One overriding theme that has emerged during INMA Generative AI Initiative Lead Sonali Verma’s conversations with more than 90 news publishers: Most GenAI being used by news media companies today is found in the newsroom, and internal uses far outweigh audience-facing ones.

For companies that use GenAI, Verma told attendees of the recent INMA Latin-American Conference she has seen an “80/20 rule” emerge: “You can get AI to do 80% of the work, but for the remaining 20%, you need a human on either side asking the right questions and vetting the output.”

Although AI is attracting massive attention right now, Verma noted AI has been around for years. For example, The Wall Street Journal’s paywall has depended upon AI since 1998 and Canada’s Globe and Mail has used an AI-driven algorithm for several years to determine where to place content based on its performance. 

What has changed and why is it creating such an uproar?

“With generative AI, we are dealing with a probabilistic model versus a deterministic one. It’s just guessing,” she said. 

Traditionally, AI followed the rules of “if this happens, then do that.” However, GenAI uses words, not code, and with fewer rules, the new models can manage greater ambiguity and complexity.

“We can scale them more quickly, but they are less accurate than the previous ones,” Verma said. “They’re more creative, they’re flexible, and they can make up stuff. The key point to remember, of course, is that we always need a human to ask the right questions and to look at the output and make sure that it is accurate and not absurd.”

According to a report released in April by the Associated Press, a survey of 200 journalists found the primary use for GenAI right now is content generation, which nearly 70% of respondents said they were using it for.

“They’ve used it for generating content like new headlines, social media posts, newsletters, text from data, and story draughts,” Verma said. “It’s also used for discovery, ideation, brainstorming, and of course, other top uses include coding, working with data, creating presentations, or drafting emails.”

She shared several case studies from INMA members to offer examples of the many ways news media companies are using GenAI, from providing article summaries to crafting chat and e-mail responses to customers to copy editing to creating an AI-generated television anchor.

As companies become comfortable with how GenAI fits into operations today, it’s important to look at where this is heading, she said.

The future of GenAI

When the Associated Press asked journalists where AI was most helpful, many wanted automation to reduce repetition or increase precision to help them do their jobs better. “They would like to have the AI do a task and have it be their job to simply review it,” Verma said.

In the future, GenAI could become more agentic and able to pair current capabilities with an understanding of what actions need to be taken, Verma said: “The core idea is that you set a task and the agent comes up with a multi-step plan to achieve its goal. So it could do research, it could write a draft, and it could suggest a headline.”

While this era is ushering in many changes, some things about journalism will not change, she promised, citing findings of a report recently published by the European Broadcasting Union: 

“Journalism still focuses on accuracy, facts, storytelling, holding power to account. These are things the machine cannot do. They also depend on trusted and stable relationships. You can’t get a story unless you have one.”

Humans will still set editorial priorities, meet and talk to people, and break big news stories.

Mistakes and best practices

News media companies entering the GenAI space can learn from those who have gone before them, Verma said. Common mistakes include building something that doesn’t add value to the business, creating something users don’t want or need, or not looking at the big picture of how it could improve other workflows within the company.

“One mistake also is that we start with projects that are too big, and so it’s hard to show quick wins,” she said. “It takes a long time to develop any learnings that we can apply to other projects. So starting with a smaller project is better.”

She offered some of the best practices to help others on their journey, beginning with establishing clear AI guidelines and standards: “What is it allowed to do? What is it not allowed to do?”

Creating an interdisciplinary team is also critical: “So [you need] a cross-functional team with people drawn from different departments with different perspectives and backgrounds and different skills. These same people will be your ambassadors. They will go back to their teams, to their departments, and help answer questions on your behalf and help get good ideas and feedback on what you’ve built. They really help socialise the product and the experiment.”

Verma also reminded everyone to keep a human in the loop and urged them to “always remember your mission and values. Don’t do anything that runs counter to that because you will regret it.”

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