A new report released today by the Antisemitism Research Center (ARC) of the Combat Antisemitism Movement (CAM) has uncovered troubling evidence that Instagram’s recommendation algorithm can rapidly steer ordinary users from mainstream wellness and fitness content toward antisemitic conspiracy theories and extremist propaganda.
The study, titled “Algorithmic Escalation: From Self-Improvement Content to Antisemitism on Instagram,” found that users did not need to search for, engage with, or express interest in antisemitic content to be exposed to it. Instead, Instagram’s recommendation system repeatedly guided users from popular self-improvement topics such as nutrition, biohacking, discipline, and physical fitness into a growing stream of antisemitic narratives, coded hate content, conspiracy theories, and even translated Nazi propaganda.
The findings raise urgent questions about how social media algorithms can inadvertently accelerate radicalization by rewarding increasingly provocative content in pursuit of engagement.
To conduct the study, ARC researchers created two new Instagram accounts designed to simulate ordinary users entering distinct self-improvement communities. One account focused on wellness and biohacking content, while the other followed fitness and discipline-oriented creators. Both accounts exclusively engaged with mainstream, non-political content.
Over three consecutive days, researchers recorded the content served to each account during 45-minute browsing sessions.
The results were striking.
The wellness-focused account was shown 59 classifiable videos during the study period. More than 32 percent were categorized as either coded or explicit antisemitic content. By the third day alone, nearly one-third of all content served to the account consisted of explicit antisemitism, including nine openly antisemitic videos during a single session.
The fitness-focused account was served 71 classifiable videos, with 24 percent categorized as coded or explicit antisemitism. On the third day, researchers documented 17 antisemitic videos in a single browsing session.
Perhaps most alarming was how quickly the escalation occurred. Researchers observed antisemitic content being recommended to the wellness account during its very first session, before the account had developed any meaningful interaction history.
“You don’t have to search for antisemitic content to find it on Instagram,” said ARC Research Associate Oliver Marks. “Our findings show that users engaging with normal self-improvement posts are algorithmically guided toward virulent antisemitic narratives and conspiracy theories. When platforms optimize for engagement without sufficient safeguards, they can end up amplifying hate to vast audiences.”
The report highlights three particularly concerning trends:
- Speed: Antisemitic content appeared almost immediately, before the algorithm had sufficient data to justify personalized recommendations.
- Volume: Explicit antisemitic content became increasingly prevalent over time, eventually representing a significant portion of recommended material.
- Convergence: Despite beginning in entirely different content ecosystems, both accounts were directed toward the same antisemitic narratives, conspiratorial themes, scapegoating tropes, and, in some cases, the exact same videos.
According to the researchers, this convergence suggests the problem extends beyond isolated online communities and points instead to structural weaknesses within the recommendation system itself.
The report builds upon ARC’s previous investigations into online antisemitism, including the exposure of networks of more than 80 AI-generated fake “rabbi” accounts on Instagram, alongside dozens more operating on YouTube and TikTok. Following that investigation, Meta removed many of the accounts identified by CAM.
CAM is now calling on Meta to conduct an urgent review of Instagram’s recommendation systems and implement stronger safeguards to prevent users from being algorithmically steered toward antisemitic and extremist content.
All documented content is archived and linked HERE in full, allowing independent verification of all findings.