Indian publisher NDTV is investing in AI tools that enhance newsroom workflows and audience products. The company prioritizes building unique software and data layers over relying solely on generic AI models. Editorial teams remain central to the process.
NDTV is taking a distinctive approach to artificial intelligence by prioritizing proprietary software, workflows, and data over simply deploying off-the-shelf AI models. For publishers and media professionals, this strategy highlights the importance of building value-added layers that are difficult for competitors to replicate, rather than relying on technology that is widely accessible.
Rohan Tyagi, Chief Product Officer at NDTV, explained that while creating AI products has become easier, sustaining their value is increasingly challenging. He said the real opportunity lies in what publishers build around AI models-such as custom software, editorial workflows, and unique datasets-rather than the models themselves.
NDTV uses an internal framework to evaluate AI initiatives, dividing them into "thin" and "thick" layers. Thin layers are quick-to-build wrappers around existing models, useful for rapid prototyping and workflow improvements, but often risk producing tools that do not address real newsroom needs. Thick layers, by contrast, require more investment and combine AI with proprietary assets, offering greater long-term value but demanding more resources and time.
Newsroom Tools and Workflow
NDTV has developed several internal tools to streamline editorial processes. One example is Echion, an infographic generator that transforms articles into visual graphics optimized for news and multiple languages. The tool leverages standard image models but adds custom templates, prompts, and branding guidelines, allowing teams to reuse designs and maintain consistency across brands and campaigns. This approach also helps address recurring issues, such as AI-generated errors in maps, by embedding corrective instructions into templates.
Another tool, Liza, acts as an analytics agent for editorial, product, social, and revenue teams. Instead of building another dashboard, NDTV integrated Liza with platforms like Search Console, Google Analytics, Chartbeat, and YouTube. Teams interact with Liza through Discord to ask questions, generate reports, and schedule updates, making analytics more accessible and actionable across departments.
Audience Products and User Experience
NDTV is also experimenting with AI-driven consumer experiences. Within its app, the publisher is testing AI-generated news feeds alongside editorially curated ones. One feature creates AI cards that combine article summaries, short videos, and GIFs. Editorial teams review and grade these summaries in the CMS, refining the output after initial inconsistencies. Tyagi noted that this grading system allows journalists to provide feedback, improving future AI outputs without requiring manual approval for every piece.
The company is piloting recommendation systems based on large language model (LLM) embeddings, aiming for more context-aware suggestions than traditional machine learning. Another experiment clusters stories around entities, offering new ways for users to navigate news content. Early results indicate these AI-driven features are helping attract new users, though NDTV is still evaluating their performance against more editorially controlled feeds.
NDTV has also launched askNDTV, an answer engine built on its news archive. Unlike standard search, this tool is designed to answer user questions using NDTV’s journalism and structured datasets, grounding responses in trusted sources. The publisher is further exploring AI-generated audio, discovering that audio scripts require different editorial standards than text. Editorial teams grade scripts before they are used in an audio feed featuring cloned anchor voices and AI recommendations.
Technical Investments and Editorial Oversight
Building these AI products has revealed the limitations of relying solely on external AI services. Running vector search and LLM embeddings across decades of articles created cost and performance challenges. To address this, NDTV invested in research to improve retrieval efficiency, developing a new database structure for LLM retrieval. This work was recognized with acceptance at an information retrieval conference, demonstrating the value of deeper technical investment.
Tyagi emphasized that not all AI workflows require the same level of editorial review. Analytics tools working with structured data pose less risk of hallucination, while AI-generated summaries and infographics benefit from closer editorial involvement. NDTV is developing grading interfaces for journalists to review and rate AI outputs, using feedback to refine future results rather than relying on a simple approval process.
For publishers seeking to differentiate their AI offerings, Tyagi’s experience suggests that the real advantage comes from what is built around the models-unique software, workflows, and data-rather than the models themselves. This approach echoes themes explored in recent coverage of AI’s impact on editorial workflows, where the focus is on practical newsroom integration and long-term value.
