Measuring the scope of pro-Kremlin disinformation on Twitter

Following Russia’s 2014 annexation of Crimea, there has been extensive concern and research on the Kremlin’s strategic use of disinformation and information warfare. Despite high-profile discussions and policy responses in the EU and US, evidence on the breadth of Russian disinformation exposure on social media remains mixed. This study aims to empirically assess the prevalence and visibility of pro-Kremlin disinformation about Crimea on Twitter. The central research question is: How prevalent is Russian disinformation about Crimea on Twitter? Rather than testing specific strategic objectives or attitudinal effects, the study descriptively maps the supply and network visibility of disinformation and compares it to competing narratives from Western media and other actors, focusing on the importance of visibility as a precondition for influence.

The paper reviews scholarship across security studies, political communication, and computational social science on Russian information warfare and disinformation. It distinguishes disinformation (intentional falsehood) from misinformation and situates Russian efforts within concepts of information and hybrid warfare. Prior work documents both overt channels (state-controlled media like RT and Sputnik) and covert operations (e.g., IRA troll accounts) that leverage social media to spread divisive content. While false content can spread faster than truthful content, multiple studies suggest misinformation comprises a small share of typical users’ online exposure. Research on Russian media control highlights the co-optation of domestic television and a shift to propaganda ‘rewired’ for the internet era, where disinformation counters international narratives. Studies show Russian TV shaping online agendas and framing Ukraine through WWII memory and anti-fascist narratives. However, there is limited empirical analysis comparing the scope of pro-Kremlin disinformation to competing narratives across languages and platforms, motivating this study’s measurement of prevalence and source impact on Twitter.

Design: Mixed-method approach combining content-centered and source-centered measures.

Data collection: Tweets from 01/01/2014 to 12/09/2016 using the Twitter Gardenhose Streaming API (10% random sample). Relevant tweets identified by Crimea-related keywords/hashtags in Latin and Cyrillic scripts (e.g., crimea, crimean, sevastopol, simferopol; and krim, krimnash, sevastopol, simferopol). Final multilingual dataset: 773,177 tweets/retweets.

Subsamples:

• Content analysis sample: 14,529 randomly sampled English tweets/retweets from 02/18/2014 to 06/18/2014 (period around the Crimean crisis).
• Network analysis sample: 266,710 retweets (multilingual) used to construct a retweet network (nodes = users; edges = retweets). Network size reported as 167,997 nodes and 222,065 edges (largest component visualized separately).

Content coding: Two trained annotators manually coded stance toward the key disinformation statement: “The Russian Federation is not carrying out a military operation in Crimea.” Categories: agree (supports disinformation), disagree (contradicts by identifying troops as Russian), neutral (mentions military presence without attribution), or unrelated. Inter-coder reliability: Cohen’s kappa = 0.85 on 97 overlapping tweets. Disinformation operationalization included both explicit and implicit support of the misleading narrative (e.g., describing armed men as local self-defense forces), a broad definition likely to overestimate rather than underestimate scope.

Source impact (network analysis): Calculated standardized in-degree (proportion of unique users in the network who retweeted a source) for top 10 Russian state-controlled outlets (e.g., RT, Sputnik, RIA/TASS) and top 10 Western news outlets as a comparison baseline. Employed Seidman’s k-core decomposition to assess impact across network layers (cores of increasing k and their peripheries), acknowledging that core users may be more effective spreaders. Robustness checks: (1) Russian-language subset (3,000 tweets coded; 170 relevant to troop presence); (2) top-5 outlets comparison; (3) time-restricted analysis during the peak crisis (02/27/2014–03/21/2014).

Content prevalence regarding troop presence (English tweets, 02/18–06/18/2014):

• 2,354 tweets referenced armed troops (16.2% of 14,529 coded tweets).
• Of these, 263 agreed with the disinformation narrative, 420 were neutral, and 1,671 disagreed (identified troops as Russian).
• Approximately 1 in 10 relevant tweets supported the disinformation narrative. For each disinformation tweet, there were about 6.4 disagreeing tweets.
• Disagreeing tweets dominated from the first week of the operation and throughout the annexation timeline.

Network impact (multilingual retweets, 2014–2016):

• In the full retweet network (k=0; 167,997 users), top Russian state-controlled outlets were retweeted by at least 2.2% of users, while top Western outlets reached at least 6.4%. Roughly 3 Western retweeters for every 1 Russian retweeter.
• This pattern persisted across k-cores and peripheries. The Western:Russian impact ratio declined among inner cores (e.g., ~1.5 at k=2; ~1.35 at k=3; minimum ~1.24 at k=7 involving 2,770 users), indicating relatively stronger Russian presence among highly engaged users, though still lower than Western.
• At the periphery (outside k=2 core), 12.4% retweeted top Russian outlets versus 27.2% retweeting top Western outlets.
• RT (@RT_com) dominated Russian impact: standardized in-degree ~1.8% (~3,000 users) across the full network, surpassing any outlet. @BBCBreaking ~1.2% ranked second; @AFP competed in inner cores. Spanish RT (@actualidadRT) ranked 4th overall (~0.92%), while @tassagency_en was ~0.17% (about 10x fewer retweeters than RT’s main account).

Robustness:

• Russian-language tweets: pro-Kremlin disinformation outnumbered countering tweets on the troop-presence subtopic, mirroring English-language pattern of counter-disinformation dominance in English and pro-disinformation strength in Russian.
• Top-5 comparison: Russian outlets’ impact remained 34.4% lower than Western (standardized in-degree 0.037 vs. 0.050) overall, with near-parity only in inner cores from k=2.
• Crisis-only window (02/27–03/21/2014): Russian outlets had even smaller impact relative to Western competitors.

Overall: Pro-Kremlin disinformation had measurable presence but was substantially outnumbered by contradictory content, and Russian state-controlled media’s visibility was concentrated primarily in RT while remaining lower in aggregate than Western outlets.

Findings indicate that while pro-Kremlin disinformation about Crimea penetrated Twitter, it coexisted with a much larger volume of countervailing information that explicitly or implicitly contradicted Russia’s denial of military involvement. Visibility—an essential precondition for influence—favored Western outlets overall, with Russian impact concentrated mainly in RT and relatively stronger only among highly engaged users in the network core. These results suggest that, even in a strategically important case for Russia, the broader Twitter information environment presented substantial competition from Western news sources and civil society actors. The study refrains from inferring the Kremlin’s strategic objectives or attitudinal effects; disinformation might aim to sow confusion or target specific subpopulations rather than dominate overall conversation. Nevertheless, the observed volumes could still be consequential by increasing information-processing burdens and potentially eroding trust. Policy and platform counter-disinformation efforts may be most effective if they focus on highly engaged core users who are better positioned to amplify narratives.

This study contributes a dual-measure approach to assessing the scope and visibility of pro-Kremlin disinformation about Crimea on Twitter, combining manual content analysis of a central narrative (denial of Russian military presence) with multilingual retweet network analysis of state-controlled outlets versus Western media. Pro-Kremlin disinformation was present but substantially outnumbered by counter-information; Russian state-controlled media’s aggregate visibility lagged behind Western outlets and was heavily concentrated in RT. The results highlight both the limits and the foothold of Russian disinformation on Twitter during the Crimean crisis and subsequent period. Future research should investigate other topics and platforms (e.g., VKontakte, YouTube), cross-platform dynamics, temporal evolution, differential exposure and consumption, demographic targeting, and causal effects on attitudes and behaviors.

• Platform and audience: Restricted to Twitter, which skews toward US and is less popular in Russia/Ukraine; users are not representative of the general population.
• Topical and temporal scope: Focused on Crimea-related tweets; results may not generalize to other issues or time periods. Network analysis is static and does not capture temporal dynamics.
• Outcome measures: Does not assess attitudes, behaviors, speed of diffusion, memorability, or persistence of narratives.
• Source of amplification: Does not quantify the role of bots/sockpuppets; if inauthentic accounts inflate pro-Kremlin retweets, human impact may be lower than observed.
• Content coding breadth: Broad operationalization of disinformation may overestimate prevalence; English-only for main content analysis (Russian subset used for validation only).