The first thing to understand about AI in global elections is how recently it became a real factor. As late as 2020, "AI in campaigns" mostly meant propensity modelling — using machine learning to score voters by likelihood of turnout or persuasion. Useful but invisible. The voter never knew an AI was involved.
By 2024, voters across four continents were having actual conversations with AI agents on the phone, watching AI-generated videos of candidates, reading AI-personalised letters, and arguing with AI-detected deepfakes of opponents. In a single electoral cycle, AI moved from back-office analytics to the front of the campaign.
This guide is the global map, condensed — what happened, what worked, what backfired, and the four patterns India should copy before the 2027 state elections begin in earnest.
The four-era timeline
To make sense of where things are, here are the four eras of AI in elections.
Era 1 — Analytical AI (2008–2018). Obama 2008 and 2012, Cambridge Analytica's work for various clients, the Modi 2014 micro-targeting playbook. Machine learning runs on the back end: which voters to target, which message to test, which neighbourhoods to door-knock. Voters never interact with the AI.
Era 2 — Generative content (2018–2022). Deepfakes emerge, mostly as gimmicks. South Korea's 2022 race introduces AI avatars of candidates. China's local elections see AI-generated images in propaganda materials. The technology is real but not yet at consumer scale.
Era 3 — Mass synthetic media (2023–2024). Argentina's 2023 presidential cycle is the watershed. Both Sergio Massa and Javier Milei deploy AI-generated images of their opponents in dystopian scenarios. Indonesia's 2024 election sees AI revival of Suharto delivering campaign endorsements. India's 2024 cycle features AI-generated videos of deceased Tamil Nadu CMs endorsing current candidates.
Era 4 — Conversational AI (2024–present). The shift from generated content to generated conversation. Voice agents that talk in 30 languages. WhatsApp chatbots that triage grievances. Two-way real-time engagement at lakh scale. This is the era India is now entering — and where the long-term transformation lies.
United States 2024: the AI cycle that wasn't
The US 2024 election was widely predicted to be "the AI election". The reality was more measured.
What happened. Both major campaigns built propensity models, used AI-generated video ads (clearly labelled as such), and ran experimental AI chatbots on their websites. Several state-level Republican primaries saw AI-generated robocalls — most famously a fake Biden call in New Hampshire telling voters to stay home, which produced an FCC ruling banning AI voices in robocalls without disclosure.
What worked. AI-driven fundraising emails — personalised by name, location, prior donation history, current news cycle — measurably outperformed templated emails on per-email yield. Estimates put the gain at 15–30%.
What did not work. Mass-deployed AI conversational agents on the voter-facing side. The legal landscape (state-by-state robocall rules, FCC's 2024 ruling) made the kind of voice campaigns that work in Indonesia and India impractical at scale.
India lesson. The US regulatory framework actively discourages conversational AI for voters. India's framework — ECI advisory + TRAI TCCCPR + DPDP — is more permissive in this dimension if AI is disclosed. This is one area where India will move faster than the US in 2026–2029.
Argentina 2023: AI as visual warfare
Javier Milei's campaign used AI-generated images aggressively — depicting opponent Sergio Massa in dystopian scenarios (China-style propaganda posters, dictator-uniform compositions, gulag imagery). Massa's team responded with AI-generated images of Milei as a Joker-like chaos figure.
What happened. Both campaigns labelled the images as AI-generated (Argentina has no formal disclosure rule yet, but both campaigns chose to). The images dominated the visual layer of the campaign — appearing on TV news, news websites, every social feed.
What worked. Speed. AI-generated visuals could respond to news cycles within hours. A traditional campaign with a graphic designer team takes 24–48 hours to produce a single high-impact image. AI cut this to under 30 minutes.
What backfired. Both campaigns lost trust on substantive issues. Voters increasingly assumed any image they saw might be AI-generated, including legitimate documentary photos. The signal-to-noise ratio of visual political communication collapsed.
India lesson. The visual AI arms race is coming to India. The 2024 cycle saw early shots (Tamil Nadu, Maharashtra). The 2027 state cycle will be the first where AI-generated visual content is a major channel. Campaigns need a deepfake-detection capability — not just to attack opponents but to defend their own candidates from AI-generated smears.
Indonesia 2024: the vernacular voice playbook
Indonesia's February 2024 election was the largest single-day electoral exercise in human history at the time (subsequently surpassed by India's 2024 cycle). Prabowo Subianto's campaign used Bahasa-native voice AI extensively — outbound calls to undecided voters, GOTV reminders, opinion sampling across the archipelago's many islands.
What worked. Reaching voters in geographically scattered constituencies (Indonesia has 17,000 islands) where physical canvassing is impractical. The voice agents could call voters in Bahasa Java, Bahasa Sunda, regional dialects, and capture sentiment that no traditional polling could surface.
What also worked. AI-generated videos resurrecting Suharto-era imagery — Prabowo positioned as continuity. Highly controversial but measurably effective in older demographics.
India lesson. Indonesia's structure (large archipelago, multi-lingual, multi-religious, fragmented electorate) is closer to India's than the US or UK structures. The voice-first playbook that worked in Indonesia maps directly to states like Tamil Nadu (rural fragmentation), Kerala (Malayalam + caste fragmentation), and Rajasthan (Marwari/Mewari dialect spread). Indian campaigns can confidently borrow from this playbook.
United Kingdom 2024: the defensive cycle
The UK's July 2024 election was striking for what did not happen. Pre-election forecasts warned of widespread deepfakes and AI-driven disinformation. The actual cycle saw minimal AI-generated political content.
Why. Three factors. First, the UK's Online Safety Act 2023 placed legal liability on platforms for AI-generated political content that misled voters. Second, all major parties pre-committed to not using deepfakes. Third, the cycle was short (5–6 weeks) and the major parties were stable in their narratives.
What did happen. AI was used heavily in the back office — for canvassing models, voter persuasion experiments, social media ad personalisation. Voters mostly did not see AI.
India lesson. A short campaign window (5–6 weeks) reduces AI's effectiveness because the technology needs time to be deployed at scale. India's Lok Sabha campaigns run 60–90 days; state campaigns run 30–45 days. There is enough runway for AI to matter, but campaigns that wait until the formal notification have already lost the AI window. Start 90 days before notification, not 30.
The four patterns India should copy
Across all four geographies, four patterns emerge that map directly to Indian context.
1. Voice-first for low-bandwidth voters
Indonesia's lesson. India's rural electorate (~70% of total voters) lives in low-bandwidth, feature-phone-dominant communication environments. Voice AI is the single channel that reaches them at acceptable engagement rates. Every other channel (WhatsApp, SMS, video) under-performs in this segment.
2. Vernacular at the dialect level
Indonesia again. India has 22 official languages and ~270 dialects spoken by >10,000 people each. Conversational AI that handles only Standard Hindi misses the actual conversation. The agent has to recognise the dialect in the first 5 seconds and switch register.
3. Disclosure and trust
Argentina's negative lesson, UK's positive lesson. Disclosed AI builds trust over time; undisclosed AI destroys it in one bad headline. The ECI's 2024 advisory got this exactly right. Indian campaigns that label AI content clearly will outperform campaigns that try to hide it — even before the regulatory enforcement catches up.
4. Defensive deepfake detection
US lesson (post-New Hampshire fake-Biden call). Every serious campaign in India 2026+ needs a 24×7 deepfake detection capability — for fake content about its own candidate. The technology to generate convincing fakes is now under $20 and 10 minutes. The technology to detect them, on a tight latency budget, requires infrastructure most state campaigns don't have. This is where to invest.
What India does that no one else does
A few things genuinely make the Indian context unique.
Bhashini. The MeitY-funded national language stack provides ASR, MT and TTS for all 22 scheduled languages, hosted in India, with permissive licensing for political use. No other country has a comparable national infrastructure.
Voter-ID-as-canonical-ID. EPIC numbers are unique, persistent and tied to the electoral roll. This makes voter outreach lists cleaner than any global counterpart — provided the campaign respects the ECI rules on usage.
Five-phase election cycles. Most countries run a single election event. India's general + state + local cycles produce a continuous five-year engagement opportunity. AI infrastructure built for one phase can be re-purposed for all five — a structural advantage Indonesian, US, UK and Argentine campaigns do not have.
Scale forces innovation. India is the only country where a single constituency is large enough to justify enterprise-scale AI infrastructure. The economic incentives for Indian-language voice AI are stronger than anywhere else in the world. Expect Indian platforms (AiSewak, Voxdonna, Haptik, Saaras, Sarvam) to lead the next wave of vernacular voice AI globally — not just inside India.
What this means for 2027 and 2029
The 2027 state elections — UP, Punjab, Goa, Manipur, Uttarakhand — are the next testing ground. Campaigns that have already started preparation in May 2026 will have:
- 6 months to build and tune their voice AI stack
- 9 months to assemble and clean voter lists
- 12 months to learn what works in their state's specific dialect map
- A full year of governance-phase AI deployment leading into the campaign
Campaigns that wait until late 2026 will be playing catch-up against a moving frontier.
The 2029 Lok Sabha cycle will be the world's first AI-native general election — every major campaign will have a conversational AI layer, every voter touchpoint will have an AI option, every grievance will be triaged by an AI before reaching a human. The global eyes will be on India for the same reason they were on Indonesia in 2024: scale.
Where to go next
- Voice AI in Political Campaigns: Technical Guide — the stack underneath
- AI vs Traditional Election Outreach — head-to-head ROI
- AI Helped Win Indian Elections: 2024 Case Studies — India-specific patterns from the most recent cycle
- Lok Sabha 2029 Voice-AI Playbook — what the next general election looks like
Most of the global commentary on AI in elections is about the spectacle — the deepfakes, the manipulated images, the legal panic. The actual transformation is quieter and far larger. AI is reshaping how candidates listen to voters, how parties allocate field resources, how manifestos are written, and how governance continues between elections. Pay attention to the boring AI. That is the one that will decide India's next decade.