Select → refine → transform

AI Refinement Workflow

Project status: Validated
Problem

AI tools generate quickly, but refinement is difficult to control.

Strategy

Reframed refinement from a regeneration problem into a control problem.

Solution

Designed multi-level refinement workflows for broad edits, precise changes, and version comparison.

Impact
  • 100% task success in Maze usability testing
Role
Lead Designer
Timeline
6 months · Aug 2025–Jan 2026
Team
3 designers on refinement use case
Tools
Figma · Maze · Usability testing · Medium

Problem

Generating content was easy. Refining it is where users lost control.

AI tools make first drafts fast, but refining outputs often forces users to choose between regenerating everything or editing manually. This makes iteration repetitive, unpredictable, and difficult to control.

Generate content

Need refinement

Users need to refine generated content

Regenerate everything

Loses structure users wanted to keep

Manual editing

Pulls users out of workflow

Research

Research showed users needed refinement they could see and control.

I analyzed existing AI tools and refinement workflows to understand how users maintain control while iterating on generated content. Existing systems tended to prioritize either broad exploration or precise edits. These findings informed three design principles that guided the rest of the design process.

Existing tools
Design principles

  1. Support multiple refinement depths

  2. Preserve structure during refinement

  3. Keep users in control of AI changes

User testing insights

We tested mid-fidelity prototypes with users through interviews and task-based walkthroughs focused on refinement workflows. Participants completed tasks across broad edits, detailed editing, merge workflows, and version history interactions while sharing where friction occurred. This helped us identify consistent patterns around control, visibility, and preserving structure.

  1. Users rarely accepted first outputs.

  2. Users wanted to refine specific parts without restarting.

  3. Users needed clearer visibility into what changed.

Problem reframing

Refinement was a control problem, not just a generation problem.

Instead of optimizing for better first outputs, we explored how users could refine AI-generated content while preserving structure, intent, and momentum.

Previous workflow

Regenerate everything

Reframed opportunity

Enable controlled refinement

Concept testing

Testing refinement interactions pointed toward one clear workflow

My concept was selected because it supported both broad and precise refinement without interrupting workflow.

01Overall refinement
Wireframe showing response options, highlighted phrase, tone menu, and refined output

Broader changes like rewrite, merge, tone, or structure.

02Inline selection
Wireframe showing inline text selection and a rewrite prompt for targeted refinement

Highlight exactly what needs refinement.

03Inline refinement
Wireframe showing inline text selection, a More Formal refinement menu, and the refined output snippet

Refine selected content without regenerating everything.

Key decisions

Three design decisions shaped a more controlled refinement workflow.

  1. Users needed refinement at multiple levels

    Output

    Paragraph

    Sentence

    Word

  2. Precise edits needed to preserve structure

    BeforeWhole output regeneration

    Try again control that regenerates the full AI output

    AfterSelected-region refinement

    Inline text selection with Ask AI and refinement options on highlighted text

    Challenge

    Regenerating full outputs often overwrote parts users wanted to keep.

    Decision

    Enabled targeted editing so users could refine selected regions without changing the rest.

    Impact

    Users kept working context while guiding AI on specific regions.

  3. Version visibility made refinement easier to evaluate

    Side-by-side comparison of Current Response and Refinement V1 with version preference controls

    Challenge

    Users needed to compare versions and understand what changed.

    Decision

    Added side-by-side version comparison and navigation.

    Impact

    Users could judge AI changes before committing to them.

Solution

Supporting refinement without restarting

We designed a refinement workflow that combined broad edits, targeted changes, and version comparison so users could steer AI outputs without losing context or starting over.

Overall RefinementBroad changes without rewriting prompts.
Detailed EditingRefine specific regions without changing everything else.
Version ComparisonReview iterations before committing to changes.

Impact

Testing validated the refinement workflow.

Maze testing validated that users understood the refinement model and successfully completed tasks across overall refinement, inline refinement, and comparison workflows.

Impact metrics

  • 100%

    Task completion across core refinement tasks

    Users completed overall refinement, inline editing, and comparison tasks successfully.

  • 0%

    Misclick rate during refinement tasks

    No navigation or shortcut confusion appeared during testing.

  • Preferred approach

    Inline refinement

    Participants consistently preferred localized edits over repeated prompting.