Hundreds of artificial intelligence (AI)-generated political campaign videos, many of which are not labeled for publication, have been widely circulated on social media ahead of Bangladesh’s 13th parliamentary elections on February 12, an extensive investigation by fact-checking organization Dimislav has revealed.
On January 4, a Facebook video showed an elderly man speaking in a crowded market. He said he would vote for the BNP’s election symbol, the rice bundle, and insisted that if elected, the BNP would offer cards on family, agriculture and health. With such support, people will be freed from daily manual labor in their old age, he added.
The next day, another video on Facebook featured a rural woman praising the scales (Daripara), the electoral symbol of Bangladeshi Jamaat-e-Islami. She explained that the symbol represents justice, neutrality and equal rights and said this is why she supports Jamaat.
However, closer inspection reveals a striking contradiction. In both videos, the scene cuts every 8 seconds and restarts from a new angle.
According to Dismislab, both clips were generated entirely by AI and were distributed as part of the election campaign without any public disclosure.
Dismislab recorded more than 800 AI-generated videos posted to Facebook, YouTube, and TikTok between January 1 and 15. Some of these videos were published on multiple platforms simultaneously.
On Facebook alone, 576 AI-generated videos were posted from 21 pages. 60% of these videos did not have an AI disclaimer. Even when labels were present, they were displayed inconsistently. Dismislab found that Facebook’s “AI Information” label only appears in the main mobile app and disappears when viewing the same video on Facebook Lite or a desktop browser.
This pattern was similar on other platforms. Seven of the Facebook pages reposted the same video on YouTube, and two posted the same content on TikTok. Of the 181 videos analyzed from seven YouTube channels, 94.48 percent did not have an AI label. There were no AI disclosures in the 50 TikTok videos.
Dismislab pointed out that Bangladesh’s relatively low digital literacy rate means users are likely to mistake AI-generated videos for authentic content. Bangladesh’s Election Commission has raised concerns about AI-generated misinformation twice, on December 12th and again on January 20th, just before the official election campaign began.
Experts cited by Dismislab warned that confirmation bias makes voters especially vulnerable during elections. People are more likely to trust and share content that aligns with their existing political beliefs, while negative propaganda targeting opponents tends to go unverified and spread quickly.
The investigation found that AI-generated characters were widely used to construct a narrative favorable to political parties. These characters included invented police officers, journalists, members of minority communities, children, and ordinary citizens.
AI videos supporting Jamaat-e-Islami feature false endorsements using “official figures” and ordinary people. In one video, a female officer claimed that the Jamaat was so popular in Bangladesh that foreign superpowers wanted it to come to power. Another alleged officer said the country would improve under Jamaat’s leadership. Dismislab confirmed that both videos were generated by AI.
Other videos featured AI characters claiming to belong to minority religious communities. In one article, a female character who identifies herself as Hindu says that Hindus will overwhelmingly vote for the Jamaat and that they will be safer if the party comes to power. She accused the leaders of “another political party” of being criminals and extortionists.
AI Content also praised Jamaat’s electoral alliances, including cooperation with the National Civic Party and the Liberal Democratic Party. One video described the Jamaat’s decision to cede seats to an alliance partner as a “second sacrifice”, while another claimed the party would contest alongside the 1971 Freedom Fighters and the 2024 “July Warriors”.
Another video presented as a street survey showed what appeared to be journalists asking seven people about their voting choices, with five strongly supporting Jamaat and two supporting BNP.
The AI-generated video promoting BNP relied heavily on child characters. In some clips, the AI-generated children wore clothes with a “bundle of rice” symbol on them.
In one video, a girl said BNP Chairman Tariq Rahman said, “This country belongs to the people” and described the “sheaves of rice” as a symbol of hope. Another child claimed that hearing Tariq Rahman’s name gave him hope for a better Bangladesh.
Dismislab also identified an AI-generated video using the digital portrait of Tariq Rahman’s daughter Zaima Rahman to deliver promotional messages and solicit votes.
One such video falsely claimed that Fatema Begum, a long-time friend and close aide of BNP chairperson Khaleda Zia, had legally become a member of the Zia family and was recognized as Tariq Rahman’s sister. Fact-checkers later debunked this claim, pointing out that Fatema Begum continued to work as Zaima Rahman’s companion even after Khaleda Zia’s death.
Both sides used AI content to discredit their rivals. Projamato’s videos frequently portrayed the BNP as an extortionist and fraudster, while simultaneously labeling the BNP’s proposed “family card” initiative as a deceptive inducement.
In one video, an AI-generated shopkeeper was seen claiming that BNP officials had eaten at his shop without paying and threatened to demand payment. The video ended with the shopkeeper pledging to vote for Jamaat.
Another video featured a Hindu woman wearing an AI-generated vermilion shakapora, urging people to confront BNP representatives with brooms as they handed out “family cards.” He said voters were becoming more aware and would support Jamaat instead.
Conversely, the AI-generated opposition campaign portrayed Jamaat as “anti-national.” In one podcast-style video, an AI character claimed that Jamaat was loyal to Pakistan, not Bangladesh. Another woman claimed that the Jamaat was trying to turn Bangladesh into Afghanistan, which failed to become Pakistan in 1971. One female AI character said she feared being stalked by Jamaat supporters “more than a pack of dogs.”
The AI misinformation also targeted individuals across party lines, including false claims linking BNP leader Mirza Abbas to the murder and “false” statements that Tasnim Jarrah, an independent candidate in Dhaka 9, had left the National Citizens’ Party to join the BNP.
Dismislab reports that the comments section of many AI-generated videos showed strong engagement, with users expressing their belief and support.
Global research supports these findings. Dismislab cited research by Feleniki Panagopoulou of Pantheon University that found that modern AI systems can understand human emotions and the nuances of language, allowing political actors to tailor messages to voters.
Bangladesh’s election law explicitly restricts the fraudulent use of AI. Section 16(b) prohibits the use of AI with fraudulent intent in election-related activities, and Section 16(g) prohibits false, misleading, or defamatory AI-generated content intended to mislead voters.
Md Ruhul Amin Mollick, Director (Public Relations) and Information Officer of the Election Commission, told Dismislab that “fraudulent intentions” mainly refer to spreading false information. He said criticism is permissible as long as it is not fabricated.
Supreme Court lawyer Jyotirmoy Barua argued that the law lacked clarity and that the mandatory AI exemption clause should have been explicitly mandated by the Election Commission.
Dismislab noted that other jurisdictions have adopted clearer rules. The Election Commission of India has made it mandatory for political ads to display prominent AI labels and occupy at least 10% of the visual frame. The European Union’s AI law also requires mandatory disclosure of synthetic political content.
Barrister Jyotirmoy said there was still a possibility that Bangladesh’s Election Commission would issue notifications to enforce disclosure requirements for the remaining elections.
methodology
Dismislab conducted daily monitoring and identified 21 Facebook pages and profiles that regularly posted AI-generated videos. All video posts from these pages from January 1 to 15, 2026 were collected and analyzed. AI-generated content was categorized by theme, cross-posted videos on YouTube and TikTok were tracked, and the presence of AI labels was systematically reviewed.
