Skip to main content
No verbal register has been decided for any Invoca agentic feature yet. This page proposes a framework for making that decision — it is not the decision itself. See Coverage, stated honestly.

What it governs

The tone and register of sentences an AI feature produces — a generated summary, an explanation, a hedge, a refusal — as distinct from data it merely renders, like a chart or a table of call metrics. It governs how “talking” content should sound; it does not govern whether uncertainty is disclosed at all (that is Trust builders) or what the feature is named (that is Identifiers: Name). It does not govern Content: Voice and tone itself — Invoca’s general brand voice, which this page inherits and narrows for AI-generated language specifically.

Vocabulary

“Talking” is not “rendering.” A generated summary sentence is talking, and in scope here. A table of call metrics beside it is rendering, and out of scope — it is governed by Foundations: Typography and the relevant component, not by this page.

Choosing a value

Reference

Nothing is decided. No Invoca agentic feature has a stated tone or register today, including Signal AI’s shipped transcript summaries. This section will be generated from a real decision once one exists.

Relationships

Personality directly inherits Content: Voice and tone’s three traits and its constraints — TITAN-VOICE-01 (plain language over technical wording), TITAN-VOICE-02 (no acronyms or insider phrasing), TITAN-VOICE-03 (concise over exhaustive) — all apply to AI-generated sentences exactly as written for the rest of the product; this page does not restate them, it narrows their application. It pairs with Identifiers: Name — a feature’s name and its speaking register are usually decided together — and with the overview’s disclosure requirement: tone must never obscure the six disclosure-and-recourse questions.

Accessibility

  • Plain language (TITAN-VOICE-01) doubles as an accessibility aid: it helps readers with cognitive or language-processing differences as much as it helps a reader outside the company understand copy without translation.
  • Tone must never substitute for or obscure the six disclosure-and-recourse answers from the overview — what the system used, how sure it is, how to check it, how to correct it, and how to get out. A warm register that omits or buries any of these is not a tone problem; it is a missing answer wearing a friendly voice.

Constraints

Divergences

Not applicable. Nothing is shipped yet to diverge from.

Gaps

  • Whether tone should shift by outcome state — more measured when uncertain, more direct when confident — is undecided.
  • Whether personality is one voice for all of Invoca’s agentic surfaces or decided per feature is the same per-feature-or-once question flagged on the overview.
  • How Invoca’s three brand traits actually read once generated by a model, rather than written by a copywriter, has not been evaluated.
Identifiers: Avatar · Identifiers: Color · Identifiers: Iconography · Identifiers: Name · AI Experience overview · Content: Voice and tone

Why it works this way

Restricting personality to Invoca’s existing three brand traits, rather than letting AI surfaces develop their own, keeps the brand from fracturing at its most visible new surface. A model’s output is the first place many users will notice the product doing something new; if it also sounds like a different product at that moment, the inconsistency compounds instead of canceling out. Scoping this page to “talking,” not “rendering,” keeps it from re-deciding typography or component rules that already govern how data displays. A generated sentence needs a tone decision because its exact words are new every time; a rendered number does not, because its presentation was already decided elsewhere.
Last modified on September 3, 2026