Enkin concluded: “There can be a tendency to believe that if an issue is not covered in a way that coincides with one’s view on the matter, it is biased. The reality is far more complex.” Thus we conclude that balance and fairness according to the CBC, especially on matters as longstanding and complex as the Israeli-Arab conflict, is achieved by an assessment of reporting over time. Rather than highlighting an article or report that might seem biased on face value, the CBC insists that a true assessment of its balance can be achieved by zooming out and looking at the bigger picture.
However, recent events suggest that public confidence in the CBC’s adherence to these very principles has come under renewed scrutiny. In October 2025, CBC President and CEO Marie-Philippe Bouchard appeared before the House of Commons Standing Committee on Canadian Heritage, where she was questioned about allegations of antisemitism within the organization and in its coverage of Israel and Jewish issues. During her testimony, Bouchard defended CBC’s journalistic practices and declined to initi-ate a formal investigation, stating that “we don’t have to stop everything to do a full investigation.” Bouchard further testified that should one have a problem with CBC’s coverage, they should contact the office of the Ombudsman, as this is the venue for filing complaints or accusations of bias.
As stated in the introduction to their JSP, Impartiality and Integrity are allegedly principles by which the CBC stands. This means that the CBC ostensibly commits to not promoting any one perspective, and avoiding any conflict of interest, in the name of facts, expertise, and credibility.
CBC’s Standards for Reporting on War, Terror and Natural Disasters:
According to the CBC, their coverage of any violence in instances of war, conflict, terror, natural disasters and crime “reflect[s] the reality of the situation.” This means that their “conflict reporting should contribute to an understanding of the reality on the ground, and tell the stories of the combatants and the people affected by the fighting. To do this we may choose to embed with the military or another faction in the struggle, or with other interest-ed parties.” This decision requires transparency, according to the CBC, which demands that the audience knows that embedding has occurred and thus reporting is restricted (and, moreover, the reasons for those restrictions must be made clear). If a decision to embed within or convey the perspective of armed factions has been made, this also means that the CBC should make an effort to cover the other perspective or side, within a reasonable timeframe.
Another subsection highlights the need for “Respect for the Suffering of Victims and their Family.” According to the CBC, their standards demand that in dealing with matters that concern victims of tragedy, they “carefully weigh both the public interest of full reporting and the need to show compassion and restraint. In such situations we are considerate and we use judgment.” When “Natural Disasters” occur, the CBC similarly aims to “provide useful information and context, especially for those most directly affected.” We see that while the CBC’s principles guide it to be accurate, providing full context, it believes that the primary way to do so is to seek out perspectives that may be underreported, more sensitive, in order to convey the full picture. As iterated in the aforementioned subsection, this means that one report may only convey one perspective, over time, all perspectives in a given incident are meant to be reported with the same depth and respect.
CBC and Matters of Religion and the Diversity of Canada:
According to CBC’s JSP, coverage of “Politics” in Canada by the news organization - whether election coverage, parliamentary matters, polling, and so forth - is provided with “equitable treatment” to all involved, though not necessarily equal broadcast time.
Concerning matters of “Language,” particularly diction and syntax, the CBC commits to “Respect and Absence of Prejudice.” This entails vocabulary that is on par with equal rights, but also respect for language when reporting concerns minority groups, because of their “influence on how minorities or vulnerable groups are perceived.” The news organization explains that their standards include:
"We avoid generalizations, stereotypes and any degrading or offensive words or images that could feed prejudice or expose people to hatred or contempt. Criminal matters require special care and precision.
When a minority group is referred to, the vocabulary is chosen with care and with consideration for changes in the language"
In effect, the CBC’s journalistic standards demand that their language choice is precise, clear, non-discriminatory nor inflammatory, and culturally sensitive, particularly with matters concerning vulnerable minority groups.
More broadly, these principles reflect the CBC’s mission “to reflect diversity,” which includes a commitment to ensuring that “All Canadians, of whatever origins, perspectives and beliefs, should feel that our news and current affairs coverage is relevant to them and lives up to our principles” and a “special responsibility to reflect regional and cultural diversity, as well as fostering respect and understanding across regions.”
METHODOLOGY:
Sympathy Analysis and Artificial Intelligence:
Bias can be understood in two forms: quantitative and qualitative. Quantitative bias is determined by the quantity, number, or frequency of a particular variable. Quantitative bias is easy to measure because it is an account of volume - one only needs to count the occurrences and compare one variable to another. It is a straightforward measure and resistant to subjective interpretation.
Qualitative bias is determined not by how often something is covered but rather how something is covered - the tone, framing and emotional slant of the coverage. This form of bias is more influential in shaping perception, yet far more difficult for humans to assess with precision and objectivity.20
In this analysis, Innohives employs a Large Language Model (LLM) as a data labeling tool. Drawing on patterns learned from a wide range of human-written language, the model identifies internally learned linguistic structures that are statistically associated with sympathetic framing in natural language.21 When applied using controlled, identical prompts, this enables the detection of sympathetic framing in CBC’s reporting not through subjective interpretation, but through the identification of recurring linguistic and tonal cues associated with emotional framing. Importantly, this process does not analyze or infer a journalist’s motivations or intentions; it is limited to examining the linguistic patterns present in the text itself.
Innohives methodology is based on LLMs’ “ability to ‘understand’ context and generate nuanced responses [which] makes them a tool with excellent potential for identifying the expression of subtle emotional expressions and providing a basis for comprehensive analysis of how sympathy is expressed in journalistic content.”22 In practical terms, these semantic signals - word choices, narrative structure, and tone - are analyzed by the LLM to identify where sympathy is present in the text. Importantly, Innohives’ analysis does not measure “sympathy” in the colloquial sense of feeling sorry for or expressing compassion toward one group or another. Rather, a sympathy analysis measures how language, tone, and framing, linguistic indicators, are used to convey emotional weight within reporting and to emphasize one side’s narrative or the others’.
By using LLMs to conduct this analysis, Innohives reduces individual subjectivity, promotes consistency in application, and enables scalability and replicability. The standardization of the methodology allows sympathetic framing to be measured systematically across a large corpus. At the same time, the results should be understood as identifying patterns rather than offering a definitive account of journalistic intention or sentiment. In essence, this approach enables the assessment of single news items as promoting the Israeli or the Palestinian narrative. These newly generated classifications are then being aggregated in different contexts for measuring the extent of the CBC’s commitment to impartiality in reporting on the Israel-Hamas war.
The Dataset:
This report examines CBC’s English-language online coverage from October 7, 2023 to June 7, 2025. The analysis is divided into two datasets:
- The initial reporting timeframe (October 7, 2023 – February 7, 2024, 4 months of war)
- The total reporting timeframe (October 7, 2023 – June 7, 2025, 20 months of war)
The same methodology is applied to both datasets. The data are divided for several reasons. Most notably, The Asserson Report: The Israel–Hamas War and the BBC analyzed BBC coverage during the first 4 months of the conflict. By separately analyzing CBC coverage over the same initial period, Innohives enables direct comparison between the two publicly funded broadcasters.
The first 4 months also capture the most concentrated period of humanitarian and military developments, during which news coverage volume was highest. As coverage volume declined over time, Innohives normalized sympathetic occurrences into proportions rather than relying on raw counts. This ensures that each article carries equal weight, regardless of overall publication volume in a given period. As a result, the analysis focuses on relative narrative treatment rather than differences in output frequency, allowing trends in sympathetic framing to be examined over time.
Innohives analyzed CBC’s online content specifically. Analyzing multiple media formats (such as television, radio, or podcasts) would require selective sampling and introduce unavoidable selection bias. CBC’s online content is centrally archived, fully accessible, and broad in scope, making it a suitable basis for examining institutional reporting patterns over time.
Data Collection:
To build the dataset, Innohives employed 2 complementary keyword-based collection strategies. First, CBC’s internal search engine was queried using predefined keywords related to geopolitical actors (e.g., Israel, Palestine, Gaza, Hamas, IDF, Antisemitism). To account for the observed variance in data collection at different time points, searches were conducted in 4 rounds between January and June 2025 to capture newly published content.
Second, Google Search was used to identify additional CBC articles containing the same keywords, allowing the inclusion of articles that may not have appeared in CBC’s internal search results due to indexing limitations. In both cases, only articles published between October 7, 2023 and June 7, 2025 were retained.
Data Refinement:
Following collection, Innohives conducted a multi-stage data refinement process. Duplicate articles resulting from overlap between collection methods were removed using similarity checks based on URLs, titles, and content. Subsequently, a second keyword filter was applied to exclude articles that did not substantively address the topic. Articles were retained only if at least 1 core keyword appeared within the article body.
The result is a final, manually validated dataset of CBC articles that robustly captures media discourse related to the Israel-Hamas war and associated issues within the defined time-frame.
LLM Analysis:
To evaluate each article’s framing, Innohives wrote a custom Python code, using the Ap-plication Programming Interface (API) made available by OpenAI, the creator of ChatGPT, to process the data. In this experiment ChatGPT-4 was the model of choice and all default settings (i.e. temperature, context, etc.) were kept. It should be noted here that ChatGPT-4's last training update was October 2023, so it had no prior knowledge of the conflict. Inno-hives asked the LLM to classify whether the presented text evoked sympathy toward spe-cific actors in the conflict. Each article was input into the model alongside a fixed prompt, which asked the following six binary questions.
This was the exact prompt used for the study:
I have the following article.
Please answer only YES or NO to the following questions:
- Does this text create sympathy for Israel?
- Does this text create sympathy for Gaza?
- Does this text create sympathy for the Israeli people?
- Does this text create sympathy for the Palestinian people?
- Does this text create sympathy for the Israeli military?
- Does this text create sympathy for Hamas?
As seen above, the model was restricted to answering “Yes” or “No” for each question. This structure allows for consistent, scalable classification of sympathetic framing across state, civilian, and militant actors. Articles could be flagged as sympathetic toward multiple actors, a single actor, or none.
