The goal of this analysis is to uncover the deeper structure of storytelling that shapes public understanding of the conflict, and to assess how that storytelling aligns with the CBC’s stated mandate and statutory obligations as Canada’s public broadcaster. Absolute neutrality in media framing is difficult to achieve, particularly in situations of war and atrocity. Journalists inevitably make choices about language, imagery, sequencing, and voice - choices that shape meaning and interpretation even when reporting is presented as factual or impartial, which is often the case, implicitly or explicitly, when it comes to public broadcasters.

This narrative framing analysis therefore matters for several reasons. First, it allows for a systematic examination of how editorial decisions such as sourcing practices, quotation patterns, and linguistic framing, accumulate over time to produce consistent interpretive patterns. Second, it makes the effects of media bias more legible to a general audience. Readers may not consciously track patterns across hundreds or thousands of articles, but they often register when coverage feels persistently one-sided, incomplete, or emotionally imbalanced. Narrative analysis helps explain why. Finally, this approach situates CBC’s coverage within a broader cultural environment, showing how its reporting both reflects and reinforces particular discourses surrounding the Israel-Hamas conflict.

Taken together, the findings of this report provide a multidimensional portrait of CBC’s coverage that moves beyond isolated articles or surface-level observations to examine how meaning is produced over time, within the larger corpus of the CBC’s coverage of the war. By focusing on narrative structure rather than isolated data points in individual articles, this report reveals not only what the CBC reported, but how its storytelling shaped public understanding of one of the most consequential and emotionally charged conflicts of our time.

Methodology:

This section outlines the technical and analytical methods used to conduct the study, including data preparation, classification procedures, and corpus-level analysis techniques. Readers primarily interested in the findings and their implications may wish to proceed directly to the Results section on page 53.

Data Collection:

For the purpose of this research, we used the dataset collected by Innohives, as described in Part I of this report booklet. Briefly, 2,789 CBC articles were collected with two complimentary collection strategies, both relying on keyword-based retrieval.

The resultant product is a final, manually validated dataset of CBC articles that robustly captures media discourse on the Israel-Hamas war and related issues of antisemitism within the defined timeframe.

Data Analysis:

To conduct the narrative analysis, we relied on a large language model (LLM), specifically ChatGPT-5, accessed via the OpenAI API. The model was used with default generation parameters; no temperature, top-p (nucleus sampling), or other stochastic sampling settings were explicitly specified by the researchers. As a result, the model operated using the API defaults in effect at the time of execution (October-November 2025). Model usage was deterministic at the prompt level in the sense that identical inputs were processed using the same fixed instructions and without parameter variation introduced by the researchers. For analysis, articles were exported from the main dataset into Python10, loaded into Google Colab using the pandas library11, and then filtered and grouped according to the specific research questions. The model’s outputs may reflect aggregate computational analysis across the full corpus, or article-by-article analysis, as indicated within the discussion of each research question, below. Because the framing analysis requires a nuanced assessment of narrative tone, framing, and linguistic choice, we supplemented the LLM’s interpretive work with targeted Natural Language Processing (NLP) tools, including keyword recognition, contextual pattern matching, and sentiment classification.

The LLM’s outputs were documented in summary tables, where we recorded the types of actors mentioned, how they were characterized, and any observed patterns in tone or narrative framing. This process allowed us to track recurring themes, compare narrative treatment across groups, and identify patterns that would not have been visible from quantitative results alone.

Research Question #1-- Civilian Perspectives in CBC Coverage: Whose Voices were Heard?

To assess how consistently civilian perspectives were included in CBC’s coverage, we conducted a systematic analysis of the full dataset of 2,789 CBC articles published between October 2023 and June 2025, using a large language model (ChatGPT-5) capable of natural-language interpretation. The analysis was conducted at the corpus level, with article-level annotation as the unit of analysis. Each article was processed individually using a fixed analytic prompt (see Appendix A) applied uniformly across the dataset, and the resulting classifications were aggregated to identify patterns across the corpus as a whole.

The prompt was applied consistently to each article’s full text, directing the model to identify quoted speech, determine whether the speaker qualified as a civilian under the specified definition, and, where applicable, classify the civilian speaker as Israeli, Palestinian, or other/unclassified. The instruction to classify conservatively under uncertainty was central to this stage of the analysis: when the speaker’s status could not be clearly established, the quote was excluded from the civilian category, and when civilian status was clear but group identity was ambiguous, the quote was classified as other/unclassified. As a result of this conservatism, some civilian quotations may be undercounted. For this reason, the analytical emphasis is not placed on absolute quotation totals, but on the consistency and direction of the representational patterns that emerge across the corpus.

Although articles were analyzed individually, the purpose of this procedure was not to interpret or evaluate specific articles in isolation. Rather, article-level annotations served as the basis for corpus-level aggregation, allowing for the identification of systematic patterns in the presence, absence, and distribution of civilian voices over time. By comparing the frequency and temporal distribution (how often and when) of Israeli and Palestinian civilian quotations, both overall and within the January-June 2024 period, this method provides insight into whose lived experiences were consistently foregrounded in CBC’s reporting and whose were comparatively marginal or absent.

Research Question #2: The De/Personalization of Actors: How does CBC Frame Human Identities?

To examine how CBC’s reporting constructs or obscures the human identities of those involved in the Israel-Hamas conflict, we conducted a systematic linguistic analysis focused on patterns of humanization and dehumanization in news coverage. Dehumanization in media discourse does not typically operate through overt language, but through cumulative representational choices about whether actors are presented as people with lived experiences or as abstract collectives, institutions, or instruments of power. These choices shape readers’ emotional engagement, moral identification, and capacity for empathy, particularly in conflict reporting where language mediates access to suffering, vulnerability, and agency.

The analysis was conducted at the corpus level, using article-level annotation as the unit of analysis. Each article in the dataset was processed individually using a fixed analytic prompt (see Appendix A) applied uniformly across all texts. For each article, the model identified whether Israelis and Palestinians were referenced in individualized, collective, or mixed terms. These article-level classifications were then aggregated across the full corpus to identify systematic patterns in how human identity was foregrounded or obscured over time. This approach allows for large-scale pattern detection while ensuring that observed asymmetries arise from consistent treatment of each article rather than from holistic or interpretive batch judgments applied to the corpus as a whole.

Research Question #3: Lexical Collocation: Who Gets to be Human?

To assess how CBC’s language constructs the human presence of Israelis and Palestinians within its reporting, we conducted a systematic lexical collocation analysis of the full dataset of 2,789 CBC articles published between October 2023 and June 2025 (See Appendix A for prompts used). This stage of the analysis focused on the semantic environments surrounding references to people - specifically the terms “Israeli(s)” and “Palestinian(s)” - rather than on state actors or geographic entities. The analysis was conducted at the corpus level, treating the dataset as a single body of text, in order to identify aggregate patterns of language use that would not be visible through close reading of individual articles alone.

Using rule-based text processing, every occurrence of the terms “Israeli,” “Israelis,” “Palestinian,” and “Palestinians” was identified across the corpus. For each occurrence, a context window of ten words preceding and ten words following the target term was extracted. These context windows were then pooled separately for each group, producing two aggregated lexical environments. Common English stop words and non-meaningful function words were removed by the LLM, leaving only meaning-bearing content words that contribute to semantic framing, tone, or thematic association. The frequency of these content words was then calculated to generate ranked lists of the most common collocations associated with each group. This approach allows for an empirical comparison of the linguistic contexts in which Israelis and Palestinians are embedded, revealing systematic differences in how each population is framed and understood. By focusing on aggregate lexical patterns rather than individual narratives, this method captures the cumulative effects of language choice that shape humanization, abstraction, and moral salience in conflict reporting.

Research Question #4: Naming Power: Whose Lives and Voices Are Individuated in CBC Coverage?

To further examine how CBC’s coverage distributes visibility, authority, and human presence across actors in the Israel-Hamas conflict, we conducted a systematic analysis of explicitly named individuals appearing in news articles. Naming is a powerful narrative device: individuals who are named are rendered legible, memorable, and socially real, while those who remain unnamed are more easily absorbed into collective or abstract categories. In conflict reporting, patterns of naming can therefore signal whose experiences are individualized, whose authority is foregrounded, and whose presence is treated as incidental or interchangeable.

This stage of the analysis was conducted at the article level, with each article processed independently using a fixed analytic prompt applied uniformly across the dataset (See Appendix A). The model was instructed to identify all explicitly named individuals within a single article and to classify each by both group affiliation and role type. By distinguishing between civilians and official or institutional representatives, and by separately tracking Israeli, Palestinian, and other affiliations, this approach allows for a granular assessment of how often named human subjects appear as private individuals versus institutional actors. Aggregated across the corpus, these classifications reveal structural patterns in whose names are attached to lived experience, whose are attached to power or authority, and whose are absent altogether.

Research Question #5: Representation in Domestic Coverage: Who Speaks for the Community?

To assess how Jewish voices are represented within CBC’s domestic coverage of the Israel-Hamas war and its spillover effects in Canada, we conducted a systematic analysis of quoted speech in articles focused on Canadian events. Quoted speech plays a central role in shaping public understanding of social conflict: it determines whose perspectives are granted legitimacy, authority, and emotional resonance, and whether particular communities are portrayed as agents, commentators, or subjects of concern. In the context of heightened domestic tension, including protests, campus encampments, and public debates about antisemitism and Israel, patterns of quotation provide a critical window into how media institutions allocate voice, credibility, and representational power.

This stage of the analysis was conducted at the corpus level, with individual articles processed only to enable consistent filtering, quotation extraction, and speaker classification. The analysis proceeded in multiple stages (see Appendix A). First, the full dataset was filtered to identify articles whose primary focus was on domestic Canadian events, excluding coverage centered on international developments abroad. From this subset, all instances of quoted speech were extracted and treated as individual observations. Each quotation was then attributed to a likely speaker and classified according to group identity and role type, with particular attention to Jewish speakers. For Jewish quotations only, an additional layer of institutional classification was applied to distinguish between private civilians and speakers affiliated with religious, advocacy, educational, political, or other institutional contexts. Aggregated across the domestic corpus, this approach allows for a systematic examination of how Jewish voices appear in Canadian coverage: whether primarily as private individuals or institutional representatives, how often they are quoted relative to other groups, and what kinds of authority or social roles they are most often assigned.