Poll by prompt: Strategists look to AI as distillers of public opinion

In a small house in Rajasthan’s Samod village, Junaid Khan (name changed on request), along with his two companions, is attempting to initiate a digital revolution in the political strategy space. They employ artificial intelligence (AI) and generative AI (GenAI) tools to map relevant issues in the upcoming panchayat-level elections.

The three men have all previously worked with the Indian Political Action Committee (I-PAC) during the 2015 Bihar and 2017 Uttar Pradesh elections, and have already applied their experience to the gram panchayat elections in Samod, Tetara, Moondli, and other areas.

“We use AI to expedite our data analysis. We gather demographic data, such as gender ratios, education levels, employment rates, and access to basic amenities among families in each gram panchayat. The AI tools then compare this data with information from past elections, encompassing socio-cultural factors that have proven decisive in each election. Ultimately, this aids us in delineating the response and sensitivity that registered voters exhibit towards these issues by region and demographic categories,” explains Khan.

He claims that this approach has facilitated the creation of more nuanced campaigns, reduced campaign costs, and enabled more accurate predictions of vote shares.

Khan also asserts that they are the first to employ AI tools like ChatGPT, DALL-E, Scribe AI, etc., for data visualisation, secondary research, comparative analysis, and data parsing on public opinion.

Regrettably, he is mistaken.

Divya Mathews and her team of prompt engineers, based in Kochi, specialise in designing and feeding precise prompts into GenAI engines to generate tailored results for their clients. In the past year alone, they have provided their services to no fewer than five strategy startups across Tamil Nadu, Karnataka, and Kerala. With the demand for prompt engineers rapidly escalating in the realm of political strategy, Mathews reveals that her team’s projects extend into March 2025.

Her team has devised questionnaires based on data from preceding panchayat elections, identified relevant social media platforms to expand client reach, and even offered suggestions for structuring campaign narratives around the key issues identified by their AI models.

Veteran election strategist Abbin Theepura, the founder of P-MARQ (Politique Marquer), however, points out that poll strategy organisations have been relying on AI for predicting election outcomes, data analysis, and interpretation for the past few years.

“We boast a team of expert data scientists who developed our AI model. Ground-level data is fed into the AI model, wherein various parameters, including caste composition and gender ratios, are computed,” he states.

The AI model produces projections for vote share and seat numbers based on the data.

The team of data scientists then engages in a comparative analysis of different scenarios, considering data from earlier elections, to determine how a shift in voting percentage might impact election outcomes, Theepura elaborates.

“If there’s a 0.5 per cent or 1 per cent shift in the voting pattern or projected vote share based on the data, the AI model predicts the updated outcome and the corresponding alteration in projected seat numbers,” he adds.

Nonetheless, there has been a marked surge in the adoption of AI models for crafting poll strategies in the lead-up to the 2024 general elections.

Anu Krishna, head of international business at Delhi-based Inductus, explains that campaign strategies rely on multiple tiers of data collection.

“There’s a central nodal office overseeing resource planning and strategy. Then we have state-specific campaign offices focused on resource mobilisation and coordination at the district level. These campaign offices coordinate with their own volunteers, who, in turn, work with party workers,” explains Krishna.

However, organisations are now increasingly outsourcing vital aspects of data collection at this level to smaller entities, such as Khan’s team, which may have already gathered data at hyperlocal tiers. This shift has prompted several localised strategists to concentrate on specialised tasks like linking clients with local media influencers, managing media narratives in remote rural areas, and devising questionnaires or surveys.

For example, while Khan and his associates have narrowed their focus to a smaller geographical area, several of the firms that Mathews’ team collaborated with chose to concentrate solely on researching keywords used in past speeches and election manifestos.

Such diversification of the poll strategy ecosystem helps address a significant challenge in employing AI tools in the realm of political consultation.

“The margin of error decreases as more data becomes available,” observes Theepura.

With smaller strategy organisations collecting and storing extensive data at the hyperlocal level, strategists operating at the state and national levels can now tap into these more detailed databases and integrate them into their AI models.

Cost remains a major impediment to the wider adoption of AI.

While more established organisations develop their own AI models to handle a broader spectrum of tasks, building such models is financially beyond the means of smaller operations like Khan’s three-member team.

Trailblazers in the field like I-PAC are closely monitoring the evolution of this ‘feeder ecosystem’.

“AI’s role in designing, implementing, and monitoring community-specific initiatives will be a critical narrative to track,” states Arindam Malakar, engagement manager, I-PAC.

Mathews posits that if the feeder ecosystem expands to encompass AI engineers and prompt designers, the expense of constructing dedicated AI models can also be outsourced.

“It’s all about establishing a truly interconnected ecosystem,” she concludes.