We use cookies to understand how you play. No weird stuff.
Analytics help us improve quizzes. Privacy policy
You clicked. That makes you one of the nerdy ones. Here's exactly how every quiz on QuizMagz gets built - from raw topic to polished result.
Every quiz on QuizMagz is generated by a pipeline of 6 specialized AI agents - not a single prompt, not a template. Each agent has one job, and the output of each step feeds into the next. A topic suggestion wins the community vote, gets picked up by the pipeline, passes a hard quality gate, and only then gets published. Here's what happens in between.
Topic Analysts
Analyzes the domain, generates a pool of candidate outcomes, and picks topic-specific personality dimensions for this quiz.*
Outcome Architects
Selects the final set of outcomes - scaled to the topic - and builds each one's dimension profile, its personality fingerprint across up to 7 dimensions.*
Question Engineers
Writes a pool of binary slider questions sized to the topic. Both poles feel like real answers. No obvious right choice.
Question Editors
Active rewriter. Removes duplicates, enforces tonal variety, and trims the pool to the final question set.
QA Reviewers
Hard pass/fail gate. Multiple criteria, each scored independently. Failure triggers a retry.
Content Writers
Writes result headlines, descriptions, and share text. The words you read at the end.
Same concept as MBTI - personality dimensions that combine to produce distinct types. Every quiz uses between 4 and 7 of its own, scaled to the topic. Every question nudges your position across these dimensions. When you finish, your profile is compared to each outcome using cosine similarity. The closest match wins. No single answer locks in your result.
After the agents finish, a deterministic TypeScript simulation runs 5,000 playthroughs of the quiz. It checks that rare outcomes are actually rare, common ones are common, and no outcome is unreachable. No LLM, no randomness - pure math. If the distribution looks off, the pipeline can regenerate before publishing.
One-shot generation produces generic outputs - the kind of quiz where every result says you're adventurous and creative. A multi-agent pipeline means each step is reviewed before the next begins. The QA gate is the only reason a quiz gets rejected instead of quietly published as mediocre.
That's it. Now go take a quiz.