I paired with AI to compress weeks of UX work into days — synthesizing requirements, modeling flows, designing hi-fi screens, and validating with PM/Eng in a fraction of the typical timeline.
- Drop-off risk mapping
- Hesitation points
- Competitive scan
- AI-extracted requirements
- Risk & constraint mapping
- PRD highlights
- Happy path flows
- Safety guardrails
- Recovery paths
- Welcome screen
- Interest selection
- First prompt UX
- PM/Eng alignment
- Copy & safety review
- Scope finalization
- TTFP tracking
- Completion rates
- Drop-off analytics
The challenge
Claude's onboarding needed to earn trust around AI privacy and safety, get users to their first valuable interaction quickly, and minimize drop-off — all in a category with no established patterns. Compressed timelines meant onboarding had to ship alongside launch.
Core Requirements
Earn trust around privacy, demonstrate value fast, minimize drop-off in a sensitive AI-first flow.
No Precedent
No established patterns for AI onboarding existed. Every decision required first-principles thinking.
Speed Constraint
Traditional workflows (PRD synthesis, IA mapping, copy testing) would take weeks. We had days.
Artifacts & deliverables
AI-accelerated artifacts that shaped every decision in the onboarding experience — from journey mapping to final shipped screens.
User journey map
Mapped the complete path to first successful prompt — identifying drop-off risks, trust moments, and success checkpoints at each step.
Trust & safety playbook
Unified messaging principles and objection-handling into a single source of truth — ensuring consistent, calm tone across every touchpoint.
- Calm, transparent tone — no anthropomorphizing Claude
- Contextual reassurance replaces disruptive modal walls
- Guardrails turn rejection into guidance — "I can't help with that, but here's a safer approach"
IA → Lo-fi wireframes
Architecture meets wireframes — mapping happy paths and recovery flows directly to frames, condensing 5 steps into 3 screens with acceptance criteria embedded per step.
Onboarding screens
Three steps, one cohesive experience — Welcome → Interests → First Prompt. Trust messaging embedded contextually; starter prompts guide without limiting creative freedom.
Final shipped welcome
The moment users reached after onboarding — simple, trust-first, ready for interaction. The culmination of every design decision.
Success metrics schema
Every design choice tied to a measurable outcome — TTFP ≤ 60s, drop-off ≤ 10%, first-session success ≥ 85%, activation ≥ 75%.
Design decisions & tradeoffs
Where speed met judgment — designing for clarity, trust, and flow under tight timelines. Each decision followed a Before → Decision → After framework.
Guided vs. free prompting
Before First-time users faced an empty text box and didn't know what to ask. Many froze or dropped off before their first message.
Decision Introduced three optional "starter prompts" — safe, engaging entry points without limiting creative freedom.
Result First-session activation jumped as users felt supported, not overwhelmed.
Trust placement
Before Onboarding began with a long modal explaining safety and data handling. It built trust but killed momentum — most users dismissed or dropped.
Decision Broke trust messaging into contextual microcopy placed at the moment of relevance ("Claude never stores personal data" under the first input).
Result Users stayed engaged and absorbed the message naturally. Safety became ambient, not an interruption.
Condensing steps
Before The flow stretched across five screens: Welcome, Safety, Interests, Permissions, and First Prompt. Slow and procedural.
Decision Merged Interests + Safety into one step and simplified permissions, trimming from five to three total screens.
Result Time-to-first-prompt dropped by 40%. The flow felt lighter, faster, and more conversational.
Edge cases & guardrails
These weren't edge cases — they were design requirements. Each one reinforced user trust through clear, ethical behavior.
Privacy
Clear disclosure before first prompt about what data is and isn't stored — shown in context, not buried in terms.
Misuse Prevention
Guardrail copy for unsafe prompts — turning rejection into guidance: "I can't help with that, but here's a safer approach."
Accuracy & Transparency
Inline "confidence framing" to set expectations around AI limits, reducing hallucination-based trust issues.
Drop-off Recovery
Resume onboarding where users left off — no restart — to protect momentum and reduce frustration.
Faster collaboration
AI workflows removed bottlenecks across Product, Engineering, and Content — compressing alignment from days to hours.
Product Management
AI-synthesized PRD highlights, risks, and open questions by EOD. Shared artifacts clarified priorities fast.
Engineering
Lo-fi → hi-fi tied directly to acceptance criteria. Decision logs reduced ambiguity in implementation.
Content Design
Strong first-pass copy to refine — messaging principles and trust microcopy provided reusable patterns.
Results & impact
Measurable impact delivered on time and on target — every metric met or exceeded.
Closing reflections
AI multiplied my velocity — craft made it shippable.
What AI Did
Handled the heavy lift — parsing PRDs, drafting flows, generating first-pass copy — so I could move from zero to options in hours.
Where Craft Mattered
Hierarchy, spacing, tone, and judgment turned drafts into shippable screens. Craft is what made trust feel effortless.
My Working Model
I treat AI like a junior teammate — fast and tireless, but always needing direction. It accelerates work; it doesn't replace design.
Pairing AI acceleration with human judgment shipped Claude's onboarding faster, safer, and sharper — and it's now how I design.