AI-generated campaign videos circulate ahead of Bangladesh polls, most lack mandatory labels

Hundreds of artificial intelligence (AI)-generated political campaign videos, many without mandatory disclosure labels, are circulating widely on social media ahead of Bangladesh’s 13th Parliamentary Election on February 12, according to an extensive investigation by fact-checking organisation Dismislab.

On January 4, a Facebook video showed an elderly man speaking in a crowded marketplace. He said he would vote for the “sheaf of paddy”, the electoral symbol of the BNP, claiming the party would provide family, agriculture and health cards if elected. He added that such support would free him from daily manual labour in old age.

A day later, another Facebook video featured a woman in a rural setting praising the scale (daripalla), the electoral symbol of Bangladesh Jamaat-e-Islami. She described the symbol as representing justice, neutrality and equal rights, saying this was why she supported Jamaat.

A closer review, however, revealed noticeable inconsistencies. In both videos, the scene cuts every eight seconds before restarting from a new angle.

According to Dismislab, both clips were entirely AI-generated and were circulated as part of election campaigning without any disclosure.

Between January 1 and 15, Dismislab documented more than 800 AI-generated videos posted on Facebook, YouTube and TikTok. Some of these videos appeared across multiple platforms simultaneously.

On Facebook alone, 576 AI-generated videos were posted from 21 pages. Sixty percent of these videos carried no AI disclaimer. Even where labels existed, they were inconsistently displayed. Dismislab found that Facebook’s “AI info” label appeared only on the primary mobile app and disappeared when the same videos were viewed on Facebook Lite or desktop browsers.

The pattern was similar on other platforms. Seven of the Facebook pages reposted the same videos on YouTube, while two posted identical content on TikTok. Of the 181 videos analysed from seven YouTube channels, 94.48 percent had no AI label. None of the 50 TikTok videos carried any AI disclosure.

Dismislab noted that Bangladesh’s relatively low digital literacy rate increases the likelihood that users will mistake AI-generated videos for authentic content. The Election Commission of Bangladesh has raised concerns about AI-driven misinformation twice, on December 12 and again on January 20, shortly before official campaigning began.

Experts cited by Dismislab warned that confirmation bias makes voters particularly 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 spread quickly without verification.

The investigation found that AI-generated characters were widely used to build favourable narratives for political parties. These characters included fabricated police officers, journalists, minority community members, children and ordinary citizens.

False endorsement using “official figures” and commoners is prominent in the AI videos that support Jamaat-e-Islami. In one clip, a female officer claimed Jamaat was immensely popular in Bangladesh and that foreign superpowers wanted the party in power. Another purported officer said the country would improve under Jamaat’s leadership. Dismislab confirmed both videos were AI-generated.

Other videos featured AI characters claiming to belong to minority religious communities. In one, a female character identifying herself as Hindu said Hindus would vote overwhelmingly for Jamaat and would be safer if the party came to power. She accused leaders of “another party” of being criminals and extortionists.

AI content also praised Jamaat’s electoral alliances, including its cooperation with the National Citizen Party and the Liberal Democratic Party. One video described Jamaat’s decision to relinquish seats to alliance partners as a “second sacrifice”, while another claimed the party would contest alongside the freedom fighters of 1971 and the “July warriors” of 2024.

A separate video, presented as a street survey, showed a supposed journalist asking seven people about their voting choices, with five strongly favouring Jamaat and two supporting the BNP.

AI-generated videos promoting the BNP relied heavily on child characters. In several clips, AI-generated children wore clothing featuring the “sheaf of paddy” symbol.

In one video, a girl said BNP Chairman Tarique Rahman had said that the country belonged to the people, describing the “sheaf of paddy” as a symbol of hope. Another child claimed that hearing Tarique Rahman’s name inspired hope for a better Bangladesh.

Dismislab also identified AI-generated videos using the digital likeness of Zaima Rahman, daughter of Tarique Rahman, to deliver promotional messages and seek votes.

One such video falsely claimed that Fatema Begum, longtime companion and aide of BNP chairperson Khaleda Zia, had been legally inducted into the Zia family and recognised as Tarique Rahman’s sister. Fact-checkers later debunked the claim, noting that after Khaleda Zia’s death, Fatema Begum continues to serve as Zaima Rahman’s companion.

Both camps used AI content to discredit rivals. Pro-Jamaat videos frequently portrayed the BNP as extortionists and fraudsters, while labelling the BNP’s proposed “Family Card” initiative as a deceptive inducement.

One video showed an AI-generated shopkeeper claiming BNP-affiliated individuals ate at his business without paying and threatened him when he demanded payment. The video ended with the shopkeeper pledging his vote to Jamaat.

Another video featured an AI-generated Hindu woman wearing vermilion and shakha-pola, urging people to confront BNP representatives offering “Family Cards” with brooms. She said voters had become more aware and would support Jamaat instead.

Conversely, AI-generated counter-campaigns portrayed Jamaat as “anti-state”. In one podcast-style video, an AI character claimed Jamaat was loyal to Pakistan rather than Bangladesh. Another alleged that Jamaat sought to turn Bangladesh into Afghanistan after failing to make it Pakistan in 1971. One female AI character said she feared being followed by a Jamaat supporter “more than a bunch of dogs”.

AI-driven misinformation also targeted individuals across parties, including false claims linking BNP leader Mirza Abbas to a murder case and “incorrectly” saying that Tasnim Jara, independent candidate of Dhaka-9, had left the National Citizen Party to join the BNP.

Dismislab reported that comment sections under many AI-generated videos showed strong engagement, with users expressing belief and support.

Global research supports these findings. Dismislab cited a study by Fereniki Panagopoulou of Panteion University, which found that modern AI systems can understand human emotions and language nuances, enabling political actors to micro-target voters with tailored messages.

Bangladesh’s electoral code explicitly restricts the dishonest use of AI. Subsection 16(b) prohibits AI use with dishonest intent in election-related activities, while subsection 16(g) bans 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 “dishonest intent” primarily refers to spreading false information. He said criticism is permitted as long as it is not fabricated.

Supreme Court lawyer Barrister Jyotirmoy Barua said the law lacks clarity and argued that mandatory AI disclaimers should have been explicitly required by the Election Commission.

Dismislab noted that other jurisdictions have adopted clearer rules. India’s Election Commission has mandated prominent AI labels in political ads, occupying at least 10 percent of the visual frame. The European Union’s AI Act also requires mandatory disclosure of synthetic political content.

Barrister Jyotirmoy said Bangladesh’s Election Commission could still issue a notification enforcing disclosure requirements for the remainder of the campaign.

Methodology

Dismislab conducted daily monitoring to identify 21 Facebook pages and profiles that regularly post AI-generated videos. All video posts from these pages between January 1 and 15, 2026 were collected and analysed. AI-generated content was categorised by theme, cross-posted videos were tracked on YouTube and TikTok, and the presence of AI labels was systematically reviewed.