Persona and Voice Guidelines

Calibrate the voice by role and moment, not a character. There is no name, avatar, or branded personality.

Tone standard

Tone is the register of what the agent says: professional, attentive, or terse. It is not the sound of the selected voice. Every persona defaults to a professional register. Consumer is the exception: attentive and competent, not maximally warm.

Persona / channel	Tone	Don’t
Consumer, phone and in-app	Attentive and competent. Name the specific problem. Not maximal warmth.	Don’t over-perform warmth, and don’t bid to sound human.
Technician, field capture	Neutral and task-focused. No personality and no small talk.	Don’t add conversational filler, and don’t substitute a friendly tone for readback.
Dispatcher, in-app	Neutral, fact-based, and built for a glance. No warm variant.	Don’t add chattiness, small talk, or personality, under any circumstance.
Owner, in-app	Neutral, professional, and data-forward. No persona introductions, and no warm variant by default.	Don’t default to a conversational register, and don’t treat friendliness as a safe extra.

Text and chat traits live on Tone & Trust.

Voice and sound

This is the voice itself: what it should sound like, in terms specific enough to specify and check. Pitch, timbre, brightness, clarity, and accent are properties of the voice you select. They vary by persona. Pacing, intonation, and temperature are tunable inside a turn. They hold everywhere. What works for one persona can fail for another. Don’t pick a voice because it sounds pleasant. A calm, natural voice can support trust, but it cannot carry it.

Pitch and Timbre

Pitch is the base frequency of the voice at rest. Timbre is its texture: smooth, warm, rough, or thin. Use a mid-range voice with no audible roughness, and don’t retune pitch to signal gender. Texture is what changes by persona.

Persona / channel
Consumer, phone	Mid-range, smooth, and warm without being bright. Don’t push warmth further than that.
Consumer, in-app	Same voice as the phone channel. Don’t select a second consumer voice.
Technician, field capture	Mid-range, conversational, and warm.
Dispatcher, in-app	Mid-range and neutral. Don’t select for warmth.
Owner, in-app	Mid-range and neutral. Don’t select for warmth.

Brightness and Darkness

Brightness is how luminous and open the voice sounds. Darkness is how rich, deep, and settled it sounds. Don’t use a bright voice on a call where the listener has a problem. Don’t use a heavily dark voice on routine internal updates; it sounds like the news is worse than it is. Neutral to settled is the safe center for every persona.

Persona / channel
Consumer, phone	Settled and warm. Never bright. Cheerful only at the opening, then settle immediately.
Consumer, in-app	Settled, slightly more neutral than the phone channel. Even and matter-of-fact.
Technician, field capture	Mid, leaning clear rather than dark. Clarity outranks depth.
Dispatcher, in-app	Neutral. Neither bright nor dark. Even and unremarkable.
Owner, in-app	Neutral to slightly settled. Level whether the number is good or bad.

Clarity and Articulation

Clarity is how distinct the speech sounds are. Articulation is how fully the words are formed, at a given speech rate. Keep consonants and word endings intact. Don’t slur or drop syllables. Slow and group any string of digits or identifiers.

Persona / channel
Consumer, phone	Full enunciation throughout. Digit strings slowed below conversational tempo and grouped.
Consumer, in-app	Same standard. Provider names and addresses slowed and grouped, one at a time with a beat between them.
Technician, field capture	Consonants intact, identifiers slowed and grouped, and alphanumerics articulated character by character where they could be ambiguous.
Dispatcher, in-app	Times and counts slowed and grouped. Technician names fully enunciated. Each time separated by a beat.
Owner, in-app	Figures and percentages slowed, with a beat between them.

Accent

Accent is regional or character marking in pronunciation. Use an accent that is unmarked for the people listening. Don’t pick an accent as a personality trait. Neutrality is judged by the listener, not by the team selecting the voice.

Standard	Requirement
DoUnmarked for the listener	Use an accent already familiar to the people who will hear it.
DoJudge neutrality against listeners	Don’t treat the team’s own accent as the neutral default.
Don’tDon’t ship accent as a persona feature	A distinctive accent is a comprehension risk and an offense risk, not a harmless personality.
Don’tDon’t assume one accent travels	The same accent that costs one listener comprehension can read as authenticity to another.

Pacing and Rhythm

Pacing is speech rate. Rhythm is where the pauses fall, including how long the voice waits before it starts talking. Hold one steady tempo across the turn. Pause at clause boundaries, not inside a phrase. Leave a short beat before you start talking.

Standard	Requirement
DoHold an even pace	One steady tempo from the first word to the last. Don’t let the rate drift upward.
DoPause at clause boundaries	Let each clause run through. Put the pause at the join.
DoLeave a brief latency before responding	A reply that starts the instant the person stops talking reads as a machine.
Don’tDon’t pause mid-phrase	A pause in the wrong place is worse than none.
Don’tDon’t clip the end of a turn	Let the turn finish, leave a beat, then continue.

Intonation

Intonation is pitch movement across a sentence, as distinct from the voice’s base pitch at rest. Move pitch with the meaning of the sentence. Fall on statements. Lift only on real questions. Change contour from one sentence to the next, and match the delivery to the news.

Standard	Requirement
DoFall on statements, lift on real questions	End a confirmation on a level or falling contour. Lift only when actually asking.
DoChange contour between sentences	Each sentence carries its own contour.
DoSoften on bad news, lift on resolution	”I found an opening tomorrow” and “I wasn’t able to book that” should not sound the same.
DoMatch contour to intent	The words and the intonation have to agree.
Don’tDon’t use flat delivery	Steady pitch, same volume, and the same rhythm on every phrase reads as a machine.
Don’tDon’t use one rising contour on routine content	Uniform overly-positive inflection reads as insincere.

Voice Temperature

Temperature is emotional variability. Low is steady and predictable; high is warm and expressive. Keep temperature low by default. Raise it only at the moment the content calls for it, then return to baseline. Sustained warmth across a whole interaction reads as performance rather than care.

Persona / channel
Consumer, phone	Low through transactional turns. Raise it when the caller describes the problem, and on bad news. Never sustain it.
Consumer, in-app	Low throughout, with a lower ceiling than the phone channel. Stay matter-of-fact on search and booking.
Technician, field capture	Low throughout. Confirm the capture; don’t encourage. Warm texture is not expressive variation during capture.
Dispatcher, in-app	The lowest of any persona. The same even register whether the news is routine or bad.
Owner, in-app	Low. The same even register whether the number is favorable or not.

Tone & Trust

Text and chat standard for CSR, Dispatcher, and Owner.

Principles

Structure first, then tone, then sound.

Patterns

How each persona’s turns are built.

AI Kit

Agent Experience

Persona and Voice Guidelines

Tone standard

Voice and sound

Pitch and Timbre

Brightness and Darkness

Clarity and Articulation

Accent

Pacing and Rhythm

Intonation

Voice Temperature

Tone & Trust

Principles

Patterns

​Tone standard

​Voice and sound

​Pitch and Timbre

​Brightness and Darkness

​Clarity and Articulation

​Accent

​Pacing and Rhythm

​Intonation

​Voice Temperature

​Related

Tone & Trust

Principles

Patterns

Tone standard

Voice and sound

Pitch and Timbre

Brightness and Darkness

Clarity and Articulation

Accent

Pacing and Rhythm

Intonation

Voice Temperature

Related