PTN—14

Style presets

Named, reusable voice and tone presets — including user-saved ones — instead of re-prompting “make it friendlier” every time.

Voice is the intent people re-express most and articulate worst. Everyone knows their team’s update memo does not sound like their customer apology, but almost nobody can specify the difference in prose — so they generate, wince, and type “make it friendlier” for the fourth time this week. A style preset captures that judgment once, gives it a name, and makes it a one-tap choice on every future request.

Naming is the design work. “Friendly” and “professional” are adjective soup — every writer means something different by them, and the preset that ships under a vague name gets applied once and abandoned. Presets named by outcome — “Customer update”, “Exec summary”, “Release notes” — carry their own instructions: the person knows when to reach for one, and the system knows what it promised. The best presets read like destinations, not moods.

Preview is what separates a preset from a gamble. Showing the same content restyled under each option — before anything is applied — turns an abstract label into a concrete decision, and it is the only honest way to sell a preset the user did not write. Once applied, the preset must layer over the person’s own prompt rather than replace it: a style is how the answer sounds, never what it says, and a preset that erases explicit instructions has overstepped its job.

User-saved presets are where the pattern compounds. The tuned voice that took four rounds of correction to reach is an asset; letting the person name it, save it, edit it, and delete it converts a good session into permanent capability. A saved preset the user cannot inspect or revise, though, is a memory they cannot audit — treat it as legible, editable state, not a frozen artifact.

Anatomy

Fig. 01 — Style presetsLive specimen

Your instruction: Write a reminder for invoice #4127 — mention the online payment link.

Pick a preset to preview the same message in that voice.

A preset row with outcome-named styles, live previews of the same content, and an editable user-saved preset.

1Outcome-based names. Presets are named for what they produce — “Customer update” — not adjective soup like “friendly”.
2Restyled preview. The same content is shown restyled under the preset before anything is applied.
3Editable saved preset. A user-saved preset can be inspected, edited, and deleted — it is legible state, not a frozen artifact.
4Prompt preserved. Applying a preset layers over the user’s own instructions; it changes how the answer sounds, never what it says.

When to use

People re-type the same tone corrections — “warmer”, “tighter”, “less formal” — across sessions.
The product serves recurring output types with distinct voices: support replies, exec updates, marketing copy.
A tuned voice takes several rounds to reach and is worth keeping once found.
Teams need consistent voice across many writers, and the preset can carry the standard.

Design considerations

01Name presets by outcome or destination, and keep the built-in roster small enough to know by heart.
02Preview each preset against the person’s actual content, not a canned sample.
03Compose presets with the prompt — explicit written instructions always win a conflict, and say so.
04Let people save a preset from a good result in one action, then edit and delete it freely.
05Show the active preset wherever it applies, so a styled output is never a mystery.
06Make “no preset” a first-class option rather than forcing every output through a voice.

Pitfalls

✕Adjective-soup presets — “friendly”, “professional”, “casual” — that mean nothing specific and get abandoned after one use.
✕A preset that overwrites the user’s explicit instructions, changing what was said in the name of how it sounds.
✕Save-only presets the person can never inspect or edit, turning their own tuned voice into a black box.
✕Sticky preset state: a style set last week silently colors today’s unrelated work.

In the wild

Claude: Styles offer built-in voices — concise, explanatory, formal — plus custom styles users create from examples, then edit and reuse across chats.
Notion AI: Change-tone actions apply named voice transforms to existing content as one-tap menu items rather than re-prompting.
Grammarly: Voice profiles and tone rewrites capture a target register once and apply it across everything a person writes.