Product design question

How would you determine how to rank posts in the newsfeed?

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

Tests understanding of ranking system design: what signals to use and how to balance engagement against long-term satisfaction.

Clarify the ranking goal: maximize meaningful engagement and time well spent, not just raw clicks or time spent, since Facebook has explicitly said this.
List candidate signals: relationship strength (interaction history with the poster), content type affinity, recency, and predicted engagement probability (like, comment, share).
Weight signals by predictive power for a satisfaction proxy, e.g. a survey-based 'meaningful interaction' score, not just click-through rate.
Address the tradeoff: pure engagement-optimized ranking favors outrage and clickbait, so add a penalty for content flagged as low-quality or misinformation.
Define success via a blended metric: session-level satisfaction survey score plus long-term retention, not single-post engagement.

Names concrete ranking signals (relationship strength, content-type affinity, recency, predicted engagement) rather than a vague 'use ML to rank'.
Explicitly raises the known tension between engagement-optimization and content quality, and proposes a countermeasure (quality penalty).
References Meta's own public shift to 'time well spent' as validation this tradeoff is real and previously acted on.
Separates the ranking model's training objective from the ultimate success metric, since optimizing directly for clicks causes exactly the outrage problem.

Proposing a ranking system that purely maximizes engagement without addressing the well-known clickbait/outrage side effect.
Giving a generic 'machine learning model' answer without naming concrete signals or how they'd be weighted.

How would you detect and down-rank clickbait or misinformation in this system?
How would you validate the ranking model isn't overfitting to short-term engagement?