At The Quint, AI is helping readers navigate long-form journalism
Rather than reworking the journalism itself, the newsroom built an AI-powered layer designed to meet different reader needs – offering summaries, key takeaways, and interactive explanations without interrupting the core article.
Founded in 2014, The Quint is a digital-first news platform in India focused on mobile audiences, explanatory journalism, and fact-checking through its WebQoof initiative. Its coverage spans politics, policy, gender, and social issues, often delivered in multimedia, mobile-friendly formats.
Like many publishers, it has had to adapt to shifting audience behaviours shaped by social platforms, search, and increasingly, AI-driven consumption.
To get support with that, The Quint joined the 2025 edition of Newsroom AI Catalyst, a WAN-IFRA accelerator programme in partnership with OpenAI that supports news publishers with their strategic AI initiatives.
The problem: high intent, low engagement
The starting point for what became NewsEasy was a mismatch between audience interest and behaviour. While long-form exclusives and deep dives were attracting page views, they were not holding attention as much as the brand would like it to for the different cohorts of users landing on its articles.
“While page views were increasing, there was an opportunity to increase the average time spent on any article even further,” said Tarun Jain, Product Head at The Quint. “Scroll depth data reinforced the pattern: presenting a clear opportunity to deepen engagement with the content.”
The NewsEasy team (L-R): Prateek Nair (Assistant Editor, Audience Engagement & Marketing), Abhilash Mallick (Editor, WebQoof), Tarun Jain (Product Head)
The gap became clearer when segmented by platform. Mobile and social audiences, particularly younger users, were engaging less with full-length articles. “They were interested, but they weren’t committed to the format,” Jain said.
Why this, and why now?
The newsroom’s diagnosis was not that the reporting needed to change, but that the presentation did, especially for younger audiences that indicated an attention deficit.
“We felt that the problem wasn’t the journalism but in how it was being presented across audience cohorts,” Jain said.
As AI tools and search interfaces reshape how audiences consume information, the expectation of a single, linear reading experience is increasingly outdated.
“Most digital publishers are essentially asking every reader to engage with a story in the same way,” Jain said. “But that’s not how the upcoming audiences seem to be engaging with content.”
For example, some readers want a quick overview; others want structured takeaways or simplified explanations.
The idea behind NewsEasy was to accommodate these different entry points within the same article, without fragmenting the journalism.
What they built: a multi-format AI layer
NewsEasy is a sticky, embedded widget that sits within a story and offers three distinct formats: a short “Article-in-Brief” summary, five key takeaways, and a Q&A-style breakdown.
From a reader’s perspective, the interaction is optional and non-intrusive.
“You open a story, and you see this AI bundle embedded within it,” Jain said. “If you’re in a hurry, you might just read the brief and move on. If you want slightly more detail, the takeaways give you that. And if you’re trying to understand a complex issue, the Q&A works really well.”
Each format is generated separately at the system level, using distinct prompts tailored to the output type. However, all outputs are strictly grounded in the source article.
“We’re very clear that the system cannot introduce new information or hallucinate,” said Abhilash Mallick, Editor, WebQoof at The Quint. “A lot of effort has gone into prompt design and guardrails to ensure the output remains faithful to the reporting.”
Implementation: layered onto existing workflows
A key requirement was that the tool should not disrupt editorial workflows. NewsEasy is applied after the story is written and edited, functioning as an additional layer rather than a replacement for any part of the process.
Editors decide whether to include the widget, typically for longer pieces, explainers, or stories where higher drop-off is expected. Once generated, outputs go through human review depending on the sensitivity of the topic.
“The system works only with the content of the article,” Mallick said. “Then there’s the human-in-the-loop approach, where editors still play a role in checking and approving the output.”
The build itself was iterative. Initial prototypes were developed within the product team, followed by close collaboration with editorial, engineering, and audience engagement teams.
Prompts were refined through repeated testing, and a feedback loop was built into the system to flag inconsistencies and improve outputs over time.
On the backend, the pipeline runs from the CMS through an automation engine into a custom backend, before rendering on individual articles on the website. Version control is maintained within prompt files, allowing for rapid adjustments.
Early impact: deeper engagement signals
The tool is currently live on a limited set of pilot articles, but early indicators suggest it is influencing how readers engage with long-form content.
“We are seeing improvements in scroll depth, especially on longer stories,” Jain said. “The targeted users are retained longer.” In several cases, overall time spent has also increased, pointing to more sustained interaction with the content.
The team is continuing to test widget placement and format effectiveness as part of an ongoing second iteration.
Lessons learned: format, cost, and control
One of the most significant shifts during development was moving away from a single-summary model to multiple formats.
“Initially, we thought one format might be enough,” Jain said. “But very quickly, it became clear that different users want different things.”
For example, the addition of multiple formats made the tool more useful and engaging.
At the same time, editorial concerns around tone and consistency required a structured approach. The team developed a centralised prompt framework aligned with The Quint’s style – clear, factual, and non-sensational – while allowing for desk-level adjustments.
Operationally, API costs emerged as a constraint. Without optimisation, costs escalated quickly, leading the team to introduce batching and caching strategies that reduced token usage by 38 percent.
Another lesson was around transparency. While the widget improved engagement metrics early on, it also highlighted the need for clearer disclosure of AI involvement to maintain trust.
What’s next
The next phase focuses on scaling and refinement. Plans include adding new formats like timelines, additional information metadata, podcasts and reader feedback mechanisms.
The team is also exploring language support and more advanced query handling to make the tool more interactive and accessible. On the infrastructure side, a cost-optimised embedding store is being considered to pre-cache high-traffic stories.
At its core, however, the approach remains consistent: adapting how stories are presented rather than rewriting the journalism.
As Jain put it, the goal is not to change the reporting, but to “build on our efforts to distribute content in personalised formats to enhance interactivity and accessibility” so that each content can be made suitable for wider audiences.
