The full-scale Russian invasion has significantly changed Ukrainian society. Distinct and traumatic experiences of living through the war and a variety of views on controversial, socially important issues create fertile ground for rising tension in the public space—particularly through mutual accusations, hate speech, and humiliation of others based on their experience. In democratic societies, such rhetoric poses a particular threat, as it undermines civic unity, reduces trust in institutions, and lowers the effectiveness of public dialogue, thereby complicating the search for consensus and the possibility of peaceful conflict resolution.
Under such circumstances, media platforms, especially social networks and messengers, which are key sources of information for Ukrainians, become not only communication arenas but also battlegrounds for interpreting events, forming images of "us" and "them," and constructing identities and social divisions. In light of this, there is a need for systematic monitoring of the content of the most important platforms for Ukrainian society, including Telegram.
Regular monitoring of the popular communication channels enables the prompt detection of manifestations of discredit toward the authorities, hostility toward specific social groups, and the humiliation or devaluation of the experiences of different individuals. Since July 2025, Civil Network OPORA has been conducting such monitoring. In this text, we describe the methodological foundations of our research.
Violent Rhetoric: Theory
By definition, any political process is competitive and involves constant rivalry between various actors—the state, political parties, movements, and even individual citizens—for power, influence, and resources. In democratic societies, this competition typically occurs openly, within institutionally defined rules and procedures, notably through elections, parliamentary debates, public discussions, or peaceful protests. At the same time, in certain situations, political actors may conclude that the established mechanisms do not allow them to achieve their goals. Under such conditions, they resort to more radical political behavior strategies, among which political violence holds a special place.
Political violence is defined as hostile, aggressive or violent acts motivated by political objectives or a desire to directly or indirectly affect political change or change in governance. The key characteristic of political violence is the presence of political motivation: the desire to restrict the activities of certain social groups, change the balance of power in society, influence the adoption of political decisions, or undermine the legitimacy of state institutions, among others. Political violence can be used by various political actors—both state and non-state.
Political violence encompasses a wide spectrum of practices—from attacks or assassination attempts on representatives of the authorities and various social groups to mass riots, terrorism, ethnic cleansing, and genocide. At the same time, political violence should be viewed as a continuum within which different political actors can gradually move from less violent to more destructive actions, and vice versa. For an individual to transition from a negative attitude towards certain societal groups or politicians to supporting violence or even committing violent acts, they must undergo a specific process: radicalization. The key idea of the concept of radicalization is that no one is born a radical. Actual physical acts of political violence are preceded by a "change in beliefs, feelings, and behaviors in a direction that increasingly justifies violence" (McCauley & Moskalenko, 2008) or "a process through which people become increasingly motivated to use violent means against members of an out-group or symbolic targets to achieve behavioral change and political goals" (Doosje et al., 2016). Furthermore, as Bosi & Malthaner note, "violence emerges as a result of relational dynamics that evolve as sequences of interaction in which mutual responses and adaptations [of actors] contribute to the gradual escalation of violence."
How does it happen that individuals or groups of people become radicalized? What becomes the key factor that causes them to change their views and become more loyal to violence?
In describing the causes of radicalization, Bjørgo, Della Porta et al. note that it often arises and spreads in an environment dominated by feelings of injustice, social exclusion, and real or perceived humiliation. In their opinion, "Nothing creates so fertile a breeding ground for political radicalisation than the feeling of belonging to the camp of those left behind" compared to other individuals or groups. The authors emphasize that the feeling of injustice forms deep mutual distrust, which in turn fuels the “us versus them” paradigm and strengthens the readiness of individuals or groups to support or resort to violent actions. Similarly, as Jahnke, Abad Borger, and Beelmann stress, a key motivational source of radicalization is the need "to be someone." This can arise for various reasons—personal discrimination, traumatic experience, or humiliation (including group humiliation)—and leads to the conclusion that “people who seek to increase or restore significance may come to commit politically violent acts”.
In contrast, McCauley and Moskalenko identify several key factors, the combination of which contributes to radicalization: “We are a special or chosen group (superiority) who have been unfairly treated and betrayed (injustice), no one else cares about us or will help us (distrust), and the situation is dire—our group and our cause are in danger of extinction (vulnerability)." The researchers emphasize that radicalization can occur at different levels: the personal level—when an individual reacts to a personal slight or a threat to their identity; the group level—when collective injustice is experienced by a community; and the mass level—when radicalization is fueled by inter-state or political conflict.
How does it happen that people begin to perceive others as "us" and "them"? Why do familiar social or political processes suddenly start to seem unfair and hostile?
Explaining how the perception of threat from others is formed, Stephan et al. (2015) note that for hostile or prejudiced attitudes to emerge, a real threat to a person or the group they belong to does not necessarily have to exist—the perception of it is enough. Even the notion that an "outgroup" might threaten the power, resources, or safety of the "ingroup” (realistic threat) or its values, culture, and lifestyle (symbolic threat) can trigger negative emotions and intensify intergroup tension. At the same time, the feeling of threat can exist at the group level (for example, the belief that migrants pose a danger to the country's economy or cultural values) and at the individual level (the fear of losing a job or a sense of being undervalued by society).
One of the key factors shaping perceptions of "one’s own" group, other groups, and the sense of justice in general is how these groups are discussed in the media and public sphere. Public communication does not merely describe reality; it constructs it. It defines who is perceived as "us" or "them," which institutions evoke trust or are devalued, and which groups become legitimate objects of criticism, mockery, or hatred. It also assigns specific traits to events, individuals, and groups. In this way, the media play a decisive role in shaping perceptions of other people, groups, authorities, society, and the sense of justice. Analyzing public communication is essential for understanding the mechanisms of violence, as it often begins not with an action, but with the language that makes it possible or acceptable.
Specifically, as Butler argues, language can not only reflect but also generate violence, since it is through language that forms of dehumanization—the stripping of certain people of their human characteristics—are established. She writes that language gives rise “to physical violence that in some sense delivers the message of dehumanization that is already at work in the culture," emphasizing that physical violence is often merely a continuation of what has already occurred on a symbolic level.
As Floss notes, a society's political culture is formed through institutions and practices that determine which specific worldviews become dominant. In this sense, the media serve as a key channel through which political values, biases, and norms are disseminated among citizens. The media determine which topics receive public attention, which voices are amplified, and which are marginalized. Held also emphasizes that the media do not merely report on violence; they actively shape the frame of its perception, indicating whom society will perceive as a "terrorist" and whom as a "freedom fighter." Thus, the media do not simply reflect reality—they influence the public understanding of violence and its permissibility.
The role of media as a catalyst for radicalization and the formation of a tolerant attitude toward violence becomes particularly palpable in the digital age, where social networks create new opportunities for the dissemination of violent and polarizing narratives. As noted in the NDI report, online violence can manifest in various forms—from campaigns to discredit political opponents to the creation of dehumanization based on political orientation, gender, sexual identity, or ethnicity. Such narratives can foster a negative online environment and, in extreme cases, contribute to calls for violent actions against specific individuals or groups.
According to NDI’s observations, the first signals of escalating violence often appear on social networks. The online space provides numerous opportunities for disseminating rhetoric that gradually normalizes violence as a political tool. Monitoring the behavior of actors and groups on social media enables early detection of trends and early warning signs of potential escalation. Digital platforms are becoming a convenient space for creating and spreading calls for political violence, which requires systemic monitoring and rapid response from the authorities, law enforcement agencies, and civil society.
How This Study Originated
The theoretical approaches described above demonstrate that physical violence driven by political views does not arise suddenly; it develops gradually, through language and narratives that normalize and justify aggression. In the Ukrainian context, this problem is particularly palpable. The full-scale war, profound social changes, and vastly different experiences of living through wartime have created conditions in which the language of violence has become a constant part of public discussion. Indeed, in the media—and especially in social networks and Telegram channels—manifestations of violent rhetoric appear with increasing frequency: from the humiliation of the dignity of others and hate speech to calls for the overthrow of the government.
In the Ukrainian context, the platforms through which information is disseminated present an additional challenge. Social networks remain the most popular source of news in Ukraine: as of 2025, 86% of Ukrainians consume news content there. Telegram holds a leading position, with 72% of users reading news on the platform. The specific nature of this platform—the anonymity of channel owners, the lack of effective moderation mechanisms, and the ability to conduct media activities outside of state regulation and professional journalistic standards—makes it particularly vulnerable to manipulation. Under such conditions, Telegram transforms not only into a convenient space for obtaining information but also into an environment where radicalizing narratives are easily formed, hate is spread, and tolerance for violence gradually grows.
Given this, a need arose to systematically investigate how prevalent narratives containing the "othering" of social groups, dehumanization, or the legitimization of violence are within the Ukrainian media space. Monitoring such narratives allows for more than just identifying key "problem zones" or groups of people who are most frequently subjected to humiliation or mockery; it also helps in understanding broader social trends, such as tracking changes in intergroup trust, attitudes toward the government and state institutions, and the rising tolerance for violence.
In this study, the Civil Network OPORA conducted monitoring of public communication in Ukrainian Telegram channels to identify signs of violent rhetoric, as well as narratives that contribute to discrimination and social exclusion of various social groups and legitimize violence against other people and the state.
Where to Look for the Rhetoric of Violence?
As noted above, due to its popularity among Ukrainians and the lack of effective content moderation mechanisms, Telegram can significantly influence the processes of radicalization and the dissemination of violent rhetoric.
To cover the key narratives and differences in the content shared by various groups of actors, Civil Network OPORA conducted monitoring of public communication in the following categories of Telegram channels:
-
15 nationwide anonymous news channels with the largest number of subscribers;
-
5 channels of popular bloggers with the largest audience, as well as the channel of Myroslav Oleshko;
-
46 regional anonymous news channels—two from each region, excluding the temporarily occupied Luhansk region and the Autonomous Republic of Crimea.
The full list of Telegram channels selected for the study is provided in Appendix 1.
The analysis used the GEMMA-3 large language model (deployed locally via the Ollama CLI environment). The model was used to automatically detect messages containing hate speech, calls for violence, or toxic political rhetoric in publications from Telegram channels. Prior to classification, the message corpus was pre-cleaned using Python (specifically the pandas library). Air raid alert notifications were removed from the general dataset.
How to Identify Violent Rhetoric?
Within the framework of this study, Civil Network OPORA focused on violent rhetoric directed against individual citizens, social groups, state authorities, or institutions. The analysis is centered on publications containing messages about violence, those that tolerate or justify aggression, as well as those that construct the image of an "other" or undermine trust in the state and its institutions.
We tracked all content that contributes to the marginalization of various population groups, the normalization of violations of their rights, and the humiliation of their dignity, as well as calls for legislative restrictions on their rights and freedoms, and their general exclusion from the political process. Additionally, we tracked content that discredits and delegitimizes government representatives and state institutions, as such material could be utilized for the escalation of protests, coups d'état, political assassinations, or murders.
Furthermore, we were interested in how violent rhetoric evolves: from everyday discrimination and manifestations of disrespect based on wartime experience, religious or ethnic affiliation, language, political views, etc., to calls to beat or kill in order to achieve political goals. This is why we examined various levels and dimensions of violent rhetoric: the normalization of violence, humiliation of others in the public space, public threats, the incitement of hostility, and the dehumanization of specific individuals or entire social groups.
We categorized the identified clusters of violent rhetoric by the level of threat to the escalation of socio-political conflicts: high, medium, and low. To do this, we were guided by two criteria:
-
Directness and Intent: Whether the content contains calls for violent actions or portrays violence as a justified or normal act.
-
Impact and Reach: The scale and systematic nature of the dissemination of such content.
The risk level was determined based on a set of characteristics, such as:
-
The presence of dehumanizing or demonizing statements;
-
The use of hate speech;
-
The presence of calls for or approval of violence;
-
Constructing the image of the state and its representatives as hostile and dangerous to their citizens;
-
"Othering" social groups and portraying them as less valuable, dangerous, unworthy, etc.;
-
Supporting discrimination against various social groups;
-
Normalizing vigilantism and manifestations of physical or verbal violence as a means for resolving conflicts, changing the behavior of others, or achieving other goals in society.
A full list of the analytical categories used in this study is provided in Appendix 2.
Technical Aspects of the Methodology
Data collection from Telegram channels was automated using a Python script developed by the Civil Network OPORA, based on the Telethon library. Data were exported monthly and stored as CSV files for further processing.
Before classification, we preprocessed the messages using Python (specifically, the pandas library), removing air-raid alert notifications.
For the analysis, we used the Ollama Gemma 3 large language model, deployed locally via the Ollama CLI. The model was applied for the automated detection of messages containing hate speech, calls for violence, or other forms of violent rhetoric in posts from the Telegram channels.
For each message, the model received an individual request via a prompt, asking it to determine the degree to which hostile or violent content was present. The processing result was a value in the range 0.0-1.0, where 0.0 indicated a completely neutral message (absence of negative emotional or violent connotations), and 1.0 indicated explicit hostile or violent content, including insults, humiliation, calls for violence, or political discredit. For further analysis, we included messages in the sample with a rating of at least 0.2. This threshold allowed us to capture not only overt hate speech but also milder forms of aggressive or disparaging rhetoric and criticism of various levels of government. All results were saved in CSV format for subsequent manual verification and contextual analysis.
Appendix 3 provides an example of the prompt text applied to each message. During operation, the model received a single input message and returned a hate speech level score (0.0-1.0), a brief explanation, and, if necessary, the target (object or group) to which the hostile statements were directed.
Classification was performed using Python 3.11. The gemma3 model was operated in instruction mode with the parameter temperature = 0.0 to ensure consistent results and reproducible decisions. Processing was conducted sequentially: each message was sent to the model as an individual request, and the results were saved in a CSV file with the fields text, rate, target, rationale, channel, and date.
In cases where the number of messages flagged by the algorithm as containing violent rhetoric did not exceed 9,000, the entire array of content was analyzed manually. If the number of such messages exceeded this threshold, a stratified sample was drawn to ensure proportional representation across Telegram channels and time periods. Subsequent content analysis was also carried out manually.
Limitations of the Study
Despite the systematic nature of our monitoring of violent rhetoric in the selected Telegram channels, this study has several limitations related to both the specifics of the sample and the nature of social media content.
-
Limitations of the Channel Sample. In this study, 67 Telegram channels were analyzed, selected based on type and number of subscribers. The combined audience of these channels is nearly 38 million users, enabling us to track the key trends and patterns in the Ukrainian segment of Telegram. At the same time, this approach does not guarantee full coverage of the nuances of violent content. There is a risk that overtly radical narratives may be disseminated through smaller or closed channels, in post comments, or in private chats not included in the sample. Identifying and analyzing such environments would require separate research.
-
Limitations of Content Distribution Platforms. This study examines messages published in the most popular categories of Telegram channels, as Telegram is currently the primary news source for Ukrainians. However, other social networks—Facebook, Instagram, YouTube, TikTok, and X (formerly Twitter)—also exert significant influence on the formation of public sentiment but were outside the scope of this analysis. It is also worth noting that radicalizing content can spread on more niche platforms, such as Discord, Twitch, Substack, and Reddit, which were not investigated in this project. Studying such environments represents a promising direction for future research.
-
Limitations of Content Sampling. The analysis was conducted using automated message classification. Before beginning the study, the team manually labeled examples of content showing signs of violent rhetoric to "train" the artificial intelligence to recognize similar messages in a large sample. Subsequently, only those messages that the AI classified as potentially violent were subjected to manual analysis. If the volume of such content was moderate (up to 9,000 posts), we analyzed it entirely by hand; for high volumes of the same type of publication, we formed a stratified sample to ensure uniform representation across channels and time periods. This approach allows for the optimization of research resources while maintaining analytical accuracy. Preliminary checks indicated that the algorithm's error rate—specifically, the false assignment of content as violent—is approximately 15%; therefore, the actual level of violent rhetoric may be slightly higher than that recorded in the sample.
Appendix 1. List of Telegram channels selected for the study.
|
Сhannel |
Link |
Subcribers |
|
Труха⚡️Україна |
3,312,213 |
|
|
Times of Ukraine: Новини | Україна |
2,407,911 |
|
|
INSIDER UA | Україна | Новини |
2,151,122 |
|
|
Реальна Війна | Україна | Новини |
1,553,320 |
|
|
Україна Online: Новини | Політика |
1,505,402 |
|
|
Телеграмна служба новин - Україна |
1,118,742 |
|
|
Україна Сейчас | УС: новини, політика |
1,200,721 |
|
|
Всевидящее ОКО: Україна |
1,118,925 |
|
|
Український Телеграм 🇺🇦 |
1,067,324 |
|
|
Новини Україна | Головний |
1,020,354 |
|
|
Інсайдер ЗСУ |
1,094,163 |
|
|
Батальйон «Монако» | Україна | Новини |
956,494 |
|
|
Україна 24/7 - новини |
875,834 |
|
|
Україна Live | Новини |
861,167 |
|
|
ППО - News🇺🇦Новини |
774,803 |
|
|
Лачен пише |
1,526,889 |
|
|
Анатолій Шарій |
1,463,853 |
|
|
STERNENKO |
852,011 |
|
|
Андрій Смолій |
386,513 |
|
|
БУТУСОВ ПЛЮС |
367,152 |
|
|
Реальна Вінниця |
230,654 |
|
|
Труха⚡️Вінниця | Винница | Новини |
149,052 |
|
|
Volynblog (Луцьк• Ковель•Володимир) |
81,168 |
|
|
Луцьк⚡️Труха |
77,326 |
|
|
ХДніпро 🇺🇦 |
520,337 |
|
|
Дніпро Оперативний 🇺🇦| Війна |
319,196 |
|
|
Типичный Краматорск |
123,169 |
|
|
Славянск / Краматорск |
107,292 |
|
|
Новини Житомира | pzhytomyr |
112,943 |
|
|
Житомир.info |
94,489 |
|
|
Ужгород⚡️Труха |
46,548 |
|
|
Ужгород LIVE | Закарпаття |
44,103 |
|
|
Труха⚡️Запорожье |
227,008 |
|
|
Это Запорожье 🇺🇦 Новости |
181,662 |
|
|
Труха⚡️Івано-Франківськ |
100,200 |
|
|
Івано-Франківськ 24/7📱 |
241,622 |
|
|
Труха⚡️Кропивницький |
150,849 |
|
|
kirovograd_official🇺🇦Кропивницький |
65,622 |
|
|
Труха⚡️Львів |
262,392 |
|
|
Львівич | Новини |
333,886 |
|
|
Новости Николаева🇺🇦 | Миколаїв |
240,140 |
|
|
Труха⚡️Николаев |
91,587 |
|
|
Хуевая Одесса |
892,991 |
|
|
Одесса INFO ⚓️ 🇺🇦 Новости | Одеса |
919,854 |
|
|
Труха⚡️Полтава |
251,360 |
|
|
ПОЛТАВА НОВИНИ | СИРЕНА |
214,546 |
|
|
Рівне Головне |
169,147 |
|
|
РІВНЕ 1283 | RIVNE ⚡️ |
124,998 |
|
|
СУМИ / SUMY GO ∆ |
160,168 |
|
|
Труха⚡️Суми |
107,052 |
|
|
Тернопіль⚡️Труха |
113,984 |
|
|
Хуйовий Тернопіль |
60,073 |
|
|
Харьков life | Харків 🇺🇦 |
670,647 |
|
|
Труха⚡️Харьков |
528,865 |
|
|
Хуевый Херсон 🇺🇦 |
326,409 |
|
|
ХЕРСОН: Non Fake |
121,476 |
|
|
Хмельницький LIVE |
264,490 |
|
|
Реальний Хмельницький | Новини |
117,708 |
|
|
Труха⚡️Черкаси |
140,927 |
|
|
Типові Черкаси |
52,290 |
|
|
ХОРОШІ ЧЕРНІВЦІ CHERNIVTSI БУКОВИНА ХЧ |
105,624 |
|
|
Чернівці⚡️Труха |
85,088 |
|
|
Труха⚡️Чернігів |
121,081 |
|
|
Твій Чернігів |
50,526 |
|
|
Київ ІНФО | Новини Україна 🇺🇦 |
1,088,056 |
|
|
Реальний Київ | Украина |
1,463,003 |
Appendix 2. List of the Analytical Categories Used in This Research
|
Analytical category |
Explanation |
Examples |
|
Ableism |
Ableism occurs when physical, mental, or cognitive conditions are used as insults or as a means of devaluation. This can take the form of a direct attack on a person with a disability or the metaphorical use of words such as "invalid," "degenerate," "psycho," "schizo," "addict," etc., as slurs. In any case, such usage reproduces stigma and discrimination against people with disabilities or health impairments. |
"What kind of a degenerate do you have to be to write something like that?" "This government is 'invalid' (crippled); they aren't capable of doing anything." |
|
Ageism |
Ageism applies to messages that humiliate or devalue people based on their age—whether they are young or elderly. This includes using age as an insult, generalizing negative traits to a specific age group, or denying a person's competence or right to participate in public life on the basis of age. |
"This deputy is an old man with dementia; he should have retired long ago." "Kids (youngsters) don't understand anything about politics." |
|
Collaborationism |
Collaborationism applies to content that describes or accuses individuals, groups, or institutions of cooperating with the enemy, occupation structures, or the intelligence services of the Russian Federation. This may involve working for Russian special services or the military ("worked for the FSB," "leaked data to the Russians"), participating in the creation or functioning of occupation administrations, aiding in the preparation of terrorist attacks or sabotage at the order of the Russian Federation, or supporting the enemy in the public sphere ("called Russian soldiers 'defenders'," "spread pro-Russian messages"). |
"A businessman worked for Russian intelligence services in Vinnytsia." "A resident of Chernivtsi was waiting for the 'Rusnya' (Russians) and called the occupiers 'defenders of the land'." |
|
Coverage of Domestic Violence |
Coverage of domestic violence refers to content containing direct mentions or descriptions of instances of violence in a domestic or private setting (within a family, between a couple, relatives, friends, acquaintances, or neighbors), regardless of the motives or outcomes. Such cases include physical, psychological, sexual, economic, and other types of violence. This category also applies when the material contains reports on court cases, detentions, punishments, or the state's and society's responses to domestic violence. It does not apply to political, military, or other forms of public violence (including protests), sports competitions, criminal incidents in public spaces without signs of personal/domestic relationships between the parties, or to metaphors. |
"A man beat his wife after an argument." "In Cherkasy, a man struck his wife with a frying pan during a conflict." |
|
Coverage of Violence |
Coverage of violence refers to content that describes or mentions instances of public violence in communal spaces (on the street, in establishments, on transport, at mass events, etc.) between acquaintances or strangers, groups of people, or within a criminal context. This includes brawls, acts of hooliganism, street assaults, violence in recreation areas or public institutions, and clashes between groups of individuals. Additionally, this category encompasses instances of violence committed by military personnel or veterans outside of combat zones (conflicts in bars or other public settings, traffic accidents, beatings, etc.). This category does not apply to violence directed at Russians or the Russian military. |
"A massive brawl between groups of young men broke out in the city center." "A soldier beat a man in a bar after returning from the front." |
|
Criticism of the Authorities |
Criticism of the authorities encompasses content that contains negative assessments, remarks, or disagreement regarding the activities of state authorities, the government, parliament, the president, or other state institutions, within the scope of reasoned critique. Unlike the "discrediting of authorities," criticism is grounded in specific decisions, actions, or policies and does not lead to total delegitimization or to branding the government as "hostile" or "traitorous." This covers remarks on the efficiency of institutional work, criticism of management decisions, and expressions of distrust toward specific policies or steps taken by the authorities. |
"The government failed to prepare for the heating season." "The authorities are not doing enough to support small businesses." |
|
Criticism of the Local Authorities |
Criticism of local authorities applies to messages containing negative assessments of the actions or inaction of local self-government bodies or their representatives (the mayor, city/regional council, or utility companies). This criticism concerns specific decisions, management errors, or failures in the performance of duties. |
"The city authorities did not repair the roads, even though the funds were allocated." |
|
Defamation |
Defamation is used to designate content containing false, unverified, or deliberately fabricated accusations directed against a specific individual or group with the intent to damage their reputation. Unlike discredit, which may be based on true facts, defamation always involves the dissemination of lies, fabrications, or rumors. This can take the form of overt fakes ("this politician sold out the country") or insinuations without evidence ("everyone knows he is a Kremlin agent"). |
"This activist receives money from the enemy." "Journalist X is a drug addict; she cannot be trusted." |
|
Dehumanization |
Dehumanization refers to messages in which individuals or groups are stripped of human traits, compared to animals, parasites, diseases, inanimate objects, or natural forces, or presented as "subhuman," "creatures," or a "mass" without individuality. Such statements lower empathy toward the group and legitimize hostility or violence against them. |
"They aren't human; they're underdeveloped cattle." |
|
Demonization |
Demonization occurs when a message portrays an opponent (a politician, party, social group, activist, or institution) as absolute evil, an enemy, or an existential threat. This typically manifests in two primary forms: dehumanizing or apocalyptic imagery (using metaphors of disease, infection, demons, "the horde," plague, etc.) and equating the target to the actions or essence of the enemy ("they are just like the Russians," "this government is no different from the Kremlin"). Demonization presents the object not as an opponent for discussion—or even as a corrupt politician—but as an existential threat. |
"Zelenskyy’s inner circle is an empire of evil." |
|
Discrediting, Humiliating, or Mocking Ukrainians Abroad |
Discrediting, humiliating, or mocking Ukrainians abroad applies to content in which Ukrainians residing outside the country are portrayed in a derogatory, contemptuous, or distrustful manner. This category includes messages that undermine the reputation or moral character of Ukrainians abroad, spread stereotypes about "lazy refugees" or "traitors," question their loyalty or patriotism, or contrast them with "real Ukrainians" who remained in Ukraine. |
"While the boys are dying at the front, they are warming themselves up in Europe." |
|
Discrediting the Authorities |
Discrediting the authorities applies when a message undermines trust in central government bodies, local self-government, and their representatives (the President, Government, Parliament, and key officials). This may occur by attributing to them corruption, betrayal, inaction, incompetence in performing their duties ("criminal government," "they sold out the country"), a lack of will to fulfill their responsibilities, activities in Russia's interests, or devaluing their intellectual abilities. It also includes mockery based on appearance, gender, professional background, ethnic or religious affiliation, family members, etc. There is a distinction between legitimate criticism and discredit: legitimate criticism is based on facts or analysis and describes specific actions or inactions of the authorities without using offensive or disparaging language. Discredit occurs when criticism is accompanied by insults, generalizations, sarcasm, or exaggeration that form a perception of the authorities as criminal, helpless, or hostile. |
"Once again they did nothing—they just sit there and steal." "These idiots in parliament are just pulling the wool over our eyes." |
|
Discrediting the Institutions |
Discrediting institutions applies when a message undermines trust in state structures and social institutions as a whole. Recipients of such messages may include the government as a system, the parliament, the courts, the army, law enforcement agencies, the Central Election Commission (CEC), educational or medical institutions, the church, and others. These messages are directed not at specific individuals, groups, or political parties, but at the institution as a whole. They present institutions as completely corrupt, incompetent, weak, passive, controlled by external forces, or redundant in society. There is a distinction between legitimate criticism and discredit: legitimate criticism is based on facts or analysis and describes specific actions or inactions of the authorities without using offensive or disparaging language. Discredit occurs when criticism is accompanied by insults, generalizations, sarcasm, or exaggeration that foster perceptions of authorities as criminal, helpless, or hostile. |
"There has never been justice in the courts—only profiteers work there." "The state is a fiction; it is incapable of resolving anything." |
|
Discrediting the Local Authorities |
Discrediting the local authorities applies to content intended to delegitimize, humiliate, or portray local self-government bodies, mayors, village heads, local councils, or administrations as hostile. This category includes accusations of treason, collusion with the enemy, bribery, total corruption, or deliberate harm to the community. There is a distinction between legitimate criticism and discredit: legitimate criticism is based on facts or analysis and describes specific actions or inactions of the authorities without using offensive or disparaging language. Discredit occurs when criticism is accompanied by insults, generalizations, sarcasm, or exaggeration that foster perceptions of authorities as criminal, helpless, or hostile. |
"The city council is stealing the community's money." "The mayor is working for the enemy and leaking information to the occupiers." |
|
Discrediting the Military |
Discrediting the military refers to content that undermines the trust, authority, or moral status of the Ukrainian military, the Armed Forces of Ukraine (AFU), or the military command. This includes attributing systemic incompetence, cowardice, betrayal, or dishonesty in official reports to the AFU, as well as mocking or devaluing the service of soldiers (e.g., calling them "meat," "good-for-nothings," etc.). This should not be confused with factual criticism of specific decisions or the provision of supplies to the troops. |
"The AFU reports are pure PR; everything is collapsing at the front." "The commanders are sitting comfortably in the heat while they drive the boys into senseless assaults." |
|
Discrediting the Mobilization |
Discrediting the mobilization encompasses messages that undermine the legitimacy and acceptability of mobilization or the work of the TRC (Territorial Recruitment Centers) as state policy. Such content portrays mobilization as illegal, criminal, senseless, selectively punitive, or intentionally harmful to citizens. Typical methods include: depicting mobilized soldiers as "cannon fodder," using the frame of "hunting" or "coercion," normalizing or endorsing violent actions against TCC staff, sensationalist generalizations (e.g., "they took everyone, even those with exemptions"), and exploiting tragedies to conclude that "mobilization = evil/harm." |
"The people-catchers are snatching men off the streets." "My boyfriend was 'bus-ified' (forcibly put into a van)." |
|
Discrediting the Politicians |
Discrediting politicians refers to content that aims to delegitimize, humiliate, or portray specific politicians or public figures associated with politics in a negative light. This category includes accusations of treason, bribery, collusion with the enemy, corruption, "selling out the country," total incompetence, or other intentionally negative characterizations. It also covers the use of derogatory language, stigmatizing labels, or sweeping accusations. |
"Poroshenko sold out Ukraine in 2014." "He is a traitor, not a politician." |
|
Discrediting the Protests |
Discrediting protests applies to content that delegitimizes or belittles protest actions, demonstrations, or rallies by portraying them as dangerous, enemy-controlled, or entirely meaningless. The primary characteristic is the creation of a frame suggesting that protesters are not acting of their own free will but are instead "tools of external forces" or a "threat to public order." |
"Russia wants to use protests in Ukraine for destabilization." "The Kremlin is instigating a new Maidan through the hands of propaganda." |
|
Disinformation |
Disinformation refers to content that deliberately contains false, manipulative, or distorted information intended to mislead the audience. This category includes disinformation "plants" (fabricated news, forged quotes, fake reports), manipulative materials (one-sided coverage, intentional distortion of facts, the use of false data or facts taken out of context), conspiracy theories, or myths and fabrications presented as facts. |
"Zelenskyy has already fled the country, and the videos of him are pre-recorded dubs." "Ukraine is being sold to the Americans for its debts." |
|
Emphasizing Regional Differences |
Emphasizing regional differences applies to messages in which a person's origin (region, city, East/West, or IDP status) is used not as a neutral context but as a basis for evaluation. This can be negative ("residents of X are to blame," "displaced persons are ruining the city") or positive ("our region is the real Ukraine," "why is there no Galician president?"). It also covers cases where a region or city is mentioned in connection with unacceptable behavior (e.g., "a fight in Lviv" → implying "typical for Lviv") even without a direct statement of causality. |
"Was the curfew cancelled in Lviv or something?" "Why is there no Galician president? We are a reliable rear..." |
|
Homophobia |
Homophobia applies to messages that contain hostile, derogatory, or discriminatory statements regarding LGBT+ people (gay men, lesbians, bisexuals, transgender individuals, etc.). This can include direct insults, stereotyping, mockery, the attribution of mental illnesses or deviant behavior, and the denial of the right of LGBT+ people to exist, have families, achieve equality, or participate in public life. |
“The March of Equality is a perversion; how can such a thing be allowed?" |
|
Humiliation |
Humiliation refers to messages in which an individual or group is mocked, shamed, or devalued through profanity, offensive labels, or sarcastic phrasing. The goal is to degrade a specific person or group, diminish respect for them, or portray them as worthless or ridiculous. |
"The speech was a total disgrace; he embarrassed himself in front of everyone again." "He is a nobody and a jester; no one takes him seriously." |
|
Incitement to Violence |
Incitement to violence applies to content containing direct or veiled calls for physical violence against individuals, groups, or communities (excluding Russians and the Russian military). This includes imperative phrasing ("beat," "kill," "shoot," "burn"), wishes for violence ("may they all be killed"), as well as veiled or indirect calls ("they need to be removed," "throw them all out"). |
"All traitors should be shot." "Hang every corrupt official." |
|
Misogyny |
Misogyny applies to messages in which women or the female identity are presented in a humiliating, hostile, or devaluing manner. This includes overt insults, stereotyping (e.g., "a woman's place is in the kitchen"), sexualization, mocking appearance or intelligence, and denying the right to participate in politics, public life, or decision-making. |
"What kind of politician is she? She's just a broad; her place is at home." "All these feminists are just hysterics." |
|
Mockery |
Mockery refers to content that ridicules or sarcastically humiliates individuals, groups, or events. The primary characteristic is not merely a joke or an instance of irony, but the deliberate portrayal of someone or something as negative, ridiculous, or absurd, with an emphasis on humiliation or discredit. This may include offensive jokes, sarcasm, mocking others' failures or suffering, and exaggeration for the purpose of devaluation. This category does not apply if the message contains light humor or irony without humiliation, concerns a neutral fact or information, or if the humor is positive or self-deprecating. |
"Those clowns gathered for a rally again 🤡." "An Arab sheikh on a Poltava beach 🇦🇪." |
|
Othering of Certain Categories of the Population |
Othering of certain categories of the population appears in content that distances a specific social group from the majority, depicts them as "alien," "hostile," or "not like us," or assigns a stigmatizing label that emphasizes their difference. This can manifest through direct opposition ("they don't belong here"), denial of belonging ("they aren't real Ukrainians"), alienating labels ("draft dodgers," "grant-eaters"), or generalizing statements that render a group "other" ("they are incapable of understanding"). |
"Draft dodgers have no right to call themselves men." "Grant-eaters (foreign-funded activists) don't work for Ukraine." |
|
Political Slur |
Political slur applies to messages that use offensive labels for political opponents, parties, ideologies, or supporters of certain political forces. Such labels do not address criticism of specific actions or policies (which may be legitimate); instead, they aim to devalue the target through negative stereotypes or abusive epithets. The difference between a political slur and the discrediting of authorities: a political slur is humiliation based on affiliation with a political group or ideology, whereas discrediting the authorities is the undermining of trust in government bodies or officials. |
"The 'Zelebobiky' (Zelenskyy supporters) are licking their leader's boots again." "The 'Porokhoboty' (Poroshenko supporters) are whining in the comments because they don't like reality." |
|
Professional slur |
Professional slur refers to messages that humiliate an individual or group based on their professional activity, using their profession as an insult or negative label. This refers not to criticism of specific actions but to the devaluation of the profession or its representatives. |
"The 'zhurnoshliukhy' (journal-whores) have written another hit piece." "Teachers are just 'stupid sovoks' (Soviet-minded) who don't teach children anything." |
|
Racial or Ethnic Slur |
A racial or ethnic slur refers to messages containing offensive, humiliating, or devaluing language regarding the ethnic or racial identity of individuals. This includes direct insults as well as "jokes," stereotypes, or labels that categorize a person or group as "others" and degrade their dignity. |
"Gypsies (Romanis) are always stealing." "The 'Zhydy' (kikes/Yids) are to blame for all problems." |
|
Religious Slur |
Religious slur applies to messages containing offensive, humiliating, or stereotypical words regarding individuals based on their religious affiliation or beliefs. This can refer to specific individuals as well as entire denominations or religious groups. |
"Muslims are always terrorists." |
|
Sexism and Objectification |
Sexism and objectification apply to content containing stereotypical, humiliating, or devaluing statements about women or men based on their sex, as well as cases of objectification where a person is discussed solely in the context of their appearance or sexual attractiveness. This includes portraying women/men through clichés, generalizations that devalue based on gender ("all men think only about sex"), jokes and messages that emphasize the body or appearance instead of personality, and treating a woman/man as a sexual or visual object. |
"+20 kg and a divorce? People in Frankivsk don't tolerate fat women." "A reminder about the leaked 18+ star photos." |
|
Stereotyping |
Stereotyping refers to content that presents generalized, simplified, or clichéd statements about groups of people as characteristics of the entire group. This can include negative or contemptuous labels, ironic generalizations, or stereotypes that fall outside the specific categories of discrimination defined elsewhere (such as sexism, ageism, religious slurs, etc.). |
"Drivers are 'deers' (idiots) because they don't look in their mirrors." |
|
Threat of Violence |
Threat of violence applies to content in which threats to cause physical harm to individuals, groups of people, government representatives, military personnel, journalists, or any other entities are expressed directly or indirectly. Such messages fuel aggression in the public sphere, justify violence, or create a perception of its permissibility and inevitability. |
"Give me the address of Deputy X, I'll take care of him myself." |
|
Tolerance for Violence |
Tolerance for violence applies to texts that do not contain direct calls for or encouragement of violence, but instead present it as an acceptable, normal, or desirable phenomenon. This can occur through approval ("that's the way to do it," "they deserved it"), normalization ("what's the big deal, everyone does it"), justification ("there's no other way to deal with them"), humor or sarcasm (memes, laughter as a reaction to violence), or presenting violence as expected, logical, and inevitable ("he got what he deserved," "that's how it was meant to be"). This category does not apply to neutral descriptions of violent facts or news reports without evaluation, nor to cases with direct calls to violence. It also does not apply to violence against Russians and the Russian military. |
"They did the right thing giving him a blow to the head."
"Serves them right; they deserved it." |
Appendix 3. Prompt for Analysis (Translation from Ukrainian)
You are a strict and precise classifier of harmful speech in Telegram messages. Your task is to determine whether the text contains hate speech, calls for violence, personal insults, discredit, or sarcastic political criticism, as well as other forms of toxic expression (vulgarity, humiliation, biased or offensive wording). Do not consider neutral discussions, informational messages, or factual news as hate speech. If the text contains a description or mention of violence (even without approval), assign it a score of no less than 0.3, as such messages are also significant for the study of the rhetoric of violence.
Provide a numerical score for the level of hate speech, where 1.0 indicates overt hostility or violent content, and 0.0 indicates a neutral message. The score cannot exceed 1.0. For further classification, messages with a rating of 0.5 and above are considered to contain hate speech. Include a short explanation (rationale) — as concise as possible, only one sentence. If the score exceeds 0.2, also specify who or what group is the object of the hostile statements. If the target is unclear, output "target": "undefined".
Examples:
-
"Ahahahahaha, so they came out in support of the presidential bill?! 😂😂😂" — hate speech, as it is a sarcastic humiliation of a political group.
-
"Missile heading for Brovary" — neutral, as it is a factual report.
-
"Only in Lviv could they get away with something like that" — hate speech, as it is a sarcastic remark about regional differences.
-
"The 'katsapy' (derogatory for Russians) struck the city" — hate speech, as the word "katsapy" is an offensive ethnonym directed at Russians.
Output the result in the format: { "rate": 0.6, "target": "political group", "rationale": "sarcastic criticism of the government" }
This material has been funded by UK International Development from the UK government; however, the views expressed do not necessarily reflect the UK government’s official policies.