Product design question

How would you use AI on TikTok to improve content recommendations for new users who haven’t built a viewing history yet?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Understanding of the cold start problem in recommendation systems and how to solve it without waiting for behavioral data to accumulate.

State the problem precisely: the feed relies heavily on watch time and interaction signals, which do not exist yet for a brand new user.
Propose an onboarding interest signal: a short category selection or a few swipe to react starter videos to get an initial preference read.
Lean on content level signals for the first sessions, such as video metadata and how similar new users have responded, rather than personal history.
Rapidly incorporate real time in session signals like watch duration, rewatches, and skips from the very first videos shown.
Define success: time to first highly engaged session for new users, a proxy for the algorithm finding the right content quickly.

Names the cold start problem explicitly and proposes both an onboarding signal and fast in session learning, not just use AI.
Distinguishes content level signals from personalization signals that require history, showing real recommendation system understanding.
Uses time to engaged session as a meaningful, platform specific metric for cold start success.

Assuming the standard personalization algorithm just works from day one without addressing the lack of data.
No metric specific to new user cold start success, only general engagement metrics that blend new and existing users.

How would you balance an onboarding quiz's friction against its value for cold start personalization?
How long does the real system likely take to find a stable interest profile for a new user, and how would you measure that?