TikTok's recommendation algorithm appears to be promoting anti-LGBTQ videos, some of which encourage targeted anti-trans violence
Olivia Little / Media Matters for America :
Context & Ripple Effects
Media Matters' report is an early entry in what became a running audit file on TikTok's For You feed: three months later, researchers documented how hashtags, video effects, and music are weaponized to promote hate against Asians, women, Muslims, Jews, Black people, and the LGBTQ community (a USA Today-covered study), and by February 2022 TikTok had responded with LGBTQ-specific policy updates including a misgendering ban (its LGBTQ safety policy overhaul).
The throughline is that TikTok's moderation answers have trailed its recommendation engine's behavior: the company judged 14 of the implicated anti-LGBTQ accounts rule-compliant even as the report flagged videos encouraging targeted anti-trans violence, and later audits — CCDH's engagement-test finding that liking self-harm content pulls more of it (CCDH's rabbit-hole experiment) and Global Witness's finding of far-right skew in For You feeds ahead of Germany's elections (the Global Witness AfD audit) — kept the same mechanism under scrutiny.
First-order effects
- Trans users and creators are directly exposed to recommended videos encouraging violence against them, while TikTok's determination that 14 of the implicated accounts violate no rules means the offending content stays up under current enforcement.
- LGBTQ advocacy groups gain documented evidence that the harm is algorithmically amplified rather than confined to individual uploads, sharpening the case against TikTok's recommendation design.
Second-order effects
- TikTok faces pressure to write identity-specific rules — which it did with the 2022 misgendering ban and age-appropriate content tools — because blanket community guidelines proved insufficient to catch this content.
- Independent researchers and NGOs institutionalize the audit playbook: engaging-with-content tests like CCDH's and feed-bias measurement like Global Witness's become the standard way to demonstrate what internal moderation misses.
Third-order effects
- If the audit pattern holds, scrutiny shifts from individual videos to the recommendation system itself, putting algorithmic accountability — not takedown volume — at the center of platform regulation debates, as Global Witness's election-cycle findings already foreshadow.
- Platforms may be pushed toward proactive design choices (age-gating, feed-level filtering) rather than reactive policy patches, since each new audit cycle shows guidelines alone don't stop amplification.
The trend: Recommendation engines keep being caught amplifying hateful and politically biased content faster than platforms can moderate it, moving accountability debates from content takedowns toward algorithm-level oversight.