cjy
AI-Generated Insights – Designing for Trust, Editability, and Clarity
Role: Freelance Product Designer
Scope: UX strategy, IA, interaction design, microcopy
Timeline: 3 weeks, April-May 2025
project overview
As part of a freelance engagement with a productivity-focused software client, I led the UX design of an AI-powered meeting recap experience that generates summaries and action items after meetings.
Although the underlying AI produced useful output, users often rewrote or ignored meeting notes due to unclear accuracy, limited control, and uncertainty around ownership. This reduced adoption and undermined the value of automation.
The solution reframed AI-generated notes as a collaborative draft, enabling fast inline edits, clear ownership of action items, and an explicit confirmation step before sharing. The resulting experience reduced manual rework and helped teams confidently act on meeting outcomes.
Client details have been anonymized to respect confidentiality.
problem statement
Although AI can generate meeting summaries instantly, users often hesitate to rely on them. Unclear accuracy, limited editability, and fear of sharing incorrect notes cause people to rewrite or ignore AI output entirely.
goals and constraints
The project focused on improving adoption and follow-through without slowing users down or introducing heavy configuration.
Goals
✅ Make AI output feel reviewable and safe
✅ Enable fast, lightweight edits
✅ Clarify ownership of action items
Constraints
⚠️ Minimal interruptions to post-meeting flow
⚠️ No modal-heavy editing
⚠️ Must scale across recurring meetings
user mental model
DRAFT → REVIEW → CONFIRM → SHARE
Users generally view AI-generated notes as “probably right, but needs review.” The experience needed to support that mental model instead of fighting it.
core design decisions
1. draft status & transparency
Users need to immediately understand that AI notes are a starting point, not a final artifact.
2. inline editing for speed
Editing AI output should feel collaborative and quick, not like correcting a mistake.
3. action item ownership
Incorrect or ambiguous ownership undermines trust and follow-through.
interaction states
Clear states are needed in order to prevent accidental sharing, lost edits, and confusion around what’s final.
What to show
One component with:
-
Default (AI draft)
-
Hover / focus
-
Edited
-
Confirmed
-
Regenerate
-
Error / unavailable
microcopy and language strategy
Language plays a critical role in shaping trust with AI systems. It should be collaborative (not authoritative), it should encourage review, and it should avoid absolutes.
(table with microcopy tweaks and why it matters)
end-to-end flow
This flow demonstrates how the feature fits naturally into a real post-meeting workflow.
-
Meeting ends → AI generates notes
-
User reviews summary
-
User edits action item
-
User confirms notes
-
Notes become shareable
-
Action items sync to tasks (optional)
edge cases and safeguards
AI systems need to gracefully handle uncertainty and incomplete data. It should anticipate failure and handle it responsibly.
Examples
-
AI unsure about ownership
-
Partial meeting transcript
-
Conflicting action items
(SCREENS)
-
Partial transcript state
-
Soft warnings
-
Guidance text
impact and measurement
Success was defined by adoption and follow-through, not novelty.
Expected outcomes
-
Less manual rewriting
-
Faster sharing
-
Higher confidence in AI output
Metrics
-
Edit vs regenerate rate
-
Time to confirmation
-
Action item completion rate
reflection and next steps
What worked:
-
Draft framing lowered anxiety
-
Inline edits preserved speed
Next
-
Confidence indicators for AI output
-
Team-level sharing preferences
-
Progressive disclosure or other shortcuts/tools for power-users
client disclaimer
Client details have been anonymized to respect confidentiality.
