UX Digest - best posts

A Review of Experiments with Synthetic Users Review of 12 studies: 9 encouraging, 14 discouraging. Synthetic users match some means but fail on details (reduced variance, shallow depth). Only 3 of 14 classic studies replicated. Best use: querying collected data—not prediction. Critical decisions shouldn't rely on them yet. Correlation ≠ equivalence From User Research to Building: Six Months Later A researcher transitioned to a "Builder" role (no official title). Key lessons: switching from no-code AI tools to Cursor + terminal was a huge unlock. Centralized tools aren't critical anymore—what matters is an "intelligence layer" (shared context, data). She helped researchers use Cursor with Qualtrics and Snowflake without SQL. Some colleagues feel AI killed creative thinking. No clear role exists—confusion is normal 🎥 NNG: Field Guide to Explaining UX Strategy Simple, relatable ways to explain complex UX strategy concepts like UX vision, goals, OKRs, and outcomes. Translate UX strategy into language anyone on your team can understand Prototyping: SONO - Designing a Mood-Based Music Discovery ExperienceSONO - Designing a Mood-Based Music Discovery Experience A case study about a music app using AI (Aria) to match songs to user emotions instead of listening history. Usability testing showed the app worked, but users found it generic: "It didn't really listen to me." Key insight: usability ≠ value. When designing around emotion, people expect the experience to feel real. The project became less about music and more about what "personal" truly means Case Study: Travel Booking Redesign of an Australian bus service with 0.29% conversion. Data showed demand existed but the booking funnel was broken. Usability testing revealed critical issues: price calendar not found, cancellation policy invisible. Fixes: calendar opens by default, specific trust strip above pay button. Testing doesn't validate designs—it breaks them AI: AI in practice - the week AI got scary, political, and expensive Anthropic unveiled Mythos—the most powerful AI ever (100% on Cybench, finding thousands of zero-day vulnerabilities)—and deemed it too dangerous for public release. OpenAI proposed robot taxes and a four-day workweek. Meta abandoned open source, going proprietary. Anthropic passed OpenAI in revenue. The one-model-fits-all era is over Basics: The Rule Nobody Teaches You - Rapport Before Research People give "safe answers," not the truth—that's the data you lose without rapport. Rapport isn't about being friendly—it's about being real. Code-switching (using their language) changes everything. Rapport opens space for their truth; leading fills it with yours. The script is a starting point. The goal isn't a smooth session—it's the truth. Keep your research questions front of mind, not the guide. Everything else is flexible Interesting: Privacy-first connections - Empowering social experiences at Airbnb Airbnb built social features with privacy by design: separate User (internal) from Profile (public). One user can have multiple profiles (Host, Guest, Experience-specific), each with its own ID. Decoupling User ID from Profile ID enables context-aware visibility and privacy controls. Goal: meaningful connections while guests control their privacy @uxdigest