TikTok 'Suggested Accounts' experiment alleged to amplify or suppress certain creators

In February 2020 an AI researcher reported that TikTok’s "Suggested Accounts" feature recommended other creators who looked similar to the account a user had just followed, raising concerns about feedback loops and visibility bias for creators. TikTok disputed the claim and said recommendations are based on collaborative filtering. Independent news outlets reported the researcher’s experiment and the platform response.

TikTok (ByteDance) · Incident Feb 24, 2020 · Indexed Jun 10, 2026 · 3 sources

Collaborative filtering and automatic featurization allegedly created feedback loops that reinforced visible traits and affected who gets recommended.
What
In February 2020 an AI researcher reported that TikTok’s "Suggested Accounts" feature recommended other creators who looked similar to the account a user had just followed, raising concerns about feedback loops and visibility bias for creators.
Incident date
Feb 24, 2020
Who
TikTok (ByteDance)
Failure mode
Brand & Safety Incident
AI surface
Recommender
Severity
Medium

What happened

A researcher (Marc Faddoul) published a casual experiment in February 2020 showing that when a fresh TikTok account followed a creator, the app’s "Suggested Accounts" recommendations often returned visually similar creators, matching attributes such as ethnicity, hair color, or age. News outlets replicated the basic experiment and reported similar anecdotal results, while TikTok told reporters it could not replicate the findings and said recommendations use collaborative filtering. The reporting framed the observation as a potential feedback loop that could reinforce existing popularity and reduce diversity of visible creators.

What broke inside the model

Failure path · mode profile · Brand & Safety Incident
  1. 01 · TriggerA user prompts the model in public view.
  2. 02 · Model stepThe model produces unsafe or off-brand output.
  3. 03 · Control gapNo filter holds the line before publish.
  4. 04 · FailureThe output goes public unchecked.
  5. 05 · ConsequenceA reputational or safety incident lands.

A contained signal crosses into output that goes public.

The alleged mechanism was recommendation logic that relied on collaborative filtering and/or automatic featurization of profile images, producing correlations between followed accounts and suggested accounts. That process can create a feedback loop where signals in followers’ behavior or image-derived features are amplified, unintentionally privileging creators who already receive more follow activity or share visible traits. TikTok denied deliberate use of race or profile-photo features in recommendations, attributing results to user-behavior patterns.

Public visibilityHigh
Regulatory exposurePossible
Customer impactMany customers
Financial impactUnknown
Time to disclosureDays
  1. PressThere’s something strange about TikTok recommendationsvox.com
  2. PressTikTok's Algorithm Shows Unintentional Racial Bias, Researcher Findsbuzzfeednews.com
  3. PrimaryIncident 117: TikTok's "Suggested Accounts" Algorithm Allegedly Reinforced Racial Bias through Feedback Loopsincidentdatabase.ai
Permalinkhttps://failureindex.ai/failures/tiktok-suggested-accounts-experiment-alleged-amplify
CitationAI Failure Index. "TikTok 'Suggested Accounts' experiment alleged to amplify or suppress certain creators" (FI-0396). Realm Labs. https://failureindex.ai/failures/tiktok-suggested-accounts-experiment-alleged-amplify (indexed Jun 10, 2026).
Share cardA branded image of this record for posts and slides.

Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0396. Full dataset at /data.

Note from Realm Labs, the Index steward

How Realm fits

Controls for this failure mode
  • Prism
  • OmniGuard
  • AI Detection & Response (AIDR)

This entry sits in the index's predictive wing: a system that scores, ranks, perceives, or steers rather than generates. Realm's runtime layer is built for the generative and agentic systems now moving into these same decision seats, where it watches a model's internal state and holds an unsupported claim or an unchecked action before it commits. The control gap on this record, an automated decision that reached people with no runtime check in front of it, is the same gap. The index keeps predictive failures on the record because the pattern carries straight into the systems shipping today.