Monthly Report September 2025
Data for September 2025 was based on 13.9M messages in 25 languages across 12 social media platforms such as X, YouTube, TikTok, Threads and Instagram.
Content warning: Presented data may contain disturbing language related to online hate speech.
Average toxicity
In September, the overall average toxicity level was 0.21, marking an increase compared to the previous month. The month began with relatively low and stable values of 0.18, persisting through the first week. Around September 9, toxicity began to rise steadily, reaching a peak of 0.25 on the 11th and 12th - the highest levels recorded during the month. Following this spike, toxicity gradually declined, stabilising between 0.21 and 0.22 from mid- to late September. By the final week, levels had returned to around 0.20-0.21, suggesting that while overall discussions remained moderately toxic, the peak observed mid-month may have been triggered by short-term events or an intensification of online discourse. Overall, the data reflect a temporary surge in toxicity during the second week of September, followed by a period of stabilisation at moderately elevated levels.
Baseline channel analysis
Overall, antisemitic and anti-LGBTQ+ content remained the most toxic domains, followed by anti-Muslim narratives. The anti-Muslim baseline maintained moderate levels, ranging between 0.22 and 0.25, while anti-refugee discourse remained slightly lower, at 0.22-0.23. Sexism and anti-Roma discourse were comparatively less toxic. Sexism-related content showed the lowest toxicity average overall, peaking at 0.20 on September 11 and stabilising at 0.17 on September 16. The anti-Roma baseline displayed mild volatility early in the month (0.16-0.21) but declined after September 6, maintaining one of the lowest averages.
Baseline Sexism
Volume: 3M of 31.9M posts (9.3%) in this channel were identified as toxic.
Platforms: Most data was observed on X (51%) and YouTube (33%).
Languages: Posts are mostly written in English (52%).
Toxicity Analysis: The average toxicity score was 0.16. Posts with very high toxicity (≥0.8) totalled 124k, and are most prominent on Chirpwire (11%).
Toxicity Categories:
50% of toxic posts were categorised as Sexism.
34% of those toxic posts contained violent rhetoric.
Commonly used words included: women, woman, stupid, girl, femmes, femme, homme, الله, quran, charlie
Key Findings: Several social media posts express strong negative sentiments and accusations. One post criticizes individuals described as "mentally disturbed left-fascists and sociopaths" for allegedly celebrating the death of a young father, and also targets a person named Kimmel, accusing them of presenting a "hateful, inhuman, and left-fascist face." The post further criticizes public broadcasters (ÖRR) for associating with such individuals. Another post uses derogatory language to describe people who are allegedly being forced into criminal acts, including sexual offenses against minors and pedophilia, linking them to "Nazi pigs." A French post contains highly offensive language directed at someone, listing accusations such as being a "pedo," a "dealer," someone who "beats his wife," involved in "embezzlement," and a "drug addict who buys from a minor," referring to them as "notorious thugs." Finally, a Spanish post uses extremely vulgar and hateful language, calling someone "scum," "piece of shit," and "son of a bitch," and questioning if they are pleased by the "murder of children."
Selected News items
• El Gobierno aprobará este mes la primera ley sobre violencia vicaria
• Abusos sexuais na Igreja: "Houve perguntas muito abusivas", denuncia associação
• Associação de vítimas de abuso na Igreja critica métodos de intimidação do grupo Vita
• El PP denuncia ante la Comisión Europea el fallo del sistema de pulseras antimaltrato en España
• Pai de Elon Musk acusado de abuso sexual infantil por cinco filhos e enteados
Baseline Anti-LGBTQ+
Volume: 681.4k of 4.5M posts (15.2%) in this channel were identified as toxic.
Platforms: Most data was observed on X (71%) and YouTube (16%).
Languages: Posts are mostly written in English (51%) and Spanish (11%).
Toxicity Analysis: The average toxicity score was 0.27. Posts with very high toxicity (≥0.8) totalled 26k, and are most prominent on Gab (3%).
Toxicity Categories:
73% of toxic posts were categorised as Sexism.
17% of those toxic posts contained violent rhetoric.
Commonly used words included: homo, women, schwule, lésbica, lesbian, lesbica, lesbiana, lesben, queer, travesti
Key Findings: The provided posts contain a significant amount of hateful and derogatory language directed towards LGBTQ+ individuals. Several posts use slurs and express extreme animosity towards gay people, transgender individuals, and lesbians. There are instances of threats of violence and dehumanizing rhetoric. Some posts also conflate LGBTQ+ identities with criminal behavior, mental illness, or political ideologies such as communism and antifa. Specific individuals and political figures are targeted with insults that often include homophobic and transphobic slurs, such as "travesti," "maricon," and "joto." The content also touches upon political discourse, with some posts linking LGBTQ+ rights or identities to specific political parties or movements. The language used is consistently aggressive and offensive, reflecting a strong negative sentiment towards the LGBTQ+ community.
Selected News items
• Diskriminierung: Sexuelle Identität ins Grundgesetz - CDU-Ministerpräsidenten starten Vorstoß
• Aufregung um queere "Hüttengaudi" beim "Aufsteirern" | Heute.at
• Anti-Song Contest in Russland beleidigt Conchita | Heute.at
Baseline Anti-Muslim
Volume: 1.3M of 7.7M posts (17.3%) in this channel were identified as toxic.
Platforms: Most data was observed on X (72%) and YouTube (19%).
Languages: Posts are mostly written in English (45%), Arabic (17%), and French (11%).
Toxicity Analysis: The average toxicity score was 0.25. Posts with very high toxicity (≥0.8) totalled 105k, and are most prominent on Chirpwire (9%).
Toxicity Categories:
79% of toxic posts were categorised as Religion.
35% of those toxic posts contained violent rhetoric.
Commonly used words included: muslim, muslims, moros, moro, الله, islam, islamic, الإسلام, quran, المسلمين
Key Findings: The social media posts reveal a strong undercurrent of anti-Muslim sentiment, often expressed with extreme prejudice and hateful language. Several posts specifically target "moros" (Moors), a term frequently used as a derogatory slur for Muslims, particularly those of North African descent. These posts accuse them of various crimes and negative behaviors, such as being "pederastas, violadores, asesinos y terroristas" (pederasts, rapists, murderers, and terrorists). Some posts call for their expulsion from Europe, with phrases like "Europa quiere no musulmanes" (Europe doesn't want Muslims) and "Oprotten kutwijf" (Get out, you bitch). There are also instances of religious intolerance, with one post stating "Muerte al Islam, muerte al musulman" (Death to Islam, death to Muslims). The sentiment extends to those who defend Muslims, as seen in the post "a los que los defienden" (and to those who defend them). Some posts express a desire for a "Catalunya independiente pero limpia y con fronteras" (independent Catalonia but clean and with borders), implying a wish for ethnic or religious homogeneity. The language used is consistently aggressive and dehumanizing, employing terms like "gentuza" (scum), "chusma" (rabble), "escoria" (scum), and "rata" (rat). There are also mentions of specific locations like Torrejón, a municipality in Spain, and broader calls for political action, such as "Gobierno a prisión" (Government to prison).
Selected News items
• "Defend Europe", "Islam = violeurs", vague de tags malsains à Namur
• Svinjske glave postavili ispred džamija u Parizu i okolici: 'Bile su na cestama, ovo je gnjusno'
• Un pied de sanglier découvert sur le parking d'une mosquée
• Rasista, seksista, mizogin i islambofob. Burmistrz Londynu nie gryzł się w język mówiąc o Trumpie
Baseline Anti-Refugee/Migrants
Volume: 289.7k of 2.5M posts (11.5%) in this channel were identified as toxic.
Platforms: Most data was observed on X (74%) and YouTube (14%).
Languages: Posts are mostly written in English (34%), Spanish (16%), and German (13%).
Toxicity Analysis: The average toxicity score was 0.22. Posts with very high toxicity (≥0.8) totalled 12k, and are most prominent on Gab (1%).
Toxicity Categories:
67% of toxic posts were categorised as Politics.
34% of those toxic posts contained violent rhetoric.
Commonly used words included: refugees, migrant, لاجئ, migrante, asylanten, immigrati, deutschland, vluchtelingen, israel, réfugiés
Key Findings: The social media posts reveal a strong undercurrent of anti-immigrant and anti-refugee sentiment across various European countries. Several posts express frustration with the perceived influx of individuals, often using derogatory terms like "Gesindel" (rabble) and "Abschaum" (scum). There's a recurring theme of associating immigrants and refugees with criminality, with frequent mentions of "Messerstecher" (knife-wielders), "Vergewaltiger" (rapists), and "Mörder" (murderers). Some posts specifically target "muslimischen, palästinensischen, arabische Flüchtlinge," expressing a desire to keep them out of Europe. The posts also criticize political leaders and policies perceived as too lenient towards immigrants, with calls to "Einwanderung stoppen" (stop immigration) and to send individuals back to their countries of origin. In Germany, there are mentions of the "Bundesregierung" (federal government) being filled with "Vollidioten" (complete idiots) and calls to stop the "Asylindustrie" (asylum industry). The sentiment extends to a general distrust of the state's definition of "Migrant," contrasting it with a more restrictive, "legal" definition.
Selected News items
• Moncloa descarta un decreto para regularizar medio millón de migrantes como pide Sumar
• Vox pide someter a una consulta popular la apertura de un nuevo centro de menas en Inca
Baseline Antisemitism
Volume: 1.4M of 6.9M posts (20.3%) in this channel were identified as toxic.
Platforms: Most data was observed on X (77%) and YouTube (13%).
Languages: Posts are mostly written in English (55%).
Toxicity Analysis: The average toxicity score was 0.30. Posts with very high toxicity (≥0.8) totalled 90k, and are most prominent on Gab (5%).
Toxicity Categories:
83% of toxic posts were categorised as Racism.
37% of those toxic posts contained violent rhetoric.
Commonly used words included: jews, اليهود, jewish, israel, judíos, juifs, يهودي, israël, żydów, juden
Key Findings: The posts display intense hostility towards Jewish people, including derogatory terms, conspiracy theories, and calls for violence. Common narratives accuse Jews of global manipulation or moral corruption and often conflate Jewish identity with the policies of the Israeli government. The language is dehumanising and aggressive, featuring references to historical atrocities and antisemitic stereotypes.
Selected News items
• Na trgovino obesil napis: Judom vstop prepovedan
Baseline Roma
Volume: 181k of 944.8k posts (19.2%) in this channel were identified as toxic.
Platforms: Most data was observed on X (59%) and YouTube (29%).
Languages: Posts are mostly written in Russian (21%), English (21%), Portuguese (13%), and Spanish (13%).
Toxicity Analysis: The average toxicity score was 0.23. Posts with very high toxicity (≥0.8) totalled 12k, and are most prominent on Gab (6%).
Toxicity Categories:
87% of toxic posts were categorised as Racism.
13% of those toxic posts contained violent rhetoric.
Commonly used words included: gitanos, gitano, gypsies, zigeuner, gypsy, roms, moros, cygan, jews, roma
Key Findings: A recurring theme is the use of derogatory language and stereotypes against Roma people, often associating them with criminality, dishonesty, and undesirable social behaviors. Several posts explicitly call for their expulsion from countries or express extreme animosity. Beyond the focus on Roma, the content also includes xenophobic remarks directed at other groups, such as Palestinians, Africans, and people from Eastern Europe, often conflating them with negative stereotypes. There are also instances of antisemitic and anti-immigrant sentiments, with some posts referencing historical atrocities and advocating for similar actions. The language used is consistently aggressive, offensive, and dehumanizing, reflecting a deep-seated prejudice and hostility towards the targeted communities.
Selected News items
• Csak Bayer Zsolttó idézett a parlamentben, de figyelmeztetést kapott Csárdi Antal
• Figyelmeztetést kapott Csárdi Antal, miután Bayer Zsolttól idézett a parlamentben
• Felvonulással, zenével és beszédekkel ünnepelték a Roma Büszkeség Napját
• "A büdös k*rva életbe már" - többen is kiakadtak Hadházy Ákos cigányozós posztjára
• Rasszista jellegű megjegyzést tett közzé a Tisza mögé besoroló kormányellenes történész
Sexism
Highest share of neutral messages (60%), suggesting comparatively balanced and less polarised discussions.
Medium-toxicity content accounts for 18.1%, while high-toxicity remains minimal (0.44%).
Overall, sexism remains the least toxic baseline among all monitored categories.
Anti-Roma
Second-highest share of neutral messages (61.7%), but with greater volatility than other baselines.
Medium-toxicity content represents 19.8%, while high-toxicity reaches 0.78%, the second-highest across all baselines.
Anti-Muslim
Neutral share of 46 %, among the highest after sexism and anti-refugee content.
Medium-toxicity content accounts for 24.5%, with high-toxicity at 1.16%, reflecting persistent polarisation and hostility.
Anti-Refugees
Neutral share of 45.4%, similar to anti-Muslim discourse.
Medium-toxicity content remains moderate (21.3%), while high-toxicity is limited to 0.46%, suggesting steady but contained toxicity levels.
Anti-LGBTQ+
Neutral share of 29.7%, the lowest among all baselines.
Low-toxicity content dominates (43.8%), while medium-toxicity (26.0%) and high-toxicity (0.48%) remain moderate.
Antisemitism
Medium-toxicity content accounts for 33.8%, while high-toxicity reaches 1.19%, the highest across all baselines, highlighting persistent and intense hostility.
Hate speech by category
These data reveal distinct thematic patterns across the six monitored baselines. Antisemitic content stands out for its strong association with racism (80%) and religion (43%), making it one of the most ideologically loaded categories. It also shows a high overlap with political discourse (39%) and threatening language (20.5%), reflecting its highly aggressive tone. Anti-Muslim narratives are similarly shaped by religious framing (75.60%) and racism (47.22%), while also integrating political elements (27.88%) and threats (19.8%), confirming their intersectional and polarising nature.
Anti-LGBTQ+ hate is primarily driven by sexism (54.7%) and ridicule (14.3%), underscoring its gendered and mocking undertones. It also records high levels of threats (17%), suggesting that hostility often escalates into violent rhetoric. Sexism-related discourse is deeply tied to sexism itself (35.1%), while also intersecting with politics (15.4%), ridicule (11.6%), and threats (14%), indicating a mix of discriminatory and contemptuous expression.
Anti-refugee content is overwhelmingly political (46.3%) and racialised (38.6%), with a notable share of threatening messages (17.7%), pointing to narratives that frame refugees as a collective danger. Finally, anti-Roma discourse is marked by racism (30.40%) and political framing (10.98%), though with comparatively lower shares of religious (4.76%) or sexist (2.7%) content.
Across all baselines, racism, religion, politics, and threats emerge as the dominant dimensions of online hate, with antisemitic and anti-Muslim content showing the highest levels of ideological and violent overlap.