Guides Lag Behind Visual Guides
Part of it is tooling history. Color palettes have had HEX and Pantone codes for generations. Typography has had point sizes and kerning values. Voice, until recently, only had adjectives — "friendly," "confident," "playful" — which sound like guidance but function more like a Rorschach test. Two writers reading "confident but approachable" will produce noticeably different scripts.
The other part is that voice guides typically cover only written tone, and it's genuinely rarer to see the spoken voice — the actual audio identity — documented with the same rigor as a logo file. That's changed because AI voice cloning gave brands something visual identity has always had: a literal, exportable, versionable asset. A cloned voice model is not unlike a locked logo file. It's a fixed reference, and it can be documented with the same specificity as a HEX code.
The Visual Style Guide, Translated to Voice
The cleanest way to build a voice guide is to walk through what a visual guide documents and find its voice equivalent for each section. The pattern maps more directly than most teams expect.
| Visual style guide element | Voice style guide equivalent |
|---|---|
| Primary and secondary logo, with locked file versions | Primary brand voice model (cloned or human), locked and named — not "whoever's free" |
| HEX / RGB / CMYK color codes | Exact voice engine parameters: stability, similarity, style exaggeration values |
| Type scale (H1-H4, body, caption) | Delivery scale: narration pace and energy for long-form, mid-form, and 6-second formats |
| Do's and don'ts with visual examples | Approved and banned phrasing, with short audio or script examples |
| Photography style (lighting, color grading, mood) | Recording environment standard: mic distance, room treatment, noise floor |
| Logo clear space and minimum sizing rules | Minimum script length or context needed before the voice sounds "on-model" |
Just as brands document primary and secondary colors with precise hex and RGB values so nobody has to guess, a voice guide should document the literal settings a synthesis engine needs to reproduce your brand voice on command.
What Actually Goes in the Document
1. The locked voice reference If you're using a cloned voice through a platform like ElevenLabs, this section names the actual voice model ID, not a description of it. Two cloning tiers matter here: Instant Voice Cloning works from 1-3 minutes of clean audio and is fine for testing, while Professional Voice Cloning trained on 30 minutes to several hours of studio-grade recording produces output that's much closer to indistinguishable from the source speaker in blind listening tests. Your guide should state plainly which tier your brand voice was built on and where the source recordings live.
2. Exact synthesis settings This is the part most teams skip entirely, and it's the single biggest lever for consistency. ElevenLabs voice generation runs on a small number of parameters that behave predictably once you understand them:
| Parameter | What it controls | Typical range for brand narration |
|---|---|---|
| Stability | How consistent versus expressive each generation sounds — low values add natural variation, high values risk sounding monotonous | 0.40-0.55 for conversational brand voice, 0.60-0.75 for formal or IVR-style delivery |
| Similarity / Clarity boost | How closely output matches the cloned source voice | 0.75-0.85; pushing to 100% can over-enunciate |
| Style exaggeration | Amplifies the source voice's mannerisms | Kept near 0 for most brand use; ElevenLabs' own guidance recommends leaving this at 0 in general |
| Speaker boost | Reinforces similarity to source at a small latency cost | On for pre-recorded assets, off for real-time or IVR |
The most common starting point across ElevenLabs' own documentation is stability around 50%, similarity around 75%, with style kept at 0 — a sensible default to write into your guide before any brand-specific tuning. Document your actual numbers, not "medium settings," the same way you'd document an exact HEX code instead of "kind of blue."
3. Delivery pace by format A brand voice guide for a company running YouTube, ads, and social needs a stated words-per-minute or pacing note for each format, because a 10-minute YouTube video and a 6-second ad bumper are not the same performance. This is the voice equivalent of a type scale — H1 doesn't behave like a caption, and your long-form narration shouldn't behave like your bumper ad script.
4. Banned and preferred phrasing, with audio examples Written tone-of-voice guides typically already do this for text. The addition for a voice guide is pairing a few of these with a short reference clip, so a new editor can hear the difference between on-brand and off-brand delivery instead of just reading about it.
5. Recording standards, if you still record human narration alongside AI voice Quiet room, fixed mic distance, consistent pacing — the same physical discipline that produces a clean Professional Voice Clone source file also keeps human-recorded segments from clashing with AI-generated ones in the same project.
Why the Exact Numbers Matter More Than Adjectives
A useful mental model: stability and style exaggeration interact, and they don't average out neatly. High stability paired with high style exaggeration tends to sound stiff and over-performed at the same time, while a slightly lower stability paired with a light touch of style exaggeration reads as warm and natural. That's not something "sound friendly but professional" tells a producer how to achieve. A number does.
This is precisely the same reason visual brand guides gave up on "make it feel premium" in favor of documented type scales and locked spacing units decades ago — vague direction produces inconsistent output no matter how good the individual producing it is. Voice is catching up to that same standard now that the underlying tool exposes real parameters instead of just a recording booth and a voice actor's instincts.
Where This Breaks Down Without Enforcement
A style guide only works if people actually open it. Even in visual branding, a majority of companies have brand guidelines on paper, but only a minority actively use them day to day — and off-brand content shows up regularly even at companies that have documented standards. There's no reason to expect voice guides will fare any better without the same enforcement: someone assigned to check new assets against the reference, and a habit of updating the document the moment you notice drift rather than waiting for a quarterly audit to catch it.
This is also where it's worth building the check-in step into your workflow rather than treating it as a separate task — if your team is producing scripts, that's a natural moment to link back to your published brand standards and confirm nothing's drifted before recording starts.
How We Use This With Clients
When we onboard a new brand for voice work, the style guide isn't a deliverable we hand over at the end — it's the first thing we build, before a single script gets recorded. We document the locked voice model, the exact synthesis parameters, the pacing rules per platform, and the phrasing guardrails, and that document travels with every project afterward so a YouTube script, a paid ad, and a social clip all draw from the same source instead of three different interpretations of "sounds like us."
If your brand has a beautiful Figma file for visual identity and nothing equivalent for voice, that's usually the fastest gap to close — and it's exactly the kind of workflow our AI Voice & Animation service handles end-to-end, from the initial voice clone through the documented settings that keep it consistent project after project.
Summary
A visual style guide exists because "make it look like us" isn't instructions — it's a hope. Voice guides have lagged behind for years because the tools didn't offer the same precision that color codes and type scales gave designers. AI voice cloning changes that: a locked voice model with documented stability, similarity, and style settings is every bit as reproducible as a HEX code, and it deserves the same document, the same enforcement, and the same respect.
If you're ready to turn your brand voice from a vague description into an actual documented, reproducible standard, get in touch and we'll help you build the reference document your future scripts and recordings can actually be checked against.