Incident Update 2︱Online Shockwaves: Reactions to the Charlie Kirk Incident on X in Canada and Abroad
Author(s) & Organization
Zeynep Pehlivan
Esli Chan
Organization: Media Ecosystem Observatory
Key takeaways
Lifespan of conspiratorial and partisan narratives was short: During the initial shock of the shooting, conspiratorial and partisan narratives emerged online globally, but quickly faded in engagement. This demonstrates how outrage-based narratives can emerge and fade just as quickly.
Clear topic segregation between news outlets, influencers, and politicians in Canada: During the peak of the Kirk-shooting discourse, news outlets provided key informational updates while influencers guided the political commentary, and politicians picked and chose their spots—engaging when it fit official narratives and leaving contentious or ritualized threads to newsrooms and creators. This reflects a division of communication roles, with each group contributing differently to the overall flow and interpretation of information.
While Canadian conversation mirrored the US, attention was limited: The thematic structure of discourse in Canada closely mirrored trends in the US; yet, the volume and degree of engagement was comparatively reduced. This demonstrates that while the Canadian information ecosystem is highly susceptible to political events in the U.S., its distinct media structures and actor dynamics can help moderate the intensity and persistence of heavily polarized online debates.
Introduction
In this incident update, we evaluate how the death of Charlie Kirk, a prominent right-wing American commentator, has provoked widespread online debate. Controversy around the legacy of Kirk has sparked discussion around the increasing partisan divide, ideological polarization, and threats of political violence. This discourse, while originating from an American context, has spilled over into the Canadian information landscape and subsequently influenced the health of the online space (e.g., insularity, segmentation, polarization). Our evaluation focuses on the topics that were most prominent and how they evolved into different interpretive narratives over time. We evaluate who captured most of the attention and shaped the conversation; and how partisan framing, free-speech arguments, and foreign-influence claims appeared in the discussion. Further, we assess how discourse around the Kirk shooting spread throughout the Canadian information ecosystem (CIE).
Figure 1. Distribution of posts on X over time
To do this, we created a dataset with the posts on X (formerly Twitter) mentioning “kirk” from September 5 to 21, 2025. Figure 1 illustrates the distribution of posts over time, with over 100,000 posts on the day of shooting on September 10, 2025.
Key questions
Our key questions focus on understanding the key topics and narratives that shaped the online discourse around the Charlie Kirk assassination. We ask the following two questions:
What did people talk about globally?
What did Canadians talk about?
Q1 : What Did People Talk About Globally?
We wanted to know which narratives dominated around the world, and how they rose and faded over time? To answer these questions, we assessed how different topics of conversation arose on X throughout the days following Kirk’s death, and the types of narratives used to frame and portray the event. To identify and group posts discussing similar themes, we employed topic modeling, a computational method that analyzes a large collection of text to automatically group together and identify recurring themes or subjects discussed in the ecosystem.
We find that the conversation on X quickly became a polarized battleground, shifting from immediate conspiracy theories and grief to sustained ideological warfare. Contrasting the number of posts (post volume) to engagement with these posts (likes), the dominant topics varied between the two. While high post volume centered on Israeli government involvement, the Epstein files, and the shooter's identity, audience reactions (likes) favored different topics. References to Elon Musk, Christianity/faith, and the Israeli government's response received the highest overall engagement and were exceptionally efficient at generating per-post likes, making them clear narrative outliers. The political Left's response, the (temporary) cancellation of the Jimmy Kimmel show, and Erika Kirk's public statements also generated significant attention, primarily driven by right-wing political commentators and news outlets.
The Four Phases of Narrative Evolution
We identified four distinct overlapping phases of public discourse on X following Kirk’s assassination. These phases include:
Phase 1: Initial Framing & Shock - High engagement framed around conspiratorial and outrage-based narratives.
Phase 2: Partisan Polarization - Left- versus right-leaning critique on responses to Kirk’s death.
Phase 3: External Narrative Dominance - Side conversations related to Kirk that emerged about wider cultural touchpoints and fallout.
Phase 4: Sustained Narrative Contestation - Persistent engagement on freedom of expression as a result of criticism versus support of Kirk’s legacy.
Table 1 below describes each of these phases in more detail (see Methodology section for information about how these phases were identified). For each phase, we describe the dominant topics, the nature of discussion and volume and engagement with the topics. The volume of posts and engagement over time corresponding to each phase are detailed in Figures 2 and 3. The phase determination process is outlined in the methodology section at the end of this update.
| Phase | Phase 1: Initial Framing & Shock | Phase 2: Partisan Polarization | Phase 3: External Narrative Dominance | Phase 4: Sustained Narrative Contestation |
|---|---|---|---|---|
| Dominant Topics of Discussion |
• Transgender identity of the shooter • Israeli Government’s involvement in Kirk’s death • Relation to the Epstein Files |
• Liberals and Democrats’ responses • Celebration of Kirk’s death |
• The cancellation of Jimmy Kimmel’s show • Erika Kirk’s response |
• Debates around free speech • Critiques of Kirk’s political stances (e.g. DEI) • Empathy and support for Kirk’s family |
| Nature of Discussion | Emerged instantaneously from the shock, focusing on the killer's link to their potential transgender identity and intent. The topics leveraged the initial shock for conspiratorial and outrage-based narratives. Attention was driven by Netanyahu's quick engagement and claims that the shooting was a distraction from the Epstein files. | A partisan lens to frame the political Left and Democrats as responsible for Kirk's death and societal violence. Political officials and commentators acted as critics, framing the celebration of Kirk’s death as morally corrupt. | Discussion shifted beyond the original event towards subsequent cultural touchpoints triggered by the death. Topics included the suspension of Jimmy Kimmel's show and Erika Kirk's public presence as the new CEO of Turning Point USA (TPUSA) and her forgiveness of the shooter. | This phase is characterized by a steady and underlying stream of discourse for the following topics, demonstrating ongoing debate. The Right (amplified by figures like Trump) framed Kirk as a martyr of free speech, while the Left highlighted perceived hypocrisy in right-leaning free speech advocacy given their crackdown on DEI-related educational materials. Free speech became a persistent battleground used by both sides to contest the other's credibility. Support for Kirk’s family was largely expressed through empathetic language, e.g., “praying for the Kirk family”. |
| Volume (number of posts) and Engagement (average likes per post) |
Posts surged on September 10 (the day of the shooting) to approximately 8,000 posts and spiked at around 10,000 posts on Sept. 11 (the day after the shooting).
Posts dropped sharply one week later (Sept. 18). Conspiracies attracted quick, short-lived attention.
Engagement was high early on but dropped sharply as speculation faded. |
Posts surged alongside Phase 1 on September 10 also at roughly 8,000 posts, then began dropping at a steady pace until September 22nd.
However, engagement remained at relatively lower levels, around 1,000–2,000 likes per day. |
Two delayed spikes revealed a revival in the narrative, with posts surging on September 17–18 (Kimmel's suspension/Erika Kirk’s public statement) and again on September 21 (Erika Kirk's message of forgiveness during the memorial).
Despite limited posts, engagement peaked on September 12 with Erika Kirk's first public statement. |
Posts emerged with Phases 1 and 2, with slightly less volume (over 5,000 posts) and held a stable average of posts per day until Sept. 20.
Engagement fluctuated over time but maintained at a moderate level of 1,000–4,000 likes per day. |
Figure 2. Volume of posts mentioning “Kirk” on X from September 5–21, 2025, segmented into four phases: Phase 1—Initial Framing & Shock; Phase 2—Partisan Polarization; Phase 3—External Narrative Dominance; Phase 4—Sustained Narrative Contestation
Figure 3. Likes (Engagement) on posts including “Kirk” on X from September 5 to 21nd, 2025 Phase 1—Initial Framing & Shock; Phase 2—Partisan Polarization; Phase 3—External Narrative Dominance; Phase 4—Sustained Narrative Contestation
Q2: What did Canadians talk about?
In Canada, what narratives emerged within the Canadian information ecosystem? Which actors led the conversation? How did discussion differ from the global dynamics? To answer, we analyzed Canada’s information ecosystem from Sept. 5–21. We identified 5,308 X posts mentioning “Kirk” from our existing Canadian seedlist, amounting to about 4% of total CIE activity.
When did the conversation peak in Canada? Examining the daily volume of posts, Figure 4 shows an early condolences spike (Sept. 10), a secondary peak related to government criticism and/or celebrations (Sept. 11), and, as offline gatherings reignited attention, a third peak, associated with vigils. As compared to the global conversational trends, Canadian engagement follows the same initial interest in the topic with increased engagement on September 10 and 11. However, Canadian online activity decreases at a relatively higher rate - while global engagement is prolonged.*
Figure 4. Distribution of number of posts for top 5 topics in the Canadian Information Ecosystem
Who talked about what? The most prominent topics in Canada following the event include: the Canadian government’s response, vigils and memorials happening across Canada, condolences for Kirk and his family, and updates and reactions to Trump’s response to the event. However, the emphasis on these topics varied significantly across actors.
Figure 5 outlines the most prominent topics influencers, news outlets, politicians and civil society organizations posted about (posts) and engaged with (likes) over the period of study. Examining the left panel of Figure 5, News outlets acted as the main information source, providing informational updates regarding the arrest of the suspect and evolution of the incident. Influencers were primarily commentators, dominating the conversation related to the celebration and condemnation of Kirk’s legacy and how the Canadian government responded to the event. Politicians were noticeably absent in discussion around polarization and blame, vigils and memorials happening in Canada, and the nature of the suspect and their arrest. The lack of engagement from politicians left space for newsrooms and creators. Lastly, civil society organizations engaged the least - those who posted were largely right-leaning Canadian groups who remarked on the positive legacy of Kirk and condemned political violence.
Figure 5. Account-normalized posts and likes by topics** in the Canadian Information Ecosystem
How is engagement distributed? Looking at the right panel of Figure 5, Politicians captured the most attention on topics such as the alleged involvement of the Israeli government (98.7% of the topic share), media updates (78.0%), and condemnation of political violence (47.8%), setting the groundwork for high-salience political debates. News outlets then amplified and framed these issues, gaining attention on updates regarding Kirk’s previous statements related to identity politics (67.3%) and Erika Kirk’s public statements (70.6%), thereby legitimizing certain narratives and shaping how issues are interpreted. Influencers, in turn, drove reactions within commentary spaces by translating these political and media narratives into affective, emotionally resonant stories that circulated within their online communities.
These findings demonstrate how different actors in the online ecosystem each play unique roles in diffusing and amplifying various elements of a story: news outlets often establish the factual foundation, influencers amplify emotional and interpretive angles, and political actors largely contribute to meaning-making. Together, they create a multi-layered and emotionally charged information environment, where facts, reactions, and meaning-making circulate and reinforce one another.
Conclusion
The online discourse surrounding the death of Charlie Kirk on X demonstrates the dynamic and polarized nature of contemporary political narrative diffusion. The incident did not result in a single, unified conversation but rather a four-phased evolution characterized by distinct patterns of volume and engagement.
In Canada, the conversation was small, only about 4% of the total X activity analyzed, but mirrored the polarized structure. The Canadian information ecosystem's response to Kirk’s shooting showed a sophisticated and layered form of polarization: news accounts moved facts quickly while influencers/politicians pushed partisan frames. The Canadian space is shaped by U.S. discourse, but its impact is moderated by local structures and a smaller, less-engaged politician cohort, which limits the intensity compared to the U.S.
Methodology
Topic modeling
The present study adopts BERTopic, a modular pipeline that combines transformer-based embeddings with non-parametric clustering to yield coherent, human-interpretable topics. In the first stage, each text instance is converted into a high-dimensional embedding using the multilingual variant of MiniLM fine-tuned for paraphrase similarity. These embeddings are then projected to a low-dimensional manifold with UMAP, which preserves the local neighbourhood structure while substantially reducing computational complexity for subsequent steps. Topic discovery is performed with HDBSCAN, a density-based algorithm that automatically determines the number of clusters and isolates noise points; its minimum cluster size parameter is scaled to corpus size to balance granularity and stability.
Once clusters are identified, BERTopic constructs a sparse class-based TF-IDF (c-TF-IDF) representation that contrasts the vocabulary of each cluster with the entire corpus, thereby highlighting discriminative terms even in highly imbalanced data. To enhance interpretability, the framework ensembles multiple representation strategies: a KeyBERT-inspired extraction ranks candidate keywords by their cosine similarity to the centroid embedding, while a Maximal Marginal Relevance step re-orders them to maximise semantic coverage and minimise redundancy. The top keywords form concise, descriptive labels that are appended to the numerical topic identifier, and each document inherits the label of its most probable cluster together with an assignment confidence.
The phase determination process
The 11 most active topics, collectively covering 90% of the data, were analytically regrouped into four distinct Narrative Phases to assess the evolution and competition between different thematic arcs.
The phase determination process moves beyond simple chronological blocks and is grounded in two core criteria:
Temporal Co-dominance (Quantitative): We identified time periods where specific clusters of topics exhibited a synchronous spike in post volume, indicating a shared, temporary shift in collective attention. This was used to establish the preliminary boundaries where the topic distribution changed drastically.
Narrative Cohesion (Qualitative): Topics identified in the same time frame were grouped only if they shared a unified thematic purpose or were directly related by cause-and-effect in the ensuing debate.
This approach treats the "phases" as analytical constructs designed to group competing narrative arcs that inform and react to one another, offering insight into the overall structure of the discourse rather than a strict, linear timeline.
*We note that the data sources for analysis of the global response (all X posts mentioning ‘Kirk’) are different from the Canadian response (all X posts mentioning ‘Kirk’ within our existing Canadian seedlist). The analysis here is a narrative assessment rather than an empirical data comparison.
**Values show the breakdown by actor type (influencer, politician, etc.) within each topic. For each topic × actor type, we count posts per account and likes per account, take the mean for that actor type, then rescale within the topic so shares sum to 100%. (Applied separately for posts and for likes.)