Reporting and research: Zyma Islam, Mir Rownak, Nourin Sultana Toma, Abdullah Hel Bubun, Tarek Hosen and Abir Ayon; Graphics: Anwar Sohel
On December 20, 2025, a female student activist at Dhaka University shared a screenshot of a Facebook post that called for violence against a fellow activist.
Alarmed, she wrote: “Getting seriously worried about … [name withheld] bhai’s well-being. Your views do not need to align with his in order to be concerned for his safety, especially in these dire times. If you recognise this person in the screenshot, or have any way to reach him, please step forward and take legal action.”
By the time this report was filed, the post drew over 10,000 reactions -- 8,500 of them “haha,” signalling mockery rather than concern.
While social media users agreeing or disagreeing with a post may react with “like,” “love,” or “haha”, a closer look at these 8,500 “haha” reactions revealed that one in every five came from suspicious profiles.
Many usernames appeared foreign, written in scripts other than Bangla or English, and lacked profile photos or personal details. Among them were accounts such as Kokou Khelios from Togo and Olivier Randrianjaka from Madagascar -- names and listed locations with no plausible connection to the post.
An investigation by The Daily Star has found that such accounts do not land on a Bangladeshi student’s post by coincidence. They are bot profiles that can be easily purchased online to attack opponents or fabricate support.
As part of this investigation, The Daily Star bought nearly 30,000 reactions across nine meme posts, using five dummy Facebook profiles and sourcing them from four separate “click farms” over two days.
Many of the same profiles delivering fake reactions for our posts were also active across the pages of dozens of political actors and at least six candidates running in the February 12 election -- supporting one side or trolling the other.
Experts warn that this kind of bot activities poses a serious risk ahead of the election, as it is often deployed to manufacture perceptions that are far from reality.
It also violates Meta’s policies.
Facebook’s Community Standards prohibit any artificial amplification of engagement, explicitly banning fake accounts, automated activity, or coordinated click farm operations to inflate reactions, followers, or shares.
“We do not allow attempting to or successfully selling, buying, or exchanging for engagement, such as likes, shares, views, follows, clicks, use of specific hashtags, etc.,” the standards state.
Yet the influence operations network uncovered in this investigation suggests that Meta’s detection systems are failing to catch them.
Meta did not respond to requests for comment sent via email.
Image
The Daily Star analysed “haha” and other reactions across 263 Facebook posts over the last two months and found at least four clusters of politically aligned users who move together on the platform like an organised army.
These clusters swarm posts from political or ideological opponents with waves of “haha” reactions, creating the appearance of mass ridicule and suppressing the visibility of targeted content.
Two of these clusters appear aligned with Bangladesh Jamaat-e-Islami, while one each is aligned with the Bangladesh Nationalist Party (BNP) and the ousted Awami League. However, no official links with any of the party were investigated.
Their operations go far beyond simple trolling. By coordinating large volumes of hostile reactions, they exploit the way Facebook’s algorithm works.
Internal Meta documents leaked in the 2021 “Facebook Papers” revealed that when a post receives signs of trolling, the platform may automatically reduce its visibility. This means that coordinated “haha” attacks do not merely mock a post, they can actively push it out of public view.
Such coordinated attacks are also easy to carry out. It only takes a willing buyer, a mobile wallet, one of the many “engagement” sellers on Facebook, and just about five minutes.
Findings show that what appears to be a genuine public backlash or ideological confrontation is often a manufactured reality, bought for a few thousand taka and used to manipulate public perceptions just days before the 13th general election.
Image
“The government asked Meta to activate five types of protections ahead of the elections. We have already met with the head of public policy in the region to that effect.”
Faiz Ahmed Taiyeb Special Assistant to the CA for Post, Telcos and ICT
The 263 posts examined come from a cross-section of Bangladesh’s political and media landscape: 55 were from pro BNP actors, 52 from pro Jamaat entities, 42 from pro Awami League figures, and 24 from National Citizen Party supporters. We also reviewed 29 posts from media outlets, 14 from individual journalists, 20 from right wing actors without identifiable party links, and 27 from left wing users.
Together, these posts drew 8,85,811 reactions. Of them, 5,49,129 (62%) were “hahas,” with at least 19,708 coming from profiles that bear all the hallmarks of bot accounts.
For this investigation, a group was identified as a cluster if its members showed explicit support for posts within their own political camp and concurrently reacted with “haha” to at least seven posts from opposing camps in tandem with other profiles.
Based on this criterion, two Jamaat aligned clusters were identified, with 326 and 16 members each, alongside an Awami League aligned cluster with 134 members and a BNP aligned cluster with 12.
Analysis shows the two pro-Jamaat clusters are the most combative, primarily targeting the pro-BNP users with 1,465 attacks within this limited dataset. Their second most frequent target was the left-wing camp, facing 1,196 attacks, while the Awami League saw a comparatively lower 207 strikes.
Two Dhaka University student leaders -- one from a leftist organisation and another from Chhatra Dal -- are among those attacked by these pro-Jamaat clusters.
The Awami League aligned cluster focused primarily on Jamaat-affiliated posts, followed by the NCP, and carried out fewer attacks on BNP and left leaning content. Their targets included an NCP candidate from Dhaka as well as the female Dhaka University student with whom we began the story.
The BNP cluster, the smallest group identified, was also the least active, striking Jamaat posts 55 times and NCP posts 30 times.
Despite their political differences, each of these clusters shared one target in common: the media. Posts from the daily Prothom Alo, The Daily Star, and New Age were frequent victims of coordinated “haha” attacks.
The Jamaat-aligned clusters attacked posts by media organisations and journalists 515 times, making them the top trollers of the media. The AL cluster targeted the media 287 times, while the BNP cluster showed minimal activity with only five recorded attacks.
However, the most prolific harassment of journalists and media outlets came from a separate right wing cluster with 234 profiles and no identifiable party link. It targeted journalistic content 970 times, nearly double the volume of the most active political clusters belonging to pro-Jamaat entities.
This group also showed strong ideological leanings, attacking left wing posts 761 times, BNP content 488 times, and Awami League content 111 times.
Image
This investigation uncovered a network of Facebook pages that openly advertise paid services to inflate social media metrics by selling reach, reactions, and followers.
The Daily Star approached four such click farms -- Sociafy, Finix IT, Tech Dreams, and Finix IT Boosting -- posing as potential clients. To test their services, we created five dummy Facebook pages and ordered reactions for meme posts shared from those profiles.
The process could not be simpler: they responded almost instantly, asking no questions about who we were or how the engagements would be used. Between January 19 and 21, we purchased 30,000 reactions, both “love” and “haha,” across nine posts created for investigative purposes.
Nayeem Uddin, the owner of Sociafy, described how the system works.
“We have contacts with people who maintain servers [collections] of Facebook profiles,” he said, adding that 500-600 people maintain such “servers” in Bangladesh.
According to him, Bangladeshi server owners build these collections by purchasing Facebook accounts for around Tk 15-16 each, while foreign profiles cost slightly less. For international accounts, they rely on servers in Thailand, Vietnam and Pakistan.
“I send the link for which reactions are being purchased to one of these people with servers, and they do the rest,” he explained.
Buying engagement is also cheap. Reactions typically cost between Tk 120 and Tk 200 per thousand, and in our case the rate dropped to Tk 500 for 5,000 reactions when purchased in bulk.
While reactions from profiles with foreign sounding names are delivered almost instantly -- sometimes 1,000 within a minute and 5,000 within five minutes in our case -- reactions from local sounding names take longer and cost more.
These “domestic” reactions usually exceed Tk 200 per thousand and may take 24 to 48 hours to arrive. However, this can be expedited to 24 hours with extra money, signalling the deployment of pre-programmed bot swarms, our investigation found.
Unlike the seamless delivery of international batches, “Bangladeshi” reactions often appear inconsistently. They may trickle in over 12 hours or arrive in irregular bulk bursts every 15 to 20 minutes, as service providers often struggle to keep “domestic” accounts ready for immediate use.
Many of these bot accounts use AI generated profile photos and contain little or no personal information.
When The Daily Star asked for “authentic Bangladeshi” profiles, they responded with slower deployment times. However, these accounts were noticeably more sophisticated, featuring active timelines, multiple photos, and patterns resembling those of real users.
“The Bangladeshi service is a little slow because these are organic,” one seller told us. Another said it is better to use “Bangladeshi bots” because they were less likely to “drop.”
“Dropping” refers to reactions disappearing if Meta detects fake profiles and removes them from the platform.
Despite being independent companies with no visible connection to each other, all four agencies appeared to draw from the same shared pool of bot profiles.
Our investigation established a direct link between these commercially available engagement services and the coordinated digital attacks across the political spectrum.
At least 354 bot profiles that we purchased to boost our meme posts were also among the accounts delivering “haha” reactions to the 263 posts analysed for this report.
But these bots do not only attack, they also generate artificial popularity.
This investigation found that at least 547 bot accounts identified in our dataset actively support or follow the Facebook pages and profiles of at least six candidates contesting the February elections.
Two of these candidates are from the BNP, two from Jamaat, one from Islami Andolan Bangladesh (IAB), and one an independent.
The two BNP candidates from Dhaka and Barguna collectively have 6,098 bot followers, with at least 356 of those bots appearing in the pages of the click farms from where The Daily Star purchased fake engagements.
The Jamaat candidates show an even larger footprint. Its two candidates -- one from Barguna and one from Cumilla -- have 5,312 such fake followers. At least 135 of the bots came directly from the same seller pages that supplied fake engagements to this newspaper.
An IAB candidate from a Brahmanbaria constituency and an independent candidate from Joypurhat also have bot followers of their own, illustrating how widespread artificial amplification has become across the political spectrum.
In several instances, the investigation discovered that politicians from opposing parties share the same bot followers, suggesting that these networks are not ideological supporters but simply digital assets available to anyone willing to pay.
Nayeem Uddin, the Sociafy owner who sold engagements to The Daily Star, is familiar with these patterns.
A first year student at Noakhali Degree College, he confirmed that politicians regularly purchase engagement services, although he declined to give any name.
“I have had experiences where an Awami League politician came and purchased ‘haha’ reactions for a post, and another BNP politician purchased ‘love’ reactions for the same post,” he said, describing a marketplace where political rivals draw from the same pool of bots.
This investigation did not examine the bot followers of all candidates. It only explored the candidates who share bots from the same click farms that sold fake engagements to the nine meme posts created for this investigation.
Of the six candidates, three did not respond to calls or text messages. The remaining three said they were unaware of any bot activity on their pages, explaining that their social media accounts are managed by party supporters or relatives.
Shahriar Islam, a Rajshahi based entrepreneur behind Tech Dream, which also sold fake engagements to The Daily Star, explained why identical bot profiles appear across so many political pages.
“We use the same servers,” he said.
Tech Dream openly advertises its services on Facebook and is currently running seven ads offering engagement boosting packages.
One of the servers Shahriar and others rely on is Fatherpanel.com, an online click farm. Its listings show that “Bangladeshi” Facebook reactions, including “haha” and “love,” cost less than Tk 1 per thousand, with delivery times between 15 and 49 hours.
It also offers a range of bot profile categories, including “100% Bangladesh,” “BD+Mixed,” “Arab,” and “Vietnam,” with Bangladeshi labelled profiles priced the highest. Payments can be made directly through bKash, Nagad, Upay, and other mobile wallets.
Saiful Islam Sami, who runs this farm, said he sources foreign bots from servers hosted in Vietnam, Pakistan, and other countries. “I am essentially a reseller for them.”
For domestic bots, he depends on local suppliers who maintain collections of Facebook profiles.
“I only provide the services,” he insisted. “I have no control over how people use them, or whether they are used politically.”
The data were collected from December 2025 till mid-January 2026 from public posts with sizable reactions. The investigation turned data on reactions across posts into a map of coordinated political “clusters” by analysing how users behaved, and grouping accounts based on their shared patterns of activity.
We aimed to maintain a balanced proportion of posts across the various political camps, but remained blind to the volume of profiles which interacted with each post. This ensured that the density of the reactions was not a result of our selection, but a reflection of the digital force deployed against specific targets.
Across 263 posts, the investigation examined “haha” reactions and supportive reactions, distributed into two distinct categories belonging to each political party. The investigation processed a dataset of about half a million reactions across the sampled posts.
The first category served to establish a baseline for support. By analysing 110 posts where users likely reacted with 'love,' we determined the political affiliation of a user. These posts frequently featured polarising rhetoric, divisive narratives, or direct disinformation.
To document “haha” reactions, no satire or humorous posts were included. Instead, neutral-toned or factual posts were prioritised, so that any “haha” reaction can be confidently interpreted as a tool for suppression or mockery rather than an organic response to humour. As many as 153 posts were analysed to determine coordinated suppression.
The investigation sorted the profiles into politically-aligned clusters based on their behaviour, i.e. if they showed explicit support for posts within their own political camp and concurrently reacted with “haha” to at least seven posts from opposing camps in tandem with other profiles.
