Communication Style

What Communication Style is

Settings → AI Voice & Style → Communication Style is where you decide how your AI sounds, as opposed to what it knows or what it is allowed to do. Five settings live on that tab: Tone of Voice, Emojis, Verbosity, Headlines and Replace em-dashes.

They are team-wide and they reach every written reply the AI produces: chat, email, social comments and DMs, and the drafts your agents see in the helpdesk. They are a baseline, not a script. A policy or a workflow step can still be specific about the wording of its own answer; Communication Style is what applies when nothing more specific has been said.

Tone of Voice 🎙️

Five options: Neutral, Professional, Friendly, Efficient and Custom.

The four presets are handed to the AI as exactly what they say, one word. Picking Friendly tells the AI to be friendly and leaves the interpretation to it. That is usually a perfectly good starting point, and it is the fastest way to move off the default.

Neutral is the default, and it says nothing at all about tone. The AI writes in its own natural voice. See the section on defaults below.

Custom opens a free-text box (up to 1000 characters) and replaces the preset entirely. This is where nuance belongs: do's and don'ts, signature phrases, openings you never want to see, formatting habits.

Example of a custom tone: "Warm and knowledgeable, never pushy. Lead with empathy ('Oh no, that sounds frustrating') rather than a greeting. Bold for the key number, never headings. At most one emoji, at the end. Never open with 'Unfortunately' or 'Thank you for contacting us'."

Two things to watch with Custom. It replaces the preset rather than adding to it, so write the whole voice, not only the fix you just thought of. And selecting Custom while leaving the box empty gives you no tone instruction at all, which is the same as Neutral.

Emojis 🙂

• Never: no emojis, anywhere.
• Less: sparingly, only where they clearly fit.
• Balanced (default): nothing is said, so the AI uses its own judgement.
• More: leans into them.

The setting does not vary by channel, and the same reply style shows up in email as in chat. So pick the level your most formal channel can live with. Less is the sweet spot for most brands: enough warmth to not read like a robot, not enough to look unserious on a complaint.

If emojis are load-bearing for your brand, say where they go in a Custom tone ("one at the end, never in the first line") rather than only turning the dial up.

Verbosity 📏

Short, Balanced (default) or Lengthy.

This is the highest-leverage setting on the tab. Long support replies are not more helpful, they are more work to read, and they bury the one sentence the customer wanted. If your drafts routinely need trimming before an agent sends them, Short fixes more of that in one click than any amount of tone wording.

Lengthy is worth it in a narrow case: complex products where the answer genuinely needs context, caveats and next steps.

Headlines 🔤

Controls whether the AI uses headings (h1, h2 and so on) to structure a reply.

• Never: no headings at all. The AI uses bold text for emphasis instead.
• Minimal: headings only when a reply is genuinely long with distinct sections. Otherwise bold, paragraphs and lists.
• Balanced (default): nothing is said.

A support email with headings in it looks written by a machine, because humans do not write emails that way. Never or Minimal is the more human-looking choice for almost every team.

Replace em-dashes ✂️

On by default, and it works differently from everything else on this tab. It is not an instruction to the AI, it is a find-and-replace applied to the finished text: every em-dash becomes a normal spaced hyphen. "Sure — done" and "Sure—done" both come out as "Sure - done". Line breaks and list indentation are left alone.

The reason it exists is that the em-dash is one of the strongest "written by AI" tells in the wild. Turn it off only if your house style genuinely uses em-dashes.

It is applied at every point where AI-written text leaves the system: sent messages and drafts, the live chat stream, AI subject-line suggestions, AI translations, and the composer's Refine action. Text an agent typed by hand is untouched unless they run Refine over it.

What "Balanced" and "Neutral" actually do 🧠

Here is the part that surprises people. The default value of each setting adds nothing to the AI's instructions. Neutral tone, Balanced emojis, Balanced verbosity and Balanced headlines are not a carefully chosen middle setting; they are silence, and the AI falls back on its own habits.

Three consequences:

• An untouched configuration is not a deliberate house style. It is the model's taste.
• If replies come out too long, too heading-heavy or too emoji-happy, you have not overridden anything yet. Move the setting off its default and it will be obeyed.
• Every setting you move off the default is one more line competing for the AI's attention. Change the ones that matter to you and leave the rest alone.

What this screen does not control

• Language. The AI always replies in the customer's language, and falls back to the market's language when that is unclear. Nothing here changes that.
• The AI's name. That lives on the Company Information tab, together with a short description of your company and what you sell. It has its own translations.
• Signatures and AI disclaimers. These are attached automatically after the reply is written, and are configured elsewhere. Do not try to reproduce them through tone.
• Rules the AI must never break. Those belong on the Guard rails tab, which is mandatory in every response. Tone is guidance; a guardrail is not.
• Edge cases and one-off requirements. The Advanced Instructions tab, or better, a policy.
• Per-market nuance. Communication Style is one setting for the whole team. If a market needs its own wording habits, use that market's Agent Instructions under Settings → Markets, which are added on top for conversations in that market.
• Phone calls. The voice agent has its own prompts and voice configuration.

Tuning it without guessing 🔧

1. Start with a preset. Friendly, Professional or Efficient, plus Verbosity if you already know your replies run long.
2. Read ten real drafts. Real conversations from your own inbox, not test cases. Write down what actually bothers you about them.
3. Change one thing. If the complaint is length, that is Verbosity. If it is structure, that is Headlines. If it is a phrase you hate, that is a Custom tone.
4. Move to Custom only when a preset keeps missing. Then write the whole voice in one go, since Custom replaces the preset.
5. Re-read ten drafts on the same topic. Voice changes are easy to imagine and easy to over-correct, so check against real output before you adjust again.

Good to know

• Changes take effect on the next AI run. There is nothing to republish.
• Communication Style also shapes Zuzana Magic in the helpdesk composer, so a reply an agent writes by hand and then refines comes out in the same voice as the AI's own.
• It reaches the AI's whole reasoning chain, not just the final sentence: the planner, knowledge base search, policies and workflow steps all see the same voice instructions.
• The Custom tone box holds 1000 characters. If you are fighting the limit, some of what you are writing is probably a rule, and rules belong in Guard rails or a policy.
• Tone will not rescue a wrong answer. If replies sound right but say the wrong thing, the fix is in your knowledge base, policies or workflows.