Greetings, everyone! I’m just back from INMA’s dynamite study tour and conference in Mumbai, a truly mind-expanding experience. You can read a quick recap of some key learnings here, but what I really want to talk about is some of the interesting moves Indian publishers are making with AI.

India is a remarkable place. Print is still thriving — and, simultaneously, younger audiences are changing the game digitally. Data is dirt cheap, and video is the dominant form of news consumption. Scale is still the name of the game.

There is a real energy and dynamism in the news ecosystem as different players hire some of the brightest talent available to jump at opportunities to reach and monetise audiences more effectively.

What can the rest of us learn from Indian publishers on AI? I’ll be writing more about them in the future, but here are some lessons from one publisher, HT Media, featured today:

• Focus and prioritisation pay off. 

• Put your audience first. 

• Put AI tools where your newsroom already operates. 

• Look for ways to reuse technology for different purposes. 

• Think big.

Thanks for reading, 

Sonali

HT Media’s “disproportionate gains” with AI

I marvel at the clarity and speed with which some Indian news companies are approaching AI initiatives. For example, HT Media, which publishes the widely circulated Hindustan Times, has a remarkably focused way of thinking about where it uses AI.

One key learning has been that leaders should not ask what AI can do. Instead, they should focus on the problem to be solved with AI, said Puneet Jain, CEO of HT Digital:

“Because AI can solve everything. But if you don’t apply the efforts and priorities on the right problem, the outcomes will not be commensurate.” 

HT’s approach involves understanding AI intervention across three levels, Jain explained:

• A level one problem is when you have an existing product, process, or workflow you want to improve using AI. Typically, it leads to productivity and efficiency gains. “There have been more disappointments than successes when it comes to this because it’s just not easy to measure cost saved in some specific areas, and, more importantly, they get lost in some other areas,” he said.

• “The second stage of AI integration happens when you look to scale a particular solution which you know already works but you are not able to scale otherwise because of complexity, difficulty, or, in general, effort required in doing it. A typical example of this would be personalisation.” 

• “The level three of the AI intervention happens when you start imagining your products or experiences which until now couldn’t have been done without AI. That’s where disproportionate gains happen.”

HT started with level one problems, but its deployments now really focus on levels two and three, Jain said. 

For example, it uses four agents working in parallel to help its editorial team determine which story to cover by examining search trends, tracking the article’s performance in real time, comparing it to similar articles produced by other publishers, and providing actionable recommendations on how to improve the story. 

“We have been able to significantly improve SEO coverage by about 50%. Wherever these recommendations are accepted, we see an immediate jump in traffic and yield of 25%,” Jain said.

“The ranking insights are very sharp, and we are now analysing the performance not a day later but in complete real time.”

Still, building the tools was only half the battle. Writers could use an interface to understand how a story was performing, but they needed to “break their regular rhythm and come to a tool and do a pull-based search,” leading to spotty adoption, Jain said. All the recommendations are now pushed into relevant Slack groups instead, where journalists are already interacting. 

HT Media has also automated the production of articles in five formats.

“For all these formats, the starting point — or the raw material — is exactly the same: the standard text-based article created by editorial,” Jain said.

“We are able to generate all of these completely in the backend without any additional effort from editorial. Of course, there’s effort that goes into experimenting it, where there’s a sign-off that happens between product and editorial. But once it is perfected, at every article that is generated in HT Media, these formats are automatically created. And we see that the engagement on these formats is significantly more than a typical text article.”

Every single reader also gets a personalised homepage now, based on their behaviour and their own inputs as well as content that other readers find engaging. 

Date for the calendar: Wednesday, August 6: GenAI Webinar

AI is incredibly helpful for reporters, particularly those who need to wade through documents or large amounts of data and find patterns and links. 

Two acclaimed newsroom innovators will join our Webinar to tell us how they are using AI for this. Please join us (or register to watch the recording)!

HT builds out AI-driven monetisation

Advertising is the largest revenue driver for HT Media, as it is for most Indian news companies.

An increasing focus of the advertising industry (especially for digital formats like display) is performance or bottom-funnel advertising, but HT had typically provided only top-of-the-funnel options for its advertising sales team. 

The publisher has now built an AI-driven full-funnel advertising platform for clients — and found that the system is also proving useful for its in-house marketing team. 

The system, called Echo, uses their in-house ad server, is integrated with a marketing suite that offers more than 20 native inventories, and marries it with an underlying data layer, letting HT showcase personalised ads at scale.

“At any given point of time, we have used AI to be able to identify the right ad, at the right inventory, at the right article, for the right brand — so that it delivers better engagement for advertisers and better outcomes.”

HT also offers features such as A/B testing, brand lift studies, and fatigue management.

“We are able to add impressions on the fly … . We’ve been able to significantly improve our ad inventory without necessarily looking to get more traffic or more users,” Jain said.

CTRs are three or four times higher than typical CTRs in news publishing. Readers are finding ads less intrusive, which shows up in a 35% increase in engagement. And ads “have entered the performance stage where we are delivering outcomes or conversions.” HT is now delivering more than two billion monthly impressions and a million conversions. 

“In fact, because this is both advertising as well as a marketing suite, we are not only limiting ourselves to improving our ad offerings but also to improving our customer lifecycle goals,” Jain said. “For example, we now know which user to target for app downloads of our news products and which user to target to source subscriptions for our news products. So the same set of ad tech is also helping us improve our reader engagement goals.”

What’s next? Jain wants to own the transactions business as well. 

“We will look to drive actual online transactions completely — originating from our systems, from our sites, and finishing on our sites using the same tech.”

Also on the road map: a WhatsApp commerce engine built around the advertising and marketing suite. 

“This is a completely conversational interface where because we already know our users — they are logged-in users, we know their interests, we know which ads they interacted with — we reach out to them with a similar product and we offer them conversational capabilities to be able to convert to a lead or, in some cases, completely to a transaction, including payments, while being in WhatsApp,” Jain said.

“Conversational commerce is already big in China. I think that is about to happen in India as well.”

Also coming up: the ability to add inventories from the open Web.

Worthwhile links

• GenAI and new audiences: Dow Jones launches a new AI language service in French; it already publishes AI-powered translations in Korean and Japanese.
• GenAI and Axel Springer: Using AI is expected of all staff, says CEO.
• GenAI and local news: From Switzerland, an idea to revitalise hyper-local news using AI.
• GenAI and answer engines: News stories are among the top sources for AI bots like ChatGPT, Claude, and Gemini.
• GenAI and layoffs: “They align suspiciously well with the rollout of large AI systems.” 
• GenAI and efficiencies: How, exactly, are they being measured?
• GenAI and content: News organisations seem to be getting more comfortable with the idea of GenAI-written content.
• GenAI and advertising: Looks like we still need humans.
• An AI diversion: An AI-generated band got a million plays on Spotify.

About this newsletter

Today’s newsletter is written by Sonali Verma, based in Toronto, and lead for the INMA Generative AI Initiative. Sonali will share research, case studies, and thought leadership on the topic of generative AI and how it relates to all areas of news media.

This newsletter is a public face of the Generative AI Initiative by INMA, outlined here. E-mail Sonali at sonali.verma@inma.org or connect with her on INMA’s Slack channel with thoughts, suggestions, and questions.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .