GenAI is changing the way Indian media companies hire journalists
Generative AI Initiative Newsletter Blog | 01 August 2024
Greetings from Delhi! I am in India for INMA’s South Asia News Media Festival, my first business trip back to the city of my birth since 1997.
So much has changed — the immense amount of money sloshing around in the economy, the landscape (new roads and an effective, widespread Metro urban train system envied by those of us who live in Toronto), the rise of influencers (who, in total, collect more in revenue than news brands do), and the telecom infrastructure to support 820 million people regularly watching at least 20 videos a day on their cell phones.
There is also a marked change in knowledge workers’ attitudes — there is now an energy and dynamism I had not noticed a generation ago, a drive to get things done with a confident, ambitious workforce.
This manifests itself in many different ways. I spent two days on a study tour of more than half a dozen publications and then two days at the conference, hearing about the different, innovative ways in which AI is being used in Indian news publications and about the challenges that they are facing.
Here’s a quick glimpse of some of what I saw.
Hiring in the GenAI era: “full-stack journalists”
Generative AI is changing hiring at news companies in countries like India, where many people speak English as their second or third language. English-language publications, which have traditionally catered to the educated elite and hired from the educated elite, are no longer necessarily looking for people who write well in English.
What has changed?
Here’s an example from Rohit Saran, managing editor at the Times of India, which has a network of about 500 reporters spread across different regions of India. These reporters speak Hindi and regional languages well, but their English is not equally or uniformly strong — so the news brand needs about 400 editors to handle their copy so it can be rewritten in clear English and run past the reporters before publication.
The reporters now file their copy by dictating five bullet points into their phones in their own languages. The Times’ GenAI tool transforms that into 300 words in English and suggests three headlines.
“Nothing goes out without at least two pairs of eyes on it, but the editing time has been drastically reduced,” Saran said.
Language is no longer a barrier, so news organisations can focus on finding experts in different fields instead.
“We are now looking for people with deep domain expertise, and we only hire people who can work well with technology,” said Nalin Mehta, editor-in-chief at MoneyControl.com. “We look at your attitude, not your technological skills — because technology is constantly changing.”
Saran agreed: “We are looking for full-stack journalists now.”
AI at work: the personalised home page
Times Internet, the digital arm of the giant Times Group, is looking at AI for the algorithmic distribution of content so each of its 270 million users is served a personalised mix of content on the homepage, based on their profile.
In an era where search and social referrals are increasingly uncertain, Times Internet is working towards a goal of driving sustained loyal, direct traffic. For the past two and a half years, it has been building a self-learning, probabilistic model that is constantly detecting patterns and which understands certain topics have a longer shelf life than others.
“We all go to Twitter several times a day because the homepage is constantly changing. That is the experience we want to imitate,” Ritvvij Parrikh said.
The system does not favour popular or trending content, or more recent articles, as editors tend to. Older content is often surfaced, with favourable results.
The algorithm has produced sustained improvements over the past 18 months, indicating it is not just a flash in the pan, Parrikh said. Times has managed to double its click-through rate compared to the manual placement of content.
Editors still place the top two stories and can override the algorithm. About 35% to 40% of traffic comes from the homepage.
Wouldn’t the cost of computation be prohibitive? Parrikh pointed to methods to keep inference very cheap and said they were relying on compression to use as little data and computation as possible, but also indicated the project had a long runway for cost recovery.
Times Internet is also finding money in GenAI, which it uses to convert content from one format to another, such as text-to-Web stories. A good Web story brings in 50,000-60,000 users, who can be monetised — and it cuts the editorial team’s effort by a third when measured in minutes. Editors still pick the best photographs for the Web story.
“We have very consciously not removed human effort from it,” said Prasad Sanyal, business head at Times Internet, underscoring the importance of keeping a human in the loop in content creation. “It does the mundane, searching for 20 images for editors to choose from. This frees up editors for higher-order functions.
“The idea is that we are increasing reach with the least amount of incremental newsroom effort.”
HT Media, another Indian publishing giant, is also using AI across its newsroom in different ways:
The complete automation of some content, such as blogs on stocks and commodities, as well as on weather and pollution levels.
Automation to augment editorial capabilities for tasks such as translation, transcription, large document analysis.
Automation to help editors with coverage:
This includes a content prioritisation process that tracks what is trending on Google and runs an algorithm that looks at how much traction HT gets from different keywords that are trending. It calculates a score that is sent to editors, who then create stories and keep monitoring the yield on the stories, as well as the competition and speed of publication.
AI also monitors content quality for HT. It identifies key factors of content quality (such as the use of active voice vs passive voice), monitors trends, and then provides suggestions to editors via dashboards.
What is on their road map?
HT plans to work on using AI more deeply for monetisation and subscriptions, as well as increasing engagement on its live blogs. It also plans to experiment with text to video.
Worthwhile links
- GenAI and search: OpenAI joins the GenAI search game.
- GenAI and advertising: A lot of testing underway at ad agencies.
- GenAI and ROI: A look at how that’s going for other industries.
- GenAI and jobs: A new contract at Ziff Davis explicitly prohibits any union member from being laid off or having their salary cut as a result of using generative AI.
- GenAI and theft: Thousands of YouTube videos were controversially used for AI training purposes.
- GenAI and ethics: Google outlines its guidelines.
A non-AI diversion
In case you missed this charming human-interest story during the peak of Euro 2024 fever, read it now.
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.