AI-based sentiment analysis of followed Twitter/X feeds
By far the most charged feed. Nearly every second post is classified as negative. Notably, the negativity rate was still 15–25 % in 2024, climbing consistently to 40–60 % from mid-2025 – suggesting an increasingly polarised account selection. Accounts like MaryanneDemasi (100 %), PGtzsche1 and WSJhealth reach 88–90 % negativity. The model detects not just insults but also accusatory questioning, exposé rhetoric and moral outrage directed at institutions.
| Account | Posts | Neg-% |
|---|---|---|
| @WSJhealth | 20 | 90% |
| @ProfessorAkston | 20 | 90% |
| @consiliumsci | 20 | 90% |
| @ID_Denmark | 19 | 89% |
| @MartinNeil9 | 34 | 88% |
| @TracyBethHoeg | 21 | 86% |
| Account | Posts | Pos-% |
|---|---|---|
| @VPrasadMDMPH | 20 | 65% |
| @MartyMakary | 19 | 42% |
| @ArmchairViro | 23 | 30% |
| @RanIsraeli | 27 | 30% |
| @Eurosurveillanc | 18 | 28% |
| @TranspariMED | 20 | 25% |
@nightglow98 is a music account with a predominantly neutral tone (85 %). The few negative posts (10 %) often arise from implicitly negative song lyrics or critical commentary. Original videos and short comments dominate the feed.
| Account | Posts | Neg-% |
|---|---|---|
| @nightglow98 | 340 | 10% |
| Account | Posts | Pos-% |
|---|---|---|
| @nightglow98 | 340 | 3% |
| Account | Posts | Neg-% |
|---|---|---|
| @SHomburg | 119 | 8% |
| Account | Posts | Pos-% |
|---|---|---|
| @SHomburg | 119 | 2% |
The smallest feed with only three accounts – but the most negative of all. Two thirds of all posts are classified as negative. With MikeBenzCyber, cohler and _____Salt___, this is no longer statistical noise but a statement about the tone of these accounts themselves. The model responds to the consistently alarmist, conspiracy-adjacent language – even without explicit profanity.
| Account | Posts | Neg-% |
|---|---|---|
| @cohler | 25 | 76% |
| @MikeBenzCyber | 21 | 71% |
| @_____Salt___ | 26 | 35% |
| Account | Posts | Pos-% |
|---|---|---|
| @MikeBenzCyber | 21 | 29% |
| @cohler | 25 | 20% |
| @_____Salt___ | 26 | 4% |
The most heterogeneous feed with 226 accounts reflects a broader range of opinion. A 32 % negativity rate is elevated but explainable by the mix of critical-conservative accounts and more factual voices. The most negative outliers – HopeRising19, LowEndNetwork (90 % each) – are English-language accounts with strong political framing.
| Account | Posts | Neg-% |
|---|---|---|
| @windscribecom | 41 | 90% |
| @MidwesternDoc | 61 | 90% |
| @durov | 20 | 90% |
| @LowEndNetwork | 20 | 90% |
| @Snowden | 20 | 90% |
| @nixcraft | 20 | 90% |
| Account | Posts | Pos-% |
|---|---|---|
| @DoktorWeigl | 20 | 55% |
| @DesmetMattias | 17 | 41% |
| @davidkorowicz | 19 | 37% |
| @hummelbubu | 202 | 33% |
| @TheBorisBecker | 68 | 32% |
| @jfodlovesyou | 19 | 32% |
Overall: 36 % of all analysed posts are negative, only 8 % positive. A keyword-based approach had previously classified ~90 % as neutral – BERT detects subtler negativity far more reliably. The feeds reflect a consistently sceptical, institution-critical milieu that is increasingly polarised.
The three Nitter feeds under examination paint a consistent picture: they do not represent a cross-section of the German-speaking Twitter/X public, but rather a thematically narrow, institution-critical information space. What unites them is a fundamental distrust of state institutions, mainstream media, the pharmaceutical industry and the scientific establishment – a milieu that perceives itself as enlightened and system-critical.
@Impf_Info is the most thematically focused feed. With 42 % negative posts and a negativity rate that more than doubled from 2024 (15–25 %) to early 2026 (45–60 %), it shows a progressive radicalisation of content selection. Core topics revolve around COVID vaccine scepticism, pandemic policy criticism, cover-up allegations against health authorities (RKI, CDC, EMA) and alternative medical narratives. Notably, reputable journals such as The Lancet, NEJM and Cochrane appear in the feed – but selectively, to lend scientific credibility to existing scepticism. The dominant rhetorical mode is not open hostility, but the pose of the concerned, rational truth-seeker fighting an overpowering mainstream.
@StHomburg is the smallest feed by account count, but with 64 % negativity the darkest of the three. MikeBenzCyber, cohler and _____Salt___ share a worldview in which state institutions are fundamentally corrupt (CIA, "Deep State", WHO), established science is an instrument of control, and conspiracy narratives (Epstein, chemtrails, elite paedophilia) are framed as revelations. The language is consistently alarmist – even without profanity, the BERT model reliably detects the negative undertone.
@SZwanglos is the most heterogeneous feed and therefore hardest to characterise. Spanning 226 followed accounts, its spectrum ranges from factual science journalists to accounts that frame violent crimes by migrants as political arguments. The 32 % negativity rate is the lowest of the three feeds, but masks the fact that the most negative accounts set particularly strong signals: statements like "Europe has committed suicide" or direct Nazi comparisons for EU policy appear alongside more sober contributions from politicians or scientists.
Methodological caveat: The BERT model used (germansentiment) was trained on German-language texts. English-language posts – which make up a significant share across all three feeds – may be classified less accurately. Short posts, irony and quoted material from external sources can lead to misclassification. The figures should therefore be understood as directional indicators, not precise measurements.