November 2025 – April 2026
Contents
Dataset: 678 TikTok accounts · 300 Telegram channels · 440 Facebook pages · 34 Facebook groups · 14 YouTube channels · 24 TV/podcast sources. November 2025 – April 2026.
Templated narratives, genuine automated coordination at the Russian-language Telegram origination layer, and a coordinated Armenian-language opposition amplifier network producing the bulk of anti-government messaging together form an active Foreign Information Manipulation and Interference (FIMI) environment around the June 7 parliamentary elections. The patterns echo Russian information operations seen elsewhere in the region – most closely Moldova's 2025 parliamentary elections, where Repsense documented the same playbook, with looser parallels to Germany and Lithuania. Rather than being built from scratch for this vote, the approach appears adapted from those earlier campaigns.
Thirteen coordinated Armenian-language opposition accounts produce the bulk of the platform's anti-government messaging. The government is no longer absent: in the closing window the Prime Minister's own account became the single most-watched political voice on the platform (15.2M views), and on operational-window chains Pashinyan himself posts fast – a 3.8-day median, sometimes leading the narrative. The lag problem is real but uneven: other government figures (Marukyan, Simonyan, Papoyan) still arrive late, and the opposition long tail collectively pulls roughly twice the PM's reach. The government is one loud, fast voice surrounded by many slower ones – present, but outnumbered.
Strict-lag analysis splits the operation in two. The fast, automated coordination sits in two places: inside Russian-language Telegram (the krepkii_oreshek–shkvarka2 pair, 1,655 near-duplicate sentences at a 0.6-hour median) and inside the Armenian-domestic outlet layer (sister-brand outlets cross-posting within 24 hours; strongest pair araratnews.am ~ civic.am). The Russian-state→Armenian-TikTok link runs at a 99–111 day lag – too slow to be a real-time laundering relay, but both ends are pushing the same Russian state line. Russia supplies the framing; the platforms reproduce it on their own clock rather than receiving a live feed. Because they reproduce it in parallel rather than relaying it, no single takedown stops the operation; durable counter-framing, not removals, is the only proportional response.
Across 15,037 comment-thread sentences from 12,207 Armenian-language Facebook comments, the audience-side reception amplifies the editorial line and never contradicts it. On two issues – the legitimacy of the EU monitoring mission and the "NGOs as Western agents" framing – 100% of classifiable comments take the Russia-aligned position. There is no visible fact-checking or counter-disinformation layer in the comments to build societal resilience.
The ruling party lost both mention share and high-reach editorial real estate across the closing window, its reach share falling 6.7 points as coordinated opposition framing gained the higher-impact slots. On the most geopolitically sensitive question – whether Armenia should diversify its partnerships or remain aligned with Russia – a 1.06-point gap separates what the public actually believes (strong preference for diversification at −0.71) from where the media supply leans (toward Russia-first at +0.35). This is the largest mismatch in the entire dataset, and it has widened since April (0.91 in April): the information environment is pushing in the opposite direction from where the majority of the population actually stands, and the gap is sharpening.
Russia executes foreign information operations on a standardized playbook. Rather than constructing unique influence campaigns from scratch, Russian actors transpose proven narratives, infrastructure, and distribution methods across target countries. The evidence from Armenia is unambiguous: Russian operators are reusing templates that worked in Moldova, Germany, and Lithuania with minimal adaptation to local context.
Two key templates were identified with direct substitution instructions. The first instructs readers to replace "Moldova" with "Armenia" and "Sandu" with "Pashinyan" – found 14 times across channels. The second propagates the "Potsdamian World Order Collapse" narrative, replicated 12 times. This is documented Russian operational methodology: template reuse, minimal localization, maximum distribution scale.
Armenia is subject to the same FIMI playbook that Russia has deployed across Europe, adapted in each case to the political fault lines of the target country – and the patterns observed here suggest that these templates are designed for reuse at scale, with infrastructure that can be repositioned from one electoral cycle to the next with predictable outcomes.
The Armenian information environment carries a consistently Russia-aligned narrative supply, but the mechanism is not a fast, automated Russian relay into Armenian feeds. Under strict-lag analysis the picture resolves into three distinct parts. First, within Russian-language Telegram there is genuine automated coordination – the krepkii_oreshek–shkvarka2 pair alone shares 1,655 near-duplicate sentences at a 0.6-hour median lag. Second, the link from Russian-state Telegram to Armenian TikTok is slow: a roughly 40-day median across the Telegram→TikTok boundary overall, and 99–111 days for Russian-state content specifically – too slow for a real-time laundering relay, even though both ends run the same Russian state line. Third, the fast operational coordination that survives strict filtering is Armenian-domestic: sister-brand outlets cross-posting within hours. Russia supplies the framing; the Armenian outlets do the fast, operational amplification.
The operational coordination is Armenian-domestic, not a Russian relay
When the propagation analysis is re-run with strict lag thresholds, the edges that survive are not Russian-state→TikTok. They are Armenian-language outlets coordinating among themselves. The strongest in-platform pair is araratnews.am ~ civic.am; the fastest operational pipeline is sister-brand outlets cross-posting within 24 hours. The 13-account opposition network's contribution is volume, replication and translation – not framing innovation, and not a Russian distribution arm.
Strongest in-platform coordination pair · Armenian opposition outlets
p < 10⁻⁴⁵
Armenian-domestic sister-brand cross-post
Armenian-domestic sister-brand cross-post
Russian influence operations in Armenia employ a five-layer narrative stack. Each layer targets a distinct emotional or cognitive vulnerability. Together they form a coherent frame that channels all grievance, fear, and frustration toward one conclusion: that Pashinyan has betrayed Armenia and that only Russian partnership can restore national security. The stack is cumulative – each layer reinforces the ones below it.
The Church narrative serves as proof of concept for full-stack integration – the only topic with confirmed presence across all five layers and confirmed cross-platform saturation: 969 Telegram messages, 452 Facebook posts, 13 videos linking Church attacks to Turkish/Azerbaijani geopolitical designs, coverage on all monitored television channels.
The narrative stack is engineered to move audiences from fear (Layer 1) through distrust (Layers 3–4) to a single policy conclusion: dependency on Russia. Each layer makes the upper layers more persuasive.
TikTok emerges as the single most powerful battleground in Armenia's information environment, and it splits cleanly in two. In the closing weeks of the monitoring period, 6,821 political videos pulled 107 million views – roughly 325 political videos a day. About 37.5% of views go to government-friendly content and 34.7% to hostile content, with measured commentary in between almost absent. Two storylines do most of the work: "the government betrayed Artsakh" (1,230 videos, 15.9M views) and "the 2026 elections will be rigged" (594 videos, 10.0M views). A coordinated network of 13 Armenian-language opposition accounts produces the bulk of the anti-government messaging – an amplifier layer whose role is volume, replication and translation, not framing innovation, and not a Russian-run distribution arm. The government is no longer absent – the Prime Minister's account is the single most-watched political voice (15.2M views) – but the opposition long tail collectively pulls roughly twice that reach. It is one loud voice against dozens.
The 13-account roster: araratnews.am, civic.am, antifakeam, oragir.news, 5tvchannel, news.armenian7, abcmedia.am, armeniatodayam, armat.media, armlife.lurer, hayeli_official, pastinfo.tv and tribune.am. The strongest in-platform coordination pair is araratnews.am ~ civic.am (316 matched sentences, p < 10⁻⁴⁵).
Six Positions Rendered Unsayable on TikTok
The following positions have been effectively excluded from TikTok's Armenian political ecosystem – not by policy, but by coordinated narrative suppression:
TikTok is not reflecting Armenian political debate – it is weaponizing algorithmic amplification to suppress it, with a coordinated clique of accounts controlling the normative conversation on the platform's most politically engaged user base.
A natural objection is that these are just outlets pushing a line detached from what ordinary Armenians think. The comment threads rule that out. We scored 15,037 comment-thread sentences from 12,207 Armenian-language Facebook comments (Armenian and Russian) over the same window, against the identical public-opinion framework. The audience-side reception amplifies the editorial framing – it never contradicts it. There is no visible network of fact-checking or counter-disinformation voices in the comments that might build societal resilience against the pro-Russian narratives. The audience is reinforcing the message, not interrogating it.
At the audience-reception layer of Facebook, the pro-government and pro-Western voice is functionally absent on every issue the public actually agrees with the government on. Whether that is a property of a self-selected commenting audience or of platform curation, the practical consequence is the same: nothing in the comments slows the narrative down.
The most concerning finding is not that the operations exist – it is that they are producing measurable effects across the entire information environment. What our monitoring reveals is a three-way misalignment: what the media is producing, what the public actually thinks, and what TikTok is amplifying are pulling in different directions simultaneously, creating an information space where different voter demographics inhabit completely different realities about the same election.
The Scale of FIMI Penetration
Across 6.42 million sentences monitored on six platforms between November 2025 and April 2026, 1,463,525 sentences – 22.8% of the entire corpus – matched one or more of the twelve FIMI narrative pairs. Nearly one in four sentences in Armenia's information environment aligns with a documented manipulation pattern, a rate consistent with a pre-positioned, industrially scaled operation rather than organic political debate.
Hostile Narrative Clusters
Topic-network aggregation across the April monitoring window identified eight hostile narrative clusters, ranked by volume, persistence, and assessed intent. Four are rated HIGH priority – requiring immediate attention – and three MEDIUM, with one WATCH-level cluster that has generated 26,327 mentions through chilling-effect amplification alone.
| Hostile Cluster | Mentions | Level |
|---|---|---|
| "Party of War" Fabrication Campaign – Opposition systematically accuses Pashinyan of inventing the war-party label to delegitimise opposition forces ahead of elections | 50,887 | High |
| EU Sanctions Threat Amplification – Coverage of EU sanctions on Georgia framed as implicit warning to Armenia; "the West punishes small countries" | 26,327 | Watch |
| Election Interference Probe Reframing – Strong Armenia investigation reframed as political persecution rather than legitimate law enforcement | 182 | Medium |
| Pashinyan Delegitimisation Campaign – Coordinated multi-angle attacks framing Pashinyan as weaker than oligarch-linked opposition figures | 166 | High |
| Kocharyan–Putin Meeting Amplification – Positions Kocharyan as parallel head of state with Kremlin backing; election interference by proxy | 61 | High |
| Russia–Genocide Equivalence Narrative – Co-opts Armenian historical memory to build sympathy for Russia's framing of the Ukraine war | 50 | Medium |
| Azerbaijan/Erdogan Sovereignty Threat – Frames the government as having surrendered sovereignty; implicit argument for returning to Russian security guarantees | 48 | Medium |
| Fabricated Content (VT Foreign Policy) – Outright fabricated content targeting Pashinyan published on known influence-operation platform | 40 | High |
The Compound Risk in Video
The video ecosystem is where these narratives do their most damage. Of 14,255 videos analysed in Phase 3, 4,590 (32%) scored high on both political agitation and viral potential simultaneously – a compound-risk cohort that has already generated 81.2 million contacts at 91% negative sentiment. Content scoring high on disinformation risk carries a mean viral potential of 7.80, well above the dataset average of 6.28, which means the algorithmic environment is not penalising disinformation but actively rewarding it with greater distribution.
How FIMI Operations Reshaped the Media Landscape
The media ecosystem shifted dramatically across the three monitoring periods, and the pattern is consistent with a coordinated operation targeting the ruling party while fragmenting and amplifying opposition voices. The primary target – Civil Contract, the ruling party – saw its reach collapse by 18.4 points in Phase 3 before partially recovering through deliberate high-reach placements. Meanwhile, opposition forces gained reach share across the board, with Armenia Alliance achieving per-mention impact parity (0.262 vs 0.260) for the first time despite producing 4.6 times fewer mentions.
Narrative Indistinguishability
Foreign-origin narratives have become deeply embedded in domestic political debate. When asked to identify the origin of statements, Armenian respondents correctly attributed foreign vs. domestic narratives at rates equivalent to random guessing (48–52% accuracy). This signals a polarised media environment where foreign narratives are already deeply engrained into the narrative landscape.
Repsense's analysis of Moldova's 2025 parliamentary elections documented a sharp acceleration in algorithmic amplification of FIMI content during the final 45 days before polling. The same escalation pattern has now played out in Armenia. With the June 7 election days away, the acceleration window has largely elapsed – and the dynamics observed in Moldova have materialised across all monitored platforms.
Assessed Risks
| Risk | Assessment | Level |
|---|---|---|
| Coordinated Volume Surge – The 13-account Armenian-language opposition amplifier network that produces the bulk of anti-government messaging is positioned to intensify output in the final days, following the Moldovan pattern of sharp pre-election spikes. | Output from this network has shown no sustained quiet period across the monitoring window; further intensification in the final week is consistent with the Moldovan precedent of sharp pre-election spikes. | Critical |
| Narrative Saturation – Foreign-origin narratives are already deeply embedded in the information landscape, with 22.8% of all monitored sentences matching FIMI patterns. | In a polarised media environment where foreign and domestic narratives are effectively indistinguishable, corrective messaging faces structural barriers in the remaining time. | Critical |
| Event Exploitation – High-visibility events are instrumentalised quickly inside the Russian-language Telegram origination layer and among Armenian-domestic sister-brand outlets (within hours), then surface on Armenian TikTok on a longer lag (≈40-day median across the Telegram→TikTok boundary). | The government's response is uneven: Pashinyan himself can move fast (3.8-day median, sometimes leading), but other officials still arrive after the opposition framing has saturated. The risk is not relay speed but a response that is fast at the top and slow across the rest of the government. | High |
| Voter Perception Manipulation – The five-layer narrative architecture (fear → dependency → institutional distrust → leader betrayal → electoral futility) is designed to suppress pro-Western turnout. | Emotional targeting of the persuadable 18–35 cohort through TikTok – where the government holds under 2% of content share – is at maximum operational intensity. | High |
| Post-Election Contestation – Pre-positioned infrastructure can pivot to delegitimisation within hours of results. | The electoral manipulation narrative (1,055 labelled videos, 517,675 FIMI-matched sentences) is pre-built for either outcome: "rigged election" if the ruling party wins, "people's will" if opposition prevails. | High |
The 45-day pre-election acceleration window documented in Moldova has now largely elapsed in Armenia, with coordinated infrastructure fully operational and actors synchronised across platforms. With the election days away, the operational focus is likely shifting from narrative amplification to post-election contestation infrastructure.
The findings above describe a diagnosis; the implications below are what they mean for how the June 7 vote will actually be fought – where the leverage sits, what kinds of response can and cannot work, and what to brace for after polls close.
The patterns echo earlier Russian information operations in the region – most closely Moldova's 2025 parliamentary elections, with looser parallels to Germany and Lithuania. The approach appears adapted from those campaigns rather than built from scratch for this vote, which means the tactics are known in advance and the counter-measures that worked – or failed – elsewhere are directly relevant here.
Thirteen coordinated Armenian-language opposition accounts produce the bulk of the platform's anti-government narrative, and TikTok's user base skews young (18–35) – the most persuadable electorate. The government is no longer ceding the platform: the PM's account is now its single most-watched political voice, and Pashinyan himself posts fast (a 3.8-day median, sometimes leading). The weakness is not the top of the government but the rest of it – other officials still arrive after the framing has saturated, and even the PM is one voice against an opposition long tail that pulls roughly twice his reach. Being present is not the same as contesting the space on equal terms.
Despite a 1.06-point narrative gap on geopolitical orientation – the media supply leaning Russia-first (+0.35) while the public sits firmly toward diversification (−0.71) – the public has not been converted; it is being out-shouted. The strongly held public position simply isn't reaching the volume the information environment gives the opposite view. Platform volume is not a proxy for the electorate's mood: anyone reading the 91–94% "rigged" framing as a forecast is reading it backwards against a public that is 52% confident in the vote. If the information environment can be challenged, public preference can reassert itself.
Because the Russian-state→Armenian-TikTok link is slow (a 99–111 day lag – too slow for a live relay, though both ends run the same Russian state line) and the genuine fast coordination is internal to two separate layers – Russian-language Telegram and the Armenian-domestic outlet network – there is no single chokepoint to close. Removing one platform, or one account cluster, leaves the rest reproducing the same framing. The only intervention proportional to a parallel-production structure is durable, sustained counter-framing that travels the same channels. Enforcement and takedowns buy time; they do not change the script.
This is the flip side of the previous point, not a reversal of it: the public's underlying preference holds in the poll, but nothing is actively defending it in the open. The comment layer shows no fact-checking or counter-disinformation response – on the issues where the public actually sides with the government, the pro-Western voice in the comments is effectively zero, and on two of them it is exactly zero. Whether that silence reflects a self-selected, already-aligned commenting audience or feed curation that buries counter-speech, the effect is the same: the resilient majority is invisible in the feed. Resilience of belief has to be converted into an active, visible defense – built deliberately and early, before the framing saturates. Once a narrative has circulated for months, late correction competes against an audience that has already absorbed it – which is why the uneven government response (fast at the top, slow across the rest) leaves so much of the field undefended.
The Russia-aligned playbook saturates the channels that reach first-time voters – Telegram and TikTok, where youth penetration runs around 50% and over-65 use is in single digits. The government's response, when it comes, lands on television, which reaches over-65s overwhelmingly but only a minority of 18–19-year-olds. The two sides are not merely out of sync in time; they are demographically desynchronised. A counter-messaging strategy that lives on television cannot reach the audience consuming the framing it is trying to answer.
The highest-consequence risk is not the campaign – it is the morning after. The most election-relevant narratives (the vote will be rigged, there is no information freedom, the country is on the wrong track) already run at 85–95% Russia-aligned share on Telegram and TikTok against a public that leans the other way. That framing is pre-built for either outcome: "rigged election" if the ruling party wins, "stolen by the West" if it does not. The infrastructure that spent six months shaping the campaign can pivot to contesting the result within hours of polls closing.
Tools & Methods: The Repsense Havel platform deployed natural language classification (166,982 sentences across 12 narrative pairs), actor network analysis (identifying coordinated relay patterns via message fingerprinting), and temporal analysis (tracking narrative latency from source to saturation). TikTok analysis included computer vision classification, audio extraction, and on-screen text recognition across 21,561 videos from 678 monitored TikTok accounts. FIMI narrative validation used semantic similarity thresholds of 85% or above across 6.42 million sentences.
Survey Integration: The PractNet/SALK public-opinion survey – a nationally representative MRP model (n = 1,955, projected to Armenia's full adult population) – ran in parallel with media monitoring (fieldwork April 8–25, 2026). Twelve paired narratives covering the major axes of the election-campaign debate, plus eight standalone items (L01–L08) on legitimacy, the EU referendum, and election integrity, enable direct comparison between media supply and public demand. The same MRP-adjusted benchmark anchors both the platform-vs-poll comparisons and the media-lean gap analysis.
Limitations: Encrypted platforms (Signal, WhatsApp) and real-time content removal mean identified operations represent a floor, not a ceiling, of actual interference. TikTok analysis covers TikTok only – no Telegram or Facebook video. Some clustering artefacts may affect individual results, and frame labels are pending native-speaker review. This analysis is non-exhaustive.
FIMI Executive Report on Armenia · November 2025 – April 2026