• Introduction

Prose

The prose stratum of the interface — the words, phrasings, labels, and linking choices through which the system takes its turn in the conversation with the user.

Scope

This foundation treats prose as a set of generative moves on the actor↔system conversation: choices that transform the structure of the exchange rather than decorate it. What to call a button, whether to narrate a save, which word to anchor a link on, how much to scaffold an empty state — these are design decisions of comparable weight to structure, affordance, and flow.

Prose as a design move

Shifting Cancel to Keep editing is a design move. It changes what the interaction does — the reader’s default, the weight of the consequence, the stance the system takes toward the interrupted work — without changing a line of structure. The same is true of the decision to say Saved instead of We’ve saved your draft: one yields the stage, the other claims it.

Prose moves are generative in the same sense structural moves are: they act on an existing centre (a button, a confirmation, an empty state, a tooltip) and produce a new one (a softer default, a warmer apology, a more trusting absence). They can be read through the same lenses as structural moves — agency shifts, learnability gains, formality loosens — because they alter the same structure.

Stance and voice

The authorial question behind every piece of interface prose is who does the system sound like? A continuum runs from narrating system — “We’ve saved your draft”, “I’m thinking about this”, “Here are three suggestions” — to yielding the stage: prose that steps back and lets the user’s own material, own vocabulary, own judgement carry the interaction.

Yielding the stage is a rhetorical move. It move treats the reader as a competent adult rather than a user needing reassurance. It is the prose counterpart of trusting the user’s judgement in an agency sense.

The tension: system voice versus user voice. Foregrounding the system’s narration helps when the system’s action is non-obvious, non-reversible, or requires the user’s trust — an AI completion that needs to explain its own uncertainty, a background agent reporting what it did. Foregrounding gets in the way when the user is in flow, the action is obvious, or the narration is there to perform presence. Transparent reasoning is a case where system voice is load-bearing; most microcopy is a case where it isn’t.

For AI-participant systems, stance is not only a prose choice but a specifiable generation input — see § Generated prose as a design surface below and AI tuning.

Register and tone as prose decision

Three modes of authored linking, after Rao (2009):

  1. Citation mode. The link functions as a footnote. The anchor names its target (“see Rao”), and the reader can follow it if they want verification. The author-reader relationship is backward-looking: the link inherits its rhetoric from academic writing. This is the default mode in most documentation, including the links in this library.

  2. Form-content blending. The link is placed on anchors that do not obviously name their target. The effect is disruption-and-reward: the reader notices the link, follows it on trust, and is surprised by what opens. The move is most legible in essayistic writing and least useful in instrumental interfaces — it asks the reader to be curious, which is a register interface prose cannot usually presume.

  3. Figure-ground-voice integration. The link compresses cultural or contextual density into minimal surface text, trusting the reader to expand the compression by following if they need to, and to stay in the flow if they don’t. The move treats the reader as a competent adult; it is the authorial-linking counterpart of § Stance and voice’s yielding the stage.

The anchoring tension: link as reference versus link as prose. Citation mode is safe and legible; it also produces the thinnest reading. Mode 3 compresses reading for competent readers and yields a richer text, but presumes the reader can do the work. Design choice: the more instrumental the surface, the further toward citation mode it should sit; the more essayistic or expressive, the further toward mode 3.

  • Dynamic hyperlinks — Dynamic hyperlinks are prose-as-affordance: the link text is the prose, and the move from authored phrase to actionable target is itself the prose move.
  • Reference — Reference links rest on a prose convention — that an authored phrase points to a canonical entity. The prose decides what counts as a name worth linking.
  • Deep linking — Deep linking surfaces a piece of system state as a shareable address; the prose around the link does the work of telling the recipient what they will land in.
  • Information architecture § Hypertext structures covers the structural dimension of linking (associative, temporal, hierarchical); the rhetorical dimension lives here.

Microcopy as assistance-register prose

Helper text, placeholders, empty states, confirmations, error messages, toasts — the thin prose that surrounds component-level interactions. Framed as design moves rather than templated outputs, these are assistive turns: short utterances that perceive (“You have no drafts”), plan (“Try adding your first tag”), recover (“We couldn’t save that — your changes are still here”), or acknowledge (“Saved”) on the user’s behalf.

The anchoring tension: silence versus scaffolding. Every microcopy surface has a no-prose option (absent helper text, blank empty state, no toast) that is sometimes better than the filled option. Over-narrated interfaces feel patronising; under-narrated ones feel indifferent. The decision is per-surface.

Microcopy touches several patterns without being owned by any. Data entry covers helper text and placeholder; status feedback covers validation and notification categories; toast covers brief confirmations; callout covers unskippable messages; notification covers the information/success/warning/error taxonomy. Each is a surface where prose-as-move decisions get made.

Generated prose as a design surface

The anchoring tension: authored prose versus generated prose. Most interfaces now hold both — authored microcopy beside model-generated completions, authored labels beside model-written summaries. Register drift between the two is a design risk: a warm empty state followed by a clinical AI completion reads as two interfaces stitched together. The design surface is the coherence between them, not either alone.

Resources & references

Related patterns

Enables

  • Collaboration — Collaboration depends on shared interpretive ground; prose is the most flexible carrier of that ground between participants.

Enacts

  • Agency — yielding the stage as locus shift
  • Learnability — prose carries domain scaffolding
  • Formality — structural demand on user input (disambiguation)
  • Conversation — turn-taking as substrate

Related

  • Information architecture — IA owns the labelling vocabulary; prose owns how those labels get phrased and where they tolerate ambiguity.
  • Transparent reasoning
  • AI tuning
  • Dynamic hyperlinks — Dynamic hyperlinks are prose-as-affordance: the link text is the prose, and the move from authored phrase to actionable target is itself the prose move.
  • Deep linking — Deep linking surfaces a piece of system state as a shareable address; the prose around the link does the work of telling the recipient what they will land in.
  • Data entry
  • Status feedback
  • Notification
  • Data & Information — generated prose as a probabilistic/inferred layer
  • Intent & Interaction — Intent & Interaction frames interaction as a sequence of turns; prose is what those turns are made of when they are text.
  • Assistance — Assistance lives largely as microcopy; each helper line is a small assistive turn made of prose.
  • Delegation — Delegation surfaces are increasingly written rather than chosen; the prose decides what the agent sounds like and what authority it claims.