Unlabeled AI-generated videos are widely circulating on social media as part of election campaigns in Bangladesh, raising concerns about platform enforcement and voter awareness, according to Dismislab, an independent fact-checking platform.
A Facebook video posted on January 4 shows an elderly man in a busy market saying he will vote for a bundle of paddy, the election symbol of the Bangladesh Nationalist Party (BNP), because of the promise of family, agriculture and a health card.
Another video posted the next day shows a woman standing in the countryside praising Jamaat-e-Islami’s election symbol, the scales, as a symbol of justice and fairness.
According to Dismislab, upon closer inspection, both videos were found to have inconsistencies. Each clip includes frequent scene cuts, with a new scene appearing approximately every 8 seconds.
Dismislab reports that both videos were generated by AI and are being circulated on social media as part of a political campaign ahead of the 13th parliamentary elections scheduled for February 12th.
Although the use of AI in political campaigns is not prohibited, none of the videos include an AI label, which violates Meta’s transparency policy.
Dismislab documented more than 800 AI-generated videos (including duplicate counts) related to the election campaign across Facebook, YouTube, and TikTok between January 1 and 15. We found that 60% of AI-related content on Facebook lacks an AI disclaimer.
The situation was similar on other platforms. Of the 21 Facebook pages identified, seven posted the same video on YouTube and two on TikTok.
After reviewing 181 videos from 7 YouTube channels, 94.48% did not have an AI label. According to Dismislab, none of the 50 videos posted to TikTok contained an AI disclaimer.
Facebook posted 576 AI-generated videos from 21 pages, but Dismislab found that the “AI information” disclaimer only appeared on the main Facebook app. If you play the same video on Facebook Lite or your computer, the label won’t appear.
Dismislab noted that in Bangladesh, where digital literacy remains low, many users misinterpret AI-generated content as genuine information.
It also reported that the Bangladesh Election Commission has expressed concerns about the proliferation of AI content on at least two occasions. The first time was on December 12th, and the second on January 20th, two days before the official election began.
Dismislab said the findings, combined with global research, show that AI-generated content can influence voters’ opinions and create an unequal electoral environment, especially if such content is circulated without disclosure.
Experts cited by Dismislab warned that voters are particularly vulnerable to AI-generated content because they tend to trust information that aligns with their existing political beliefs.
This trend is exacerbated during elections, when such content is more likely to be accepted and shared without verification, especially if it contains negative claims about political opponents.
According to Dismislab, the AI-generated videos feature a wide range of fabricated characters, including alleged police officers, politicians, and ordinary citizens, appearing to support political parties and promote favorable narratives.
The analysis found that content supporting Jamaat-e-Islam frequently used false endorsements from “government officials” and ordinary people, while some videos linked to the BNP used AI-generated children to solicit votes.
Both sides were also found to be using AI-generated videos to discredit their opponents.
Pro-Jamaat content frequently depicts the BNP as a usurper or deceiver, while anti-Jamaat-targeted campaigns use similar tactics to portray the party as “anti-national,” Dimislav reported.
