Product design question

How would you use ML to improve Facebook 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

Whether you can propose a concrete machine learning application with a clear input, output, and training signal, not a vague 'use AI' answer.

Clarify what part of Newsfeed to improve: ranking order, content type mix, or spam and low-quality filtering, since ML applies differently to each.
Pick ranking order as the target, since it has the clearest measurable impact on engagement and satisfaction.
Define the ML task precisely: a model predicting probability of meaningful engagement, like a comment or long watch, for each candidate post per user.
Identify training signal: historical engagement data labeled with post type, poster relationship, and time spent, avoiding purely click-based labels which reward clickbait.
Address a known risk: engagement-only labels can amplify sensational content, so add a secondary model or penalty for predicted negative sentiment or misinformation likelihood.
Define success as a lift in meaningful interactions and time well spent survey scores, validated through an A/B test before full rollout.

Names a precise ML task, predicting meaningful engagement probability, rather than a vague 'personalize the feed with AI' statement.
Flags the real risk of engagement-only training labels amplifying sensational content, a documented issue for feed ranking systems.
Proposes validating through A/B testing before rollout, showing awareness that ML changes to ranking need careful measurement.

Saying 'use AI to personalize the feed' without specifying the actual prediction task or training signal.
Ignoring the risk that optimizing purely for engagement can amplify low-quality or sensational content.

Chapters of the AI PM course that teach what this question tests.