Chat has its own register
Live chat sits between the formality of email and the speed of texting a friend. Too stiff and you sound like a form letter; too casual and you sound careless with someone's problem. The right register is warm, plain, and brief — a competent human being who is glad to help and does not waste your time. Most brand voice guides were written for marketing pages and translate badly to a support chat.
Consistency is a trust signal
When every agent sounds like a different company, visitors notice — it reads as disorganized. A shared voice makes the whole team feel like one reliable place, whether the visitor reaches the newest hire or the veteran. That consistency is worth designing for, especially as the team grows past the point where everyone absorbed the tone by osmosis.
Give guardrails, not a script
The instinct to enforce voice with rigid scripts is exactly wrong for chat. Word-for-word scripts make agents sound like robots, and visitors can feel it instantly. Instead, give principles and examples: how we open, how we say no, how we handle a mistake, how we sign off. Show a good version and a bad version side by side. Trust agents to hit the tone in their own words.
The hard moments define your voice
- Saying no: honest and kind beats corporate hedging. "We can't do that, but here's what we can do" is your voice under pressure.
- Owning a mistake: a plain apology without excuses does more for the brand than any tagline.
- Not knowing: "Let me find out" said confidently keeps trust; bluffing destroys it.
Keep canned responses human
Saved replies are where voice quietly dies — they get written once, formally, and pasted forever until every chat sounds automated. Write canned responses in your actual voice, keep them editable in the moment, and review them the way you review anything else that speaks for you. A macro is a starting point, not a final word.
Let the AI layer match the voice too
As an AI layer handles more first replies, its tone becomes part of your voice whether you planned it or not. Give it the same guidance you give agents, and review what it says the way you review a new hire's chats. In MyLiveChat, canned responses and AI answers draw on content you control, so the voice stays yours as the volume scales.
Three voice decisions worth writing down
Most tone guidance is too abstract to act on. Three concrete decisions do more work than a page of
adjectives, because they resolve the arguments that actually come up.
First, formality: do you use contractions, first names, and short sentences, or not? Pick one and
be consistent, because mixing them within a team reads as inconsistency rather than range. Second,
humour: is it allowed, and where does it stop? A useful default is that humour is fine when things
are going well and disappears entirely the moment someone is frustrated.
Third, apology: how do you say sorry, and how often? Teams tend to either apologise for
everything, which dilutes it to noise, or avoid it defensively, which reads as evasion. Agree a
middle position — apologise once, specifically, for the thing that actually went wrong, then
move to the fix — and you have resolved most tone disputes before they happen.
Where voice breaks, predictably
Voice rarely fails in ordinary conversations. It fails in three places, and they are the same
three places for almost every team.
It fails when refusing. Under the discomfort of saying no, warm writers turn stiff and formal,
reaching for passive constructions and policy language that sounds like a different company wrote
it. The fix is to keep the refusal in the same register as the rest: plain, direct, and human, with
the reason stated once.
It fails at escalation, where the handover often reads as a shrug. And it fails at the close,
where a warm conversation ends on a pasted sign-off that undoes the warmth. These three moments are
worth writing example wording for, because they are where the gap between your intended voice and
your actual voice is widest — and where visitors are paying most attention.
Most arguments about tone are actually arguments about formality, and they stall because the two get treated as one setting with a single slider from stiff to chatty.
They are separate. Warmth is how much the writing acknowledges the person; formality is how much it observes convention. You can be warm and formal, which is what good hospitality sounds like. You can be cold and informal, which is what a rushed agent sounds like when they have adopted the vocabulary of friendliness without any of the attention.
Separating them makes the decision tractable. Most teams want high warmth and moderate formality, and can then adjust formality by audience without touching warmth at all. A team supporting procurement officers and a team supporting hobbyists can share one warmth standard and differ entirely on contractions, exclamation marks and first names.
It also explains the most common complaint about scripted chat, which is that it reads as insincere. That is rarely a formality problem. It is warmth being performed by vocabulary rather than delivered by attention: the words say we are delighted to help while the reply ignores half the question. Fixing it means reading what was asked, not adding more friendly adjectives.
Voice shifts within a single conversation
Voice is usually written as a fixed property of the team, but it is not fixed across a conversation. The same agent should not sound identical at the opening, in the middle of diagnosis, and at the close, and a voice guide that ignores this produces uniformly wrong writing somewhere.
The opening carries the most personality, because it is establishing that a person is present. The middle should get plainer as the work gets more technical: when you are giving instructions to follow, warmth is expressed through clarity and short steps rather than through phrasing. The close widens again, because that is where you confirm the outcome and leave the door open.
Agents who sound wrong often have the shape inverted. They open briskly, which reads as impatient, and then become chatty during the part where the customer is trying to follow instructions, which reads as not taking the problem seriously. Neither is a vocabulary problem and neither is fixed by a word list.
This is a more useful thing to coach than a tone adjective. Ask an agent to read one of their own transcripts and mark where the register changes; most people can see the mismatch immediately in their own writing, and the correction tends to stick better than a rule handed down. It is one of the more productive things to look for in a regular transcript review.
What to measure
Read for voice in transcript review rather than trying to measure it directly. A simple question
— does this sound like us? — asked over a handful of conversations a week catches drift
earlier than any metric.
Watch the refusal and escalation moments specifically, since that is where the drift concentrates.
Sampling only routine conversations will tell you your voice is fine when it is not.
Check your canned library against the voice decisions on a schedule. Snippets are written once and
pasted thousands of times, so they carry more of your voice than any individual agent does.