As Assam heads into polling on April 9, the election is being fought as much on screens as on the ground.
A voter scrolls past a video of Congress leader Gaurav Gogoi reciting Pakistan's national anthem on Republic Day. It never happened. Another clip places him alongside Pakistan Army Chief General Asim Munir, framing him as a foreign-backed operative. Elsewhere, reels warn that Assamese identity is under threat, casting Bengali-origin Muslims as "infiltrators". In between, an AI-generated young Assamese woman appears in soft-lit, vlog-style videos, blending cultural pride with political messaging.
None of it is real. But all of it is shaping what voters see.
A new report by the Netherlands-based Foundation Diaspora in Action for Human Rights and Democracy (DAHRD) has found widespread circulation of AI-generated content ahead of the Assam elections. Between November 2025 and early April 2026, at least 432 AI-generated posts were shared across Facebook and Instagram. These posts collectively drew over 45 million views and more than 100,000 likes.
The report describes what it calls a "six-tier content factory"—a coordinated pipeline of 273 accounts with a combined reach of over 400 million, designed to push political narratives at scale. At the top sit pages linked to leaders like Assam Chief Minister Himanta Biswa Sarma, which feed into party pages, friendly media outlets and local networks before content moves into WhatsApp groups.
At the bottom layer, high-volume AI accounts generate large amounts of propaganda, allowing a single narrative to be repeated, amplified and made unavoidable for anyone online.
A significant share of the content targeted Congress leader Gaurav Gogoi.
An election engineered online
The report analysed over 2,400 posts across Facebook, Instagram and X between November 2025 and early April 2026.Of these, at least 432 were likely AI-generated, together crossing 45 million views, the report found. Their formats are consistent and optimised for reels:
- Deepfake-style clips showing Gaurav Gogoi in fabricated scenarios—singing Pakistan’s anthem or appearing with General Asim Munir. These mimic news edits, with captions, background scores, and stitched footage.
- Meme reels and slideshows, often under 30 seconds, with text like “Assam under threat” or “demographic invasion”, paired with stock or unrelated visuals.
- AI “talking head” videos, where synthetic faces—often young Assamese women—speak directly to the camera in Assamese or Hindi.
Most of these posts carried no AI disclosure, despite visible artefacts—unnatural lip-sync, overly smooth faces, inconsistent lighting.
What stands out is the scale and the speed of distribution. The same videos appear across multiple accounts within hours, which the report cites as evidence of coordination rather than organic spread.
The network behind the reels
Behind the content is a layered network of accounts, ranging from official political handles to anonymous high-output creators.
At the top sit pages linked to the Bharatiya Janata Party and the Assam Chief Minister. Speeches by Sarma on "illegal immigrants" and Assamese identity were clipped and circulated widely, the report found, often becoming the raw material for downstream content. Rhetoric on identifying "illegal immigrants" was reframed into reels claiming the imminent deletion of Muslim voters—a shift from policy talk to fear messaging.
A second layer of supporter and ideological pages—with names invoking Assamese identity and nationalism—repackaged this material into reels with captions like "Save Assam" or "Protect Indigenous Rights".
The third layer is the most opaque with high-volume AI content creators.
The Instagram account Politooons (@politooons) stood out in the report's findings. It generated 40.2 million views—88% of all AI content tracked—across 102 posts, and racked up 4.57 million likes. Its polished AI-generated reels, in Assamese and Hindi, pushed BJP-aligned themes including anti-infiltration narratives and the projection of Sarma as a strong leader.
The report describes Politooons not as an individual creator but as a coordinated content operation.
Another account, Axomiyafiles, contributed over 161,000 views through AI-generated "Assamese girl" avatars that speak directly to the camera in Assamese or Hindi, blending cultural pride with political cues.
Other accounts—NewsNexus Assam, Assam Insight, Voice of Axom and Truth Axom—drove high engagement through AI videos, bulldozer visuals and eviction footage, collectively pulling in tens of millions of views.
A coordinated deepfake campaign targeting Gogoi also ran across accounts including iambjpofficial, djrfiles and BJP4Assam.
Many of these accounts show no clear affiliation—no ownership details, no political disclosure, and minimal bios.
“Propaganda-to-policy pipeline”
The report describes the broader ecosystem as a "propaganda-to-policy pipeline", where online narratives don't just shape opinion but feed into real-world action.
Much of the content targets Assam's Bengali-origin Muslim population, often pejoratively labelled "Miya".
According to the report, the loop runs in four stages. It begins with a narrative—AI-generated posts, amplified by statements from Sarma, frame Muslims as a demographic threat. The language then mirrors policy—claims about deleting "Miya votes" are followed by voter roll revisions and constituency changes that reduce representation. Next comes law: property restrictions and exclusionary measures. Then enforcement: posts celebrating evictions and displacement.
"The loop is complete: narrative, amplification, policy, enforcement — each feeding the other," the report notes. For Bengali-origin Muslims at the receiving end, this is not just online discourse but something that shapes everyday realities, from documentation to housing.
The enforcement gap
On paper, the Election Commission of India had a framework in place — the Model Code of Conduct, ad pre-certifications and coordination with platforms. In practice, the report found, enforcement was missing.
It documents 119 violations, including 84 high-severity cases and 15 linked to Sarma's personal handle. None of these triggered a public takedown, flag or penalty, and all the posts remained accessible weeks later.
The report flags timing as part of the problem. Oversight kicked in after the first wave had already spread, and digital campaigning moved much faster than regulation.
Platforms showed a similar pattern. Meta and X did not act on flagged content.
Sixty-four AI-generated posts on Facebook and 108 on Instagram remained unlabelled, while 31 deepfake videos targeting Gogoi stayed online without any warnings.
Even long-running campaigns went untouched. A three-year "pushback" operation of 68 posts celebrating evictions saw zero takedowns.
The only exception was a viral AI video showing Sarma firing at two Muslim men, with captions like "No Mercy" and "Foreigner Free Assam". Even there, the report notes, the video was likely taken down by the account itself, not the platform.
The report frames this as a shift from inconsistent moderation to near-total inaction, leaving voters with few reliable signals to distinguish what is manipulated, misleading or false.