Synthetic Users and UX Research: When Fake Data Misleads

There’s a principle most UX practitioners carry like a badge of professional pragmatism: imperfect research beats no research. Five scrappy user interviews with a rough prototype? Still worth something. A guerrilla hallway test with whoever’s in the office? Better than shipping blind. It’s a sensible heuristic — until synthetic users entered the room and quietly broke it.

The Specific Danger of Synthetic Confidence

Kyle Soucy’s analysis on UX Design CC draws a precise and uncomfortable distinction. Every form of imperfect research that came before synthetic users — small samples, rushed recruitment, a prototype held together with Figma tape — still had one irreducible quality: it involved contact with a real human and their actual behavior. The limitations shaped what you could conclude, but the signal was real.

Synthetic users generate statistically plausible-sounding responses from language models trained on aggregated behavioral data. The problem isn’t that they’re wrong. It’s that they’re confidently, fluently, articulately wrong in ways that look exactly like findings. When a product team in Manila or Jakarta sees a synthetic user “session” reporting that a checkout flow feels intuitive, that response carries the rhetorical weight of a real data point — without any of the epistemic grounding. False confidence, Soucy argues, is categorically more dangerous than acknowledged uncertainty. You can act on uncertainty carefully. You build on false confidence at speed.

What This Means for Design Teams Operating at Scale

For growth-stage brands across Southeast Asia managing multi-platform design systems — simultaneously optimizing for Shopee listings, LINE OA interfaces, and mobile web — the temptation to use synthetic users as a research shortcut is structurally embedded in the workflow. Research timelines compress. Real user recruitment in Manila, Ho Chi Minh City, and Bangkok simultaneously is expensive and logistically messy. Synthetic users look like the elegant middle path.

The failure mode is specific: synthetic users reflect the training data’s median user, not your market’s actual edge cases. A Grab-adjacent super-app flow that works for a synthetic composite of “Southeast Asian mobile user” will systematically miss the grandmother in Chiang Mai navigating a feature for the first time, or the Gen Z shopper in Jakarta who has five payment apps open simultaneously and zero patience for friction. These aren’t edge cases for conversion — they’re your volume.

The tactical discipline here is about classification, not rejection. Use synthetic users for hypothesis generation and stress-testing information architecture before recruitment. Never use them as validation. If a finding from a synthetic session would change a design decision, that decision requires real users before it ships.

---

Naming as Infrastructure: The Underestimated Design System Problem

Vitaly Friedman’s practical guide to naming UI components on Smashing Magazine addresses something that looks like housekeeping but functions like architecture. Inconsistent naming conventions in a design system don’t just slow down handoffs — they fracture the shared mental model between designers, developers, and product managers, which compounds every time a component gets repurposed across a new channel.

For teams managing multilingual interfaces across Thai, Bahasa Indonesia, Vietnamese, and English simultaneously, naming discipline is even more load-bearing. A component called promo-banner in one language context and hero-offer-block in another creates downstream chaos in content management systems and translation workflows. Friedman’s recommendation to build naming conventions around function rather than appearance — what a component does, not what it looks like — is directly applicable here. A time-sensitive-offer-container survives a redesign. A red-countdown-strip doesn’t.

The business case for naming conventions is measurable: design teams at Atlassian have reported significant reductions in component duplication after implementing systematic naming. For a mid-size brand managing design assets across three markets and two platforms, component bloat is a real budget leak — duplicated components mean duplicated QA, duplicated updates, and duplicated design reviews.

Craft Discipline and the Art Director’s Trap

Sergio Membrillas, profiled by It’s Nice That, describes the uncomfortable dual identity of someone who works fluently as both an illustrator and an art director — two disciplines with different value systems and different ways of being taken seriously. The tension he articulates is professionally familiar to anyone who has watched UX researchers get dismissed in a product review because their work didn’t look like “real data.”

There’s a design-adjacent business lesson buried in his experience: the people who earn creative authority in commercial contexts are usually those who learned to translate their discipline’s internal standards into the language their stakeholders already respect. An art director who can connect typographic hierarchy to time-on-page metrics gets a seat at the growth table. A UX researcher who frames qualitative findings in terms of drop-off rates and revenue impact doesn’t get overruled by someone waving a synthetic user report.

This isn’t capitulation to metrics culture. It’s recognizing that good design work — whether it’s a Membrillas illustration or a carefully recruited five-person usability study — has to survive contact with commercial decision-making. The craft earns its place by speaking the room’s language without abandoning its own.

Key Takeaways

• Treat synthetic users as a hypothesis-generation tool only — any finding that would change a shipping decision requires validation with real users before acting on it.
• Name UI components by function, not appearance, to build design systems that survive redesigns and scale cleanly across multilingual, multi-platform Southeast Asian markets.
• Design and research professionals who connect their craft to measurable business outcomes — conversion, revenue, drop-off — are structurally harder to override in product and growth reviews.

The deeper question synthetic users are forcing isn’t whether AI tools belong in a UX workflow — they clearly do. It’s whether the teams adopting them have built the institutional discipline to classify them correctly. Fast tools in undisciplined processes don’t save time. They accelerate the wrong decisions. How is your team currently drawing the line between using AI to think faster and using it to think less?

---

At grzzly, we work with marketing and product teams across Southeast Asia who are navigating exactly this tension — faster tooling, higher design complexity, and audiences who have very little patience for experiences that weren’t built with them in mind. If you’re rethinking your UX research process, your design system foundations, or how your creative decisions connect to commercial outcomes, we’d enjoy that conversation. Let’s talk