Imagine scrolling through YouTube at midnight and meeting a split screen. Sheikh Hasina sits on one side, Jamaat Ameer Dr Shafiqur Rahman on the other, and the title promises a face-to-face clash on Khaled Muhiuddin’s talk show.
More than 352,000 viewers watched it, and thousands argued in the comments as if they had witnessed live television. They had not. No studio, no host, and no debate ever existed.
A second video was bolder still. Its caption said Hasanul Haq Inu was speaking live right after leaving jail while Prime Minister Tarique Rahman listened. In reality, editors arranged old footage to mimic a real-time exchange, and Inu has been in custody since 2024.
Both clips are cheapfakes: No artificial intelligence, just cutting, a split screen, and a trusted program’s borrowed look.
A cheap illusion at industrial scale
Neither case is an accident. A recent study by FactWatch identified around 300 similar items and then examined several platforms.
According to findings, the YouTube channel Rajnoitik Studio and the Facebook page Bangla News240 collectively published more than 300 fabricated shows that accumulated roughly 517K followers.
The fabricated broadcasts typically target sensitive political topics, elections, or public controversies to shape public opinion or push partisan agendas. The casting is telling. Fabricated statements and deepfakes regularly target key political leadership.
Why does the talk-show format work so well? A studio duel squeezes messy politics into a contest anyone can follow, and the host lends borrowed authority. A reel then strips away the channel behind it, so viewers see a clip, not its source. People struggle most when fakes wear a news format.
By combining psychological manipulation with media impersonation, these AI-generated “talk shows” and deepfakes weaponize local cultural sensitivities to shift public opinion. A single deepfake is easy to ignore, but hundreds of them create an illusion of reality.
If a viewer scrolls and sees others echoing the same political talking point, they begin to believe that the entire country feels that way.
One fake show persuades a few people; a hundred of them feel like a national mood.
Prof Sumon Rahman, founding editor of FactWatch, has warned that confirmation bias and timing can confuse even media-literate viewers.
This is a tactic of manufactured majoritarianism. A report reveals that nearly 30,000 fake reactions were purchased from four click farms. They cost roughly Tk 120 to 200 per thousand, and some arrived within minutes. A review of 263 posts counted 885,811 reactions, 62% of them “haha,” and at least 19,708 came from bot-like profiles.
“How artificial intelligence and synthetic reality shaped Bangladesh’s 2026 election,” a study concluded that coordinated narrative manipulation had become a structural feature, introduced with Bangladesh’s first AI-saturated election, where seeing was no longer believing.
Who is behind the screens?
No single mastermind appears in the record. Instead, several actors overlap. The click-farm probe found politically aligned clusters moving in tandem, yet it did not examine official party links. The administrators of those talk-show channels never answered the questions.
Some operators work alone. Anthropic’s September threat report, relayed by bdnews24, described one operator in Gaibandha who rotated 29 AI accounts over about 16 months and produced roughly 1,500 fake headlines. The company found no evidence that the Awami League ran or funded the network.
Others are plainly commercial.
FactWatch exposed a counterfeit Prothom Alo page with an invented interview with Abdullah Abu Sayeed that steered readers toward a trading platform; meanwhile, the Daily Star documented 52 deepfake videos and 31 pages offering fake Family Cards for likes.
Consequently, “who” is less a mastermind than a marketplace.
What the numbers say
FactWatch’s data offers a before-and-after picture. In its 2024 election study, video made up 71.9% of 157 unique hoaxes, and only three, or 1.9%, were deepfakes.
By 2026, another analysis of 72 AI-manipulated cases found that 49% invented what leaders did or said, and 28% used AI-edited photocards to fake quotes.
After Osman Hadi was shot on December 12, 2025, fact-checkers documented within two days fake photocards in the names of RTV, Kalbela, BBC Bangla, and Amar Desh, an AI image of a suspect, and an AI clip in which a hand grips a man yet belongs to no one.
A widely shared photograph of Shadik Kayem with the suspect was synthetic, yet BNP’s Ruhul Kabir Rizvi cited it at a rally and later apologized.
The burden cannot rest on viewers
For years, the standard advice has been to tell citizens to verify before sharing. That advice is sound, yet it is unfair on its own. A viewer on a bus, with a phone and little data, cannot reverse-search a split screen.
After the Cyber Security Act of April 2026, draft amendments would force platforms to remove harmful AI content, and Transparency International Bangladesh and other rights groups fear a return to Digital Security Act-style repression.
A vague takedown power is easily aimed at critics, and real journalism can also be dismissed as fake. Any rule should therefore target fraud, impersonation, and paid fake engagement, with judicial review and public reporting, rather than “misinformation” in general.
Broadcasters and newsrooms have a duty. Every outlet whose face or microphone is borrowed should publish a quick, searchable “this is not ours” notice, and should consider visible watermarks on original clips.
The click-farm evidence shows that candidates from several parties share the same bot pool. A public pledge never to buy engagement, never to circulate unverified cards, and to apologize quickly when fooled would cost nothing. Rizvi’s apology was a start; it should become a norm.
The cost of a counterfeit
Return, finally, to that midnight scroll. The harm of a fake talk show is not only that someone believes a debate occurred.
Political factions in Bangladesh do not just create fake videos for entertainment; they use them to manufacture deliberate, high-stakes narratives.
A falsehood once believed does not stay online. It walks into real politics.
A studio that never existed can cost a few hours and a laptop. Rebuilding trust in real ones will cost far more.
Zulker Naeen is a Research Coordinator, Centre for Critical and Qualitative Studies (CQS) and Adjunct Faculty, Department of Media Studies and Journalism (MSJ), University of Liberal Arts Bangladesh (ULAB). Views expressed are the writer’s own.