Most people in Armenian diaspora groups could not vote in Armenia’s June 2026 parliamentary election. The rules let diplomats, military personnel and their families vote abroad. Everyone else had to travel home.
So why did political messaging in those groups rise from 30 percent of diaspora-related posts in early spring to 54 percent after the vote? A DFRLab report published on 4 November 2026 tries to answer that. It should be read with its limits in view.
What was measured
The authors collected 204,552 posts from 83 public groups that presented themselves as Armenian diaspora groups. The posts ran from 1 March to 18 June 2026 and came from 7,395 authors. Meta’s Content Library was used to find the groups. Apify scraped the posts.
The report is by Givi Gigitashvili of DFRLab, with Andra Martinescu and Marius Dima of Qriton Technologies and Andreas Sjöstedt of Erebus. I could not trace Qriton or Erebus beyond the report’s own credits.
Two disclosures matter here. Meta supplied the data tool, and Meta owns the platform being studied. Separately, DFRLab is part of the Atlantic Council. The Atlantic Council’s 2023 honor roll lists Meta among donors giving $1,000,000 or more. That list covers the whole Council, not DFRLab alone. The report carries no funding statement.
None of this makes the findings wrong. It tells you who had access to the data and who has a stake in the platform.
1. A few accounts wrote a large share
The report finds heavy concentration. The ten largest groups held 49 percent of posts. The most active 1 percent of authors wrote 38 percent. About 65 percent of posts were reshares with little or no original text. Only 12 percent drew any comment.
Twenty accounts out of 6,586 active ones, under 0.4 percent, wrote 33,969 of the 175,797 pre-election posts. That is about a fifth.
In the three weeks before the vote, fourteen of the twenty posted at least 50 percent more key-issue posts per week than they had in March. Some raised it five to seven times. Eight cut it sharply in the ten days after.
Heavy posting is not fakery. DFRLab says many of these accounts appear to belong to real individuals and labels them with codes rather than names. The pattern is still a useful signal. A group can look like a crowd while a few dozen accounts set what it talks about.
2. Messaging aimed at people who could not vote
The report sorts political posts into several kinds:
- Calls for the diaspora to return and vote.
- Posts discouraging return, especially aimed at Armenians in Russia. One post said, “if you come to Armenia to vote, remember why you left.”
- Posts framing diaspora rights as a grievance, citing disenfranchisement and entry bans.
- Posts casting diaspora money as vote buying.
- Fights over who speaks for the diaspora.
Consider the audience. These posts cannot move a ballot from abroad. They can move other things: who travels, who donates, who tells relatives at home how to vote, and who feels the government is against them. The posts can still reach voters through relatives in Armenia.
3. Event spikes
The authors flag topic spikes against a 14-day baseline. The largest was the 24 April genocide commemoration, at 13.9 times baseline. Russia-related vocabulary peaked at 3.8 times baseline on 2 and 3 April, around the Pashinyan and Putin meeting in Moscow.
These spikes are mostly ordinary news-driven attention. The report does not claim otherwise. Their use is as a baseline. When a spike has no news event behind it, that deserves a closer look.
4. Copy-and-paste across groups
Here the report turns to coordination. Some accounts posted the same item into many groups within seconds.
One example is the account the report calls QLT5, which appeared in 38 groups. About 84 percent of its posts were shares from a page linked to the account owner’s organization. On 25 May an article about “Western Azerbaijan” went into at least 49 groups. Of the 40 copies with timing data, 37 appeared within eleven minutes. Some copies appeared in the same second.
The authors say this suggests Facebook’s multi-group sharing tools. Their data could not confirm which tool was used. Same-second posting points to automation or a bulk-posting feature. It does not show who ran it or why.
5. Seventeen seemingly inauthentic accounts backing Prosperous Armenia
The report identifies 17 accounts in two clusters that it describes as seemingly inauthentic. Some had little personal information and used stock or website photos. Cluster membership came from manual review of profiles, not from an automated test.
On the evening of 4 May, ten accounts from the first cluster posted the same slogan backing Prosperous Armenia, the party of Gagik Tsarukyan. The report quotes it as “On 7 June I choose N7 #we_are_gold.” Between 20:40 and 21:13 CEST they posted fifteen copies into six diaspora groups. A new account posted every two to four minutes. Each group was covered by its own small set of accounts.
That division of groups is the strongest sign of planning in the report. Seventeen accounts is also small.
Armenian fact-checkers have separately described a larger network of fake pages and accounts backing a different opposition figure, Samvel Karapetyan. Media.am fact-checkers, reported by JAMnews, described 17 pages created on 18 February and at least 60 fake accounts backing Samvel Karapetyan. The two findings come from different researchers and different evidence. Together they suggest fake support circulated for more than one candidate. I have not matched the two networks, and nothing here shows they are connected.
I left out one item in the report. It describes eleven accounts posting a photo of Tsarukyan “on August 13, 2026,” which falls after the vote and outside the study window. The report does not explain the date. It may be a typo.
What This Is Not
The report names accounts but not the people behind them, does not say who directed the activity, and does not measure any effect on voters. The heavy posters mostly amplified anti-government and pro-opposition sources, which is a finding about what they shared and not a finding that an opposition party ran them. From the text I read, the report did not analyze coordinated posting that favored the government, so a one-sided result may reflect where the authors looked.
How to read a group like this
Use these checks on any large group that claims to speak for a community:
- Count the authors. If a few accounts write most posts, you are reading them, not the community.
- Check how many posts are original. If two thirds of posts are reshares, few members are writing their own posts.
- Look at timing. Identical text in several groups within seconds means a tool, or a person copying fast.
- Check the profile. Stock photos, no history, and posting only about one party deserve doubt.
- Ask who the post can move. If the audience cannot vote, ask what the post wants them to do.
If you moderate a group, sort posts by account and look at the top twenty authors. Then compare what they posted in the three weeks before an election with the ten days after. A sharp drop once the vote is over is worth noting.
This article is part of Decipon’s Manipulation Breakdowns series, examining specific influence operations through the Influence Tactics Protocol.
Sources: