Contents
A familiar FIMI architecture was active throughout the campaign: Russian-state Telegram channels, Armenian-language relay networks and a coordinated TikTok amplification clique. The model closely resembled earlier operations in Moldova and, to a lesser extent, Germany and Lithuania. Yet Civil Contract still secured 49.8% of the vote on the highest turnout since 2017.
Russia combined public pressure with a quieter narrative campaign. The first generated headlines but failed to shift the election result. The second—built around Telegram networks, TikTok amplification and coordinated narratives—continued operating long after the votes were counted.
A coordinated group of 13 accounts dominated Armenian political TikTok, while government voices remained largely absent. Official accounts consistently responded after narratives had already spread, leaving the platform's most influential political space effectively uncontested.
Russian-state channels set the themes, Armenian-language Telegram channels adapted them, and TikTok accounts amplified them for younger audiences. Because the system relies on parallel replication rather than a single distribution channel, removing individual accounts does little to disrupt it.
Platform narratives were consistently more extreme than public opinion, often by wide margins. Yet the electorate ultimately proved more resilient than the information environment suggested. The online conversation leaned heavily toward Russia-aligned narratives; the vote did not.
Russia runs foreign information operations from a standardized playbook, transposing proven narratives and infrastructure across target countries with minimal localization. Armenia carries the same constellation of indicators seen in earlier operations – most closely Moldova’s 2025 vote, with looser parallels to Germany and Lithuania. The clearest tell is the ignition: as late as March 2026 the “Western Realignment” narrative that would dominate the campaign had exactly one matched mention. Thirty days later it was everywhere, switched on fully formed.
Armenia is subject to the same FIMI playbook Russia has deployed across Europe, adapted to local fault lines – templates designed for reuse at scale, on infrastructure that repositions from one electoral cycle to the next with predictable outcomes. The 7 June result confirms the most important parallel of all: as in Moldova, the operation shaped the information space without capturing the vote.
The Armenian information environment carries a consistently Russia-aligned narrative supply, organised across three distinct layers. Layer 1 is the framing layer: Russian-state Telegram channels set the script – talking points, framings, emotional register – without addressing Armenians directly. Layer 2 is distribution: roughly fifty domestically operated Armenian-language Telegram channels produce content on the same script, independently and within hours. Layer 3 is amplification: a 13-account TikTok clique repackages Layer-2 content for the youngest, most opinion-formative demographic. Russia supplies the framing; the Armenian outlets do the fast, operational amplification.
The government arrived 100–140 days late
The decisive asymmetry is timing. The same-operator distribution layer lifts a narrative onto the audience-facing platform within 1–47 hours, so the reaction window for first-mover counter-framing is roughly 24 hours. The government did not use it. Across the propagation chains, official accounts arrived at a 100–140 day median: the Prime Minister's official account posted at a 139-day median, after TikTok amplification 97% of the time; armgovernment at 135 days, 90% late. The state was not a fast voice surrounded by slow ones – it was structurally late, entering chains long after the framing had set.
The Russian playbook in Armenia runs on a six-narrative stack. Each narrative targets a distinct cognitive or emotional vulnerability; together they form a coherent frame in which all grievance, fear and frustration channels toward one conclusion. The stack is self-reinforcing: if democracy is a facade, elections are rigged; if the West uses Armenia, NGOs are agents; if Russia is the only guarantor, diversification is suicide. Every one of the six grew through the campaign and none declined – and the two democracy narratives grew fastest, because that is where the attack was aimed: at the legitimacy of democratic governance itself.
The ignition was fast. Five topics arrived together in April 2026 and displaced the background policy discourse that had dominated the prior five months, growing in lockstep from 20,572 weekly mentions in summit week to 42,039 in the final week of May. "NGOs are foreign interference" carried the largest TikTok amplification of the stack – a +71-percentage-point gap between platform stance volume and public opinion – and it targets EU democratic support directly.
The stack had to be challenged as a system; contesting one narrative leaves the others intact. Nobody did. For eight months a user on Armenian TikTok saw what looked like universal agreement – while the public held the opposite view on most issues.
TikTok is the platform where the manipulation gap is widest. Comparing the share endorsing the hostile, Russia-aligned pole in public opinion against Armenian TikTok stance volume, every single one of the twelve narrative pairs is amplified on the platform – by between +28 and +71 percentage points. The platform does not reflect Armenian political debate; it inflates one side of it.
Thirteen Accounts, One Ecosystem
A thirteen-account coordinated clique anchored every major anti-government narrative on Armenian TikTok, cross-posting as a functionally single operation: 5tvchannel, antifakeam, araratnews.am, armeniatodayam, armlife.lurer, armpolitics, civic.am, mesropmanukyan, news.armenian7, oragir.news, pastinfo.tv, patriotisch and tribune.am. The tightest pair, araratnews.am and civic.am, shared 316 near-duplicate sentences across 8 narratives at a 3.6-hour median lag (p < 10⁻⁴⁵).
The Debunking Trap
Counter-messaging failed for a structural reason visible in cluster #803, the anti-Pashinyan chant "Qez tesnem, cavd tanem". Three single-day surges shared the same spike pattern – two opposition-originated, the third the Speaker of the National Assembly rebutting it. The platform could not distinguish defence from amplification: the Speaker's 20-sentence rebuttal counted, volumetrically, as 20 more sentences of anti-Pashinyan amplification. Platforms count volume, not intent – when a defender repeats a hostile frame, the audience sees the frame restated and the algorithm pushes it further.
Contesting narratives by quoting them is counterproductive. The reaction window is a day, and quoting the attack reopens it. Counter-framing has to lead the agenda rather than rebut it – and land within the day, before the same-operator network lifts the narrative onto the audience-facing platform.
A natural objection is that these narratives were largely confined to sympathetic media outlets. The comment layer suggests otherwise. Across 15,037 sentences from 12,207 Armenian-language Facebook comments, audience reactions almost invariably reinforced the dominant framing rather than challenged it.
On two issues—the EU monitoring mission and the claim that NGOs act as Western agents—every classifiable comment aligned with the Russia-oriented position. Whether the result of self-selection, platform incentives or audience preferences, the effect was the same: the narratives faced little visible resistance.
Yet Armenians ultimately voted differently from what the online conversation suggested. The silence of pro-government and pro-Western voices in comment sections makes that divergence all the more striking.
The operation succeeded at one level and failed at another, and the distinction is the central finding of this report. Measured across twelve binary narrative scales over the nine-week campaign window, seven ended on the constructive side and five on the hostile side. The constructive wins cluster where facts anchor the debate – the peace process, institutions, the Turkey border. The hostile wins cluster on identity and alignment – democracy, sovereignty, geopolitical orientation – where framing, not facts, decides. The campaign was lost in the information environment even as Civil Contract won at the ballot box.
But the loud coercion campaign did not move the ballot. Moscow's open pressure – gas-price warnings, the Wildberries and Ozon suspensions, the Jermuk water block, the ambassador's recall – turned the vote into a referendum on Europe, and the referendum went Europe's way. The pro-European vote held at 49.8% on the highest turnout since 2017. Strong Armenia finished second at 23.3%, Armenia Alliance third at 9.9%; the two together hold roughly a third of the incoming parliament. The diversification preference the public held before the campaign – the strongest single lean in the dataset – survived three weeks of escalating economic coercion intact.
The Coverage Ranking Did Not Predict the Ballot
Media visibility was a poor guide to the result. Across 858,750 article mentions (1 April – 11 June), Civil Contract owned 40.2% of coverage, Armenia Alliance 16.5%, Prosperous Armenia 15.0% and Strong Armenia just 7.5%. Yet Strong Armenia, fourth in coverage, finished second at the ballot, while Prosperous Armenia, third in coverage, fell out of parliament entirely. Strong Armenia's reach ran Telegram-first (104M, the only party where Telegram out-reaches Facebook and TikTok) – precisely the platform mix the article count undercounts.
| Party | Coverage share | Result |
|---|---|---|
| Civil Contract – Facebook-first reach (1.02B); Pashinyan carried 45% of the party's coverage | 40.2% | 49.8% |
| Strong Armenia – Telegram-first reach (104M); Karapetyan campaigned under house arrest | 7.5% | 23.3% |
| Armenia Alliance – TikTok-first reach (293M); Kocharyan second-most-covered person in the race | 16.5% | 9.9% |
| Prosperous Armenia – Facebook-first reach (177M); third in coverage, below threshold at the ballot | 15.0% | Out |
The findings suggest that influence operations can shape visibility more easily than preference. What people encounter online is not always what they believe—and not always how they vote.
Repsense's analysis of Moldova's 2025 parliamentary elections documented a sharp acceleration in algorithmic amplification of FIMI content during the final weeks before polling. The same escalation played out in Armenia, following a textbook coercion arc that generated a high-volume media event every 48–72 hours into the final week.
| Window | Event | Function |
|---|---|---|
| 20–21 May | Russia restricts flower imports from Armenia | First concrete action beyond rhetoric – calibrating the pressure dial |
| 24–27 May | US Secretary of State Rubio visits Yerevan; CEC invalidates seven candidate registrations | Western-engagement high point, diluted; first institutional-process cluster feeding the "facade" frame |
| 27–29 May | Russia threatens to rip up the gas deal; EAEU summit pressure | The turning point – an economic instrument placed publicly on the table |
| 30–31 May | Wildberries & Ozon suspend Armenian sellers; ambassador Kopyrkin recalled to Moscow | Trade coercion plus a diplomatic summons, calibrated to peak days before the vote |
| 2–3 Jun | Detentions and searches; €2.2M EU visa-liberalisation support; BTK railway opening | Police-operation cluster feeding delegitimisation; EU re-enters with substance, late |
The arc ran as designed and the audience absorbed it – and rejected it. The verdict from the pre-election assessment held: the pressure did not convert votes. What it did instead was raise the salience of the European question to the highest point of the campaign and lay the groundwork for post-election contestation. The most election-relevant narratives – that the vote would be rigged, that democracy is a facade – are pre-built for either outcome, and the infrastructure that spent eight months shaping the campaign can pivot to contesting the result within hours of polls closing.
Moscow ran two campaigns. The loud one – open coercion – made the European question the question of the election and lost it. The quiet one – the 13-account clique, the Armenian-language Telegram layer, the self-reinforcing narrative stack – was never contested, and it does not demobilise on election day. The loud campaign is over. The quiet one is still running.
The findings above describe a diagnosis confirmed by an outcome. The implications below are what they mean now that the vote is in – where the leverage sits, what kinds of response can and cannot work, and what does not end when the polls close.
The tactics were adapted from Moldova 2025, with looser parallels to Germany and Lithuania – known in advance, with counter-measures that worked or failed elsewhere directly relevant here. And the most important Moldovan parallel repeated: the pro-European result held. The operation shaped the information space without capturing the vote.
Thirteen coordinated accounts anchored the anti-government narrative for the youngest, most opinion-formative electorate. The government's late climb – 2% to 4% to 8% of narrative volume, arriving 100–140 days behind the framing cycle – shows that being present is not the same as contesting the space on time. The contest needs a standing capability in Armenian, where the under-30 audience actually is, not an election-cycle project.
Despite a manipulation gap of +28 to +71 percentage points on every narrative pair, the public was not converted; it was out-shouted. The 49.8% pro-European result confirms that platform volume is not a proxy for the electorate's mood.
Because the layers reproduce the framing in parallel rather than relaying a live feed, there is no single chokepoint to close. Removing one platform or one account cluster leaves the rest reproducing the same script. The only proportional intervention is durable, sustained counter-framing that travels the same channels – and that leads the agenda rather than rebutting it. In the one week the EU set the terms, every narrative scale improved and hostile messaging fell to its lowest point of the cycle.
The 13-account clique, the Armenian-language Telegram layer and the self-reinforcing narrative stack all remain in place after 7 June. The most election-relevant narratives are pre-built for post-election contestation. The next cycle is already underway; the quiet campaign continues whether or not anyone is watching for it.
The Russia-aligned playbook saturates the channels that reach first-time voters – Telegram and TikTok – while government and EU messaging, when it comes, lands episodically and late. The EU's two constants this cycle, the High Representative and the Ambassador in Yerevan, show what continuous on-the-ground presence delivers; engagement built around summits decayed within a fortnight.
Coverage: continuous monitoring of 1,226,186 mentions across 12 platforms (Armenian, Russian and English content – social posts, traditional media and the comment layer) from November 2025 to June 2026. The operational structure was reconstructed from 1.78 million sentences across 11,896 propagation clusters, with 24 narratives tracked.
Tools & methods: the Repsense platform deployed natural-language stance classification, actor-network analysis (coordinated relay detection 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. Coordination signals were validated against cluster-overlap probability thresholds (the tightest pair, araratnews.am ↔ civic.am, at p < 10⁻⁴⁵).
Survey integration: the PractNet/SALK public-opinion survey – a nationally representative MRP model (n = 1,955, projected to Armenia's adult population) – ran in parallel with media monitoring. Twelve paired narratives plus standalone legitimacy and EU-referendum items enable direct comparison between media supply and public demand, anchoring both the platform-vs-poll gap analysis and the media-lean comparisons.
Election-day platform split: Telegram carried 11,177 mentions on 7 June – 47% of the day's entire volume, the crisis platform where the contest concentrates; Facebook reached 515M (population-scale emotional amplification); TikTok reached 87M (disproportionate reach per post, the youth platform). Project: Building Armenia's Resilience Network Against Foreign Manipulations, implemented by PractNet.
Limitations: encrypted platforms (Signal, WhatsApp) and real-time content removal mean identified operations represent a floor, not a ceiling, of actual interference. Some clustering artefacts may affect individual results, and a subset of frame labels remain pending native-speaker review. This analysis is non-exhaustive.
FIMI Executive Report on Armenia · November 2025 – June 2026