The TikTok caliphate: How jihadist supporters exploit algorithmic recommendations and evade content moderation

Summary

This article examines how supporters of ISIS, al-Qaeda, and affiliated jihadist organizations circulate propaganda on TikTok despite the platform’s stated zero-tolerance policy on violent extremism. Drawing on a modified grounded theory analysis of over 300 pro-jihadist videos collected through an algorithmically primed research account, the authors inductively derive a typology of five evasion strategies that embed extremist messaging within TikTok’s participatory vernacular culture. They theorize this as “everyday extremism” — a collapse of spectacle into ritual and mundane platform aesthetics — and argue that effective response requires infrastructural critique of recommendation systems rather than reactive content removal alone.

Key Contributions

• One of the first in-depth empirical studies of jihadist activity on TikTok, addressing a literature skewed toward far-right extremism.
• An ideology-independent typology of five moderation-evasion strategies portable across platforms and movements.
• The concept of everyday extremism, capturing how propaganda dissolves into platform-native memes, sounds, and vernaculars rather than functioning as terrorist spectacle.
• A reframing of policy debate from content takedown toward infrastructural critique of TikTok’s algorithmic architecture.
• Concrete governance proposals, including TikTok’s inclusion in GIFCT, culturally informed human moderation, and user-driven annotation mechanisms.

Methods

The authors used a modified grounded theory approach combined with algorithmically assisted snowball sampling. A dedicated anonymous TikTok account, accessed through a US-based VPN, was used to “radicalize” the recommendation feed via Arabic and English hashtags, emojis, and references to canonical jihadist figures and events. Over 300 videos were bookmarked between January and July 2024 and analyzed multimodally (after Kress and Van Leeuwen) through iterative constant comparison, observational notes, and screenshots. A follow-up check in October 2025 found that only 25 of the original videos remained accessible, itself treated as analytical evidence of platform ephemerality.

Findings

• Five overlapping evasion strategies emerged: audio camouflage, meme infiltration, blurred intent, emoji codes, and bait-and-switch.
• Audio camouflage: pitch/tempo manipulation of canonical nasheeds (e.g., “Dawlati Baqiah,” “Tora-Bora”) and disguised metadata titles such as “Contains: My Love.”
• Meme infiltration: jihadist messaging grafted onto Roblox, Fortnite, and Inside Out 2 memes, or trending sounds like Mareux’s “Lovers from the Past” used to glorify perpetrators.
• Blurred intent: visual obfuscation of banned figures (Bin Laden, al-Baghdadi, al-Adnani) and ISIS iconography that defeats image classifiers while remaining legible to in-groups.
• Emoji codes: the black flag and raised index finger (tawhid) signal allegiance while bypassing text-based moderation.
• Bait-and-switch: videos open with Pride Month, July 4th fireworks, or Ronaldo highlights before pivoting to extremist content.
• Trivial textual alterations (e.g., “1S1 S”) bypass keyword filters that block direct ISIS-related searches, revealing basic design flaws in moderation infrastructure.

Connections

This paper extends the largely far-right focus of online radicalization scholarship into jihadist contexts on TikTok, and its typology of evasion strategies is directly comparable to work on adversarial creativity and platform vernaculars in adjacent communities — see Rothut2026-wt and Nangle2026-yo on far-right content circulation and moderation dynamics. The infrastructural-critique stance also resonates with platform-governance arguments in Bouchafra2026-ts, while its algorithmically assisted account-based methodology parallels approaches in Bailard2024-pj.

Podcast

A research-radio episode discusses this paper: Listen