Seeking Help, Facing Harm
Auditing TikTok mental health recommendations under help-seeking vs distress intent
Research question
Do recommender systems distinguish between help-seeking and distress expression within the same sensitive domain, or do they collapse both into an undifferentiated “mental health interest” signal?
Audit design
We run a controlled 7-day audit of TikTok’s For You Page using 30 fresh accounts and LLM-guided agents. The design crosses:
- Search framing: distress-initiated vs help-initiated
- Interaction strategy: MH-engaged vs MH-avoidant vs passive-observer
This yields six experimental conditions and isolates stated intent (search) from revealed behavior (watch/skip patterns).
Key findings
1) Behavior dominates outcomes. Engagement rapidly saturates feeds with mental health content (about 45% of daily recommendations), while avoidance and passive viewing reduce but do not eliminate exposure (about 11–20%).
2) Help-seeking changes the mix, not the risk. Help-initiated searches yield more potentially supportive material, but potentially harmful content persists at low but non-zero levels, including suicide/self-harm content. Help-seeking does not act as a safety boundary.
3) Avoidance reduces volume, not safety. Suppressing mental health content lowers overall exposure, but does not reliably improve what the remaining content contains.
Why this matters
The results suggest limited sensitivity to user intent signals in a sensitive domain, and motivate context-aware safeguards that treat help-seeking as safety-relevant rather than just topical.