Unlocking Insights from Millions of Voter Records

One of the most ambitious undertakings involved analyzing nearly 22 million voter records across the Indian states of Bihar, Tamil Nadu, and West Bengal. The data, released by the Election Commission as image-based PDFs in Hindi, presented a significant challenge. Processing approximately 90,000 files in Bihar alone – covering 6.5 million records – required a novel approach.

Instead of manual data entry, The Hindu employed optical character recognition (OCR) to convert the images into machine-readable text, followed by translation into English. Crucially, large language models were then utilized to generate SQL queries from natural language prompts, eliminating the need for reporters to write complex database commands. This streamlined process allowed for rapid analysis of the vast dataset.

The analysis revealed concerning patterns. In Bihar, a disproportionate number of women were removed from voter rolls despite higher rates of male out-migration. Furthermore, a significant percentage of deleted voters across multiple polling booths were incorrectly marked as deceased, many under the age of 50. These findings prompted further investigation and ultimately led to scrutiny from the Supreme Court of India, which directed the Election Commission to release complete deletion records.

The Hindu responded by creating a searchable database of deleted names and reasons, publishing state-level investigations that sparked public debate and, in some cases, corrections to the voter rolls. As a senior editor noted, “These were not conclusions drawn by AI. The hypothesis was ours. The political and social context was ours. AI helped us process the scale.”

<h2>Building Interactive Election Maps Without Code</h2>
<p>The application of AI extended beyond document processing to the creation of interactive election maps for the 2019 and 2024 Indian general elections. These maps allowed users to filter results by region, state, rural-urban classification, and urban clusters. Remarkably, the application was built entirely using prompts in ChatGPT, Gemini, and Claude – without a single line of code written manually.</p>
<p>The team broke down the interface into its core components – filters, maps, and list views – and used the LLMs to generate annotated code for each. This approach not only accelerated development but also facilitated verification.  Previously, such projects would have required dedicated in-house engineers or external volunteers.  Now, the team could bypass these bottlenecks, adhering to critical journalistic deadlines.</p>
<p>“Deadlines are sacrosanct in journalism,” explained a lead editor. “Now we don’t have to extend them because we’re waiting for technical help.”</p>
<img src="https://cdn.wan-ifra.org/wp-content/uploads/2026/03/05080739/Screenshot-2026-03-05-at-1.37.31-PM-1024x586.png" alt="Interactive Election Map" width="600" height="343">

<h2>Measuring Heat Stress at Street Level</h2>
<p>Innovation wasn’t confined to digital projects. In Chennai, where extreme summer temperatures pose a significant health risk, <em>The Hindu</em> investigated how heat stress varies across different occupations. Utilizing AI-assisted guidance, the newsroom assembled low-cost, Arduino-based devices to record temperature and humidity every 10 seconds. The entire project cost between ₹15,000-₹20,000 (approximately $180-$240), demonstrating the potential for impactful reporting with limited resources.</p>
<p>Data was collected from a cook, a fisherman, an industrial worker, and an autorickshaw driver over a 24-hour period. The resulting heat index measurements, combining temperature and humidity, revealed significant disparities in exposure, peaking at 69°C (156.2 F) in one instance.  The data was visualized to highlight these differences.</p>
<img src="https://cdn.wan-ifra.org/wp-content/uploads/2026/03/05080853/Screenshot-2026-03-05-at-1.38.36-PM-1024x475.png" alt="Heat Stress Measurement Device" width="600" height="279">
<p>Following publication, the Tamil Nadu government announced a heat management plan and initiated further studies using similar devices. This project exemplified the power of combining hardware experimentation with data storytelling, with AI playing a crucial role in design and troubleshooting.</p>

<div style="background-color:#fffbe6; border-left:5px solid #ffc107; padding:15px; margin:20px 0;"><strong>Pro Tip:</strong> When using LLMs for code generation, always thoroughly test and verify the output.  AI-generated code can contain errors or inefficiencies that require human oversight.</div>

<h2>The Human Element: AI as a Powerful Tool, Not a Replacement</h2>
<p>The integration of AI into <em>The Hindu’s</em> data journalism pipeline – encompassing hypothesis formation, data collection, cleaning, analysis, visualization, and publication – is not about replacing journalists, but augmenting their capabilities. The team categorizes its work into five types: simple trend analysis, correlation studies, factor analysis, causal investigations, and deep-dive accountability reporting.</p>
<p>AI now assists with web scraping, unstructured document processing, database query suggestions, and front-end interface building. However, human oversight remains paramount.  One instance highlighted the need for human insight when an AI-generated script processed documents sequentially, slowing down analysis. A technologist suggested multi-threading, which, when prompted to the model, resulted in a significantly more efficient version.</p>
<p>“You need human insight to tell it what to optimize,” emphasized a senior editor.  The newsroom also avoids using AI to draw editorial conclusions, recognizing that the risk of “hallucinations” is lower in structured tasks where outputs can be directly tested.</p>
<p>Over the past decade, data journalism at <em>The Hindu</em> has evolved from visual enhancements to traditional reporting to a dedicated function with data journalists, designers, and editorial coders. A notable project involved an analysis of excess deaths during the COVID-19 pandemic, which estimated that official death counts were underreported by a factor of five to six. While initially contested, subsequent analyses by the World Health Organization and official data revisions validated the newsroom’s findings.</p>
<p>Today, data-driven reporting is fully integrated across all operations, driving higher subscriptions and engagement.  As one editor stated, “We want a more informed audience. This kind of work helps us move in that direction. Across projects, AI does not replace journalistic judgement. It expands the scale at which it can operate.”</p>
<p>What ethical considerations should news organizations prioritize when implementing AI-powered journalism tools?  How can we ensure that AI enhances, rather than diminishes, the quality and trustworthiness of news reporting?</p>

Frequently Asked Questions About AI in Journalism