AI for UX Researchers Quiz

Answers and explanations appear as you take the quiz.

1. What is the core limitation of synthetic users?
   • They are too expensive to run
   • They give plausible, averaged answers from training patterns rather than lived experience
   • They can only answer in English
   • They always refuse to answer interview questions
2. Which is a reasonable use of synthetic users according to the guide?
   • Replacing all usability testing
   • Reporting findings to executives as customer evidence
   • Piloting a discussion guide to find confusing questions
   • Setting product pricing
3. Why should researchers ask AI for verbatim quotes with timestamps?
   • So each quote can be checked against the source, since models can misattribute or invent quotes
   • Because timestamps make reports look more professional
   • Because AI cannot summarize without timestamps
   • To shorten the transcripts
4. What does Cohen's kappa measure in AI-assisted tagging?
   • The speed of transcription
   • The number of themes in a dataset
   • The sentiment of each comment
   • Agreement between model labels and human coding, adjusted for chance
5. Why is an AI summary not the same as research synthesis?
   • Synthesis weighs evidence, notices contradictions and ties findings to decisions, while summaries can smooth over outliers
   • Summaries are always longer than synthesis
   • Synthesis can only be done on paper
   • There is no difference
6. Why can't synthetic users replace usability testing specifically?
   • Usability tests require video cameras
   • They cannot show what people actually do, only generate what someone might say
   • Usability testing is illegal with AI
   • Synthetic users are too critical of designs
7. When should a codebook for AI tagging be written?
   • After the model finishes tagging
   • Never, because models create their own codes
   • Before the model sees the data, with definitions, examples and counterexamples
   • Only if the dataset is under 100 items
8. What role do embeddings play in analyzing large feedback sets?
   • They encrypt participant data
   • They record audio
   • They replace the need for a researcher to read feedback
   • They represent text numerically so similar comments can be clustered