At Inithouse — a studio shipping a growing portfolio of products in parallel — we build tools that help people understand themselves better. When we launched Origin Of You, a self-discovery app combining five personality frameworks into one AI-generated portrait, we faced a classic UX problem: how long should the quiz be?
The length–accuracy trade-off
Short quizzes feel snappy but produce shallow results. Long quizzes produce richer data but people abandon them midway. We tracked completion rate as our primary metric from day one — because a personality analysis means nothing if nobody reaches the end.
Our first prototype had 80 questions. Completion rate sat around 30%. We cut to 45 questions. Completion jumped, but the AI portrait quality dropped — too few data points for the five systems we map to (MBTI-style cognitive functions, Big Five, Enneagram, attachment style, and a values framework). We landed on a structure that collects 120+ data points from fewer questions by varying question types.
Three question types, one quiz
Instead of endless Likert scales, we mixed three formats:
Multiple choice — quick, low cognitive load. Works well for binary-ish personality dimensions. "When you enter a room full of strangers, you usually..." with four options. One question, two data points (extraversion + social anxiety proxy).
Scaled response — a 5-point agreement scale for nuanced dimensions like openness or conscientiousness. We limited these to clusters of 3–4 in a row, then broke the pattern with a different format. Monotony kills completion.
Scenario-based — short vignettes: "Your friend cancels plans last minute for the third time. You..." These map to multiple systems simultaneously. One scenario question can feed data into Enneagram type, attachment style, and conflict resolution pattern. High data density per question.
The mix keeps users engaged. We observed that scenario questions had the highest per-question dwell time but also the lowest skip rate — people found them interesting enough to think through carefully.
Mapping logic: questions to systems
Each question feeds into one or more of the five personality systems. We built a weighted mapping matrix where every answer option carries scores across relevant dimensions.
A scenario about handling criticism might contribute to:
- Big Five Neuroticism (emotional reaction weight)
- Enneagram core fear (which fear is triggered)
- Attachment style (withdrawal vs. pursuit pattern)
The weights aren't equal. Some questions are "anchor" questions — high-confidence indicators for a specific system. Others are "signal" questions that add small increments across multiple dimensions. We found that 8–10 anchor questions per system, supplemented by shared signal questions, gave us enough confidence for a meaningful AI portrait.
This approach meant we could collect data for five systems without asking five times as many questions. The quiz feels like one coherent experience, not five tests stapled together.
What we measured
Completion rate was the headline metric, but we tracked more:
- Drop-off curve — where exactly people quit. Early exits meant the intro was wrong; mid-quiz exits meant fatigue; late exits meant the final stretch dragged.
- Time per question type — scenario questions took roughly 12 seconds, multiple choice around 5, scales about 4. We used this to pace the quiz: start with quick wins, place scenarios in the middle when engagement peaks, end with easy scales.
- Result satisfaction — a simple thumbs up/down after seeing the portrait. We measured whether longer engagement (slower, more thoughtful answers) correlated with higher satisfaction. It did.
Across our portfolio at Inithouse, we've observed similar patterns in other products. When we built the card selection flow at Tarotas — a tarot web app in five languages — dwell time on the result page was the real engagement signal, not just completion. And at Here We Ask, our conversation card game, we learned that pacing between question types matters more than total count.
What we'd change
If we redesigned the quiz today, we'd add adaptive branching — skipping questions the model already has high confidence about. We'd also test a "progress preview" showing which personality dimensions are taking shape as you answer.
Quiz design for personality tools isn't about asking the right questions. It's about asking the fewest questions that still produce results people recognize as true.
Try the quiz yourself at Origin Of You — five personality systems, one portrait.
Inithouse is a lab building many products at once. This is part of our Building in Public series.
Top comments (0)