Product design question

How would you create a feature on Facebook to combat depression, suicidality and self harm?

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

Product design in an extremely high stakes, sensitive domain, requiring careful balance of detection, intervention, and user trust.

Clarify the scope: detecting signals of depression, suicidality, or self harm in posts and connecting the user to appropriate help, not diagnosing.
Identify stakeholders: the at risk user, their friends and family who might also see concerning posts, and crisis support organizations.
Design detection carefully: use a combination of automated language pattern detection and user or friend reporting, since automation alone risks false negatives on subtle signals.
Design intervention with care: when a signal is detected, surface a supportive, non alarming prompt connecting the user directly to crisis resources like a hotline, not a public flag.
Involve mental health experts and crisis organizations in designing the actual intervention language and flow, not just engineering.
Define success as connection rate to crisis resources when triggered, and false positive rate kept low enough to avoid user distrust, tracked separately, never total post removals as a success metric.

Treats this as a support and connection problem, not a content moderation or removal problem, showing the right instinct for this domain.
Involves outside experts, crisis organizations and mental health professionals, rather than assuming engineering can solve this alone.
Gives an illustrative number, for example assuming a target of connecting flagged users to a resource within seconds of detection.
Explicitly avoids using post removal or account suspension as a success metric, which would be actively harmful here.

Treating this primarily as a content moderation problem to remove or hide posts rather than a support intervention.
Proposing fully automated intervention without human or expert oversight, which is unsafe in this domain.

How would you avoid false positives that make users distrust or avoid the platform?
How would you measure the real world impact of this feature responsibly?