The actor tells the system what to change about its output — shorter, warmer, restructured, this part only — and gets a revised version rather than a fresh draw or a manual edit.
Forms
Recurring forms: quick-action verbs on the output (expand, shorten, restyle, transform, summarise), free-form refinement commands in the conversation, and scoped revision where a selection plus an instruction rewrites just that span.
Resources & references
- Shape of AI — Expand, Restyle, Restructure, Transform, Summary
- AI UX Playground — Response Refinement, Magic Edit
- aiuxpatterns.com — Result Actions
To-do
Generated content hosts the “Actions” stub this decomposes; migrate as this matures. The boundary with prompt needs drawing: a revision is a delta on standing intent, not a rearticulation of it.
Consequences
- iteration proceeds by deltas: the actor directs the change in their own words without remaking the output or rearticulating the request
- the instruction acts on the whole unless scoped, so an untargeted revision churns the parts that were already right
- the actor shaped the result without ever touching it, which leaves authorship — and responsibility for the details — ambiguous
Related patterns
Enacts
- Agency — the actor steers the output with judgement the system executes
Complements
- Selection — a selected span scopes the revision to the part that needs it
Alternatives
- Regeneration — naming the change versus drawing again on the same intent
- Editing in place — saying what to change versus changing it by hand
Related
- Text lens — a lens changes how the content reads; a revision changes what it is
Preceded by
- Generated content — acts on output the generation activity produced