Newsroom leaders have heralded artificial intelligence tools as having the potential to remake journalism — for example, by allowing journalists to spend less time on mundane tasks, such as scouring police blotters, to focus on real reporting.
But realizing AI gains in the newsroom hinges on realistic expectations and managerial buy-in, according to research exploring AI use at global news organizations.
Two recent academic papers feature researchers embedded at the Associated Press and the BBC, capturing a key period when the industry was beginning to grapple with what artificial intelligence means for reporters, editors and the future of journalism writ large.
The first paper explores how journalists’ expectations about AI affected an AP effort to engage local newsrooms with AI tools. The second paper looks at how BBC journalists used AI tools — and the value journalism research can bring to news organizations learning to adapt to new technologies.
Here are four big takeaways from the research.
- While AI tools have been hyped on social media and in traditional media, it’s important for those in charge of AI trials or experiments to manage expectations, as AI tools often need to be tailored to the specific needs of particular newsrooms.
- AI tools by and large require a human in the loop — journalists to ensure that what is produced is accurate, whether it’s an AI-generated summary of a press conference, a draft news story or an analysis of a dataset.
- Embedding researchers in newsrooms can mean making university resources available to help news organizations develop AI tools that meet their needs, and newsrooms that work with academic researchers may avoid becoming pigeonholed in how they use AI, based on expectations or biases.
- Buy-in from newsroom managers is critical to the success of any type of experimentation with AI technologies.
For the first project, the authors observed dozens of online meetings held from March to October 2023 as part of the AP’s initiative to help local newsrooms understand and adopt AI. These included private meetings between project managers and newsroom staff and public webinars meant to inform the broader news industry about AI tools and their implications for journalism.
“The core idea of the AP’s AI initiative for small newsrooms is rooted in a specific narrative of the future: AI is going to dramatically change journalism as an industry,” the authors write in “AI Hype and its Function: An Ethnographic Study of the Local News AI Initiative of the Associated Press,” published in January in Digital Journalism.
They explain in the paper that those transformational expectations about AI, as laid out by the initiative, “shaped the perceived importance of the initiative” and were critical in securing grant funding.
As the initiative moved forward, the authors observed project managers tempering expectations about AI, particularly as they began producing the educational webinars and working with newsrooms on AI-related projects meant to tackle specific challenges.
“What we did is really capture this snapshot in time where AP and both program managers were trying to experiment with these tools and gather people to do that and build up excitement to do it,” says one of the authors of the paper, Nadja Schaetz, a postdoctoral researcher at the University of Hamburg in Germany.
The John S. and James L. Knight Foundation funded the first two years of the initiative, which began in August 2021. Program managers Aimee Rinehart and Ernest Kung, whose positions are now funded by the AP, according to the paper, first set out to survey newsrooms on their AI readiness, gathering 192 responses from news leaders across every state, the District of Columbia, Guam and Puerto Rico.
“What most newsrooms and their audiences need is the automation of basic information, such as social media content and high school sports scores,” Rinehart and Kung wrote in a March 2022 report based on the survey responses and follow-up interviews with leaders from 25 newsrooms. “They need help managing information overload.”
The initiative soon benefited from a bit of propitious timing.
In November 2022, with the survey complete and the webinars under way, OpenAI released ChatGPT.
Newsroom leaders quickly realized the inherent promise — and began considering the pitfalls — of AI tools that could generate written articles and perform other tasks once firmly in the realm of human cognition.
Virtually overnight after ChatGPT was released, the AP’s AI webinars went from attracting fewer than 100 people to attracting upwards of 1,500 people, the authors write in the paper.
“The changed momentum was, therefore, not grounded in showing the direct result or emerging profits of the individual projects,” write Schaetz and co-author Anna Schjøtt, an AI researcher at the University of Amsterdam. “Rather, it was because the project started to reflect an investment in collective expectations.”
A narrative began to take hold across the news industry in 2023: AI could free up journalists stuck performing mundane or rudimentary tasks, such as parsing information from police blotters, to focus on deeper, nuanced reporting.
The authors note that local newsroom representatives participating in the AI initiative said AI tools had freed up time for “original journalism,” helped to reduce burnout, and that these tools should be viewed not as a replacement for reporting but a complement to it.
By examining journalists’ expectations around AI heading into the initiative, the authors also note that newsroom staff and leaders should be aware that expectations shape assumptions about the future.
“If we have very definitive claims about the future what happens is that we make the future by doing that,” Schaetz says. “We’re kind of narrowing down what can happen.”
Inside the Local News AI initiative
Following the survey, the Local News AI initiative began working with five newsrooms to explore how AI tools might help them tackle specific challenges.
- For KSAT-TV in San Antonio, Texas, that meant automated transcripts and summaries from live news events, such as press conferences, with the aim of “allowing reporters to work on other stories,” according to an AP case study on the project. “As technology stands today, there is no such thing as a perfect AI-powered transcript. It was determined that a human would need to be in the loop in between the transcription and summarization processes.”
- Michigan Radio, the NPR affiliate at the University of Michigan, already had an app to transcribe city council meetings. The app was enhanced to improve transcript quality and alert reporters when transcripts were ready, with journalists “able to choose keywords and get alerts whenever those words showed up in the transcript of a meeting,” according to the case study on this project.
- For El Vocero, a newspaper serving all of Puerto Rico, participation in the initiative resulted in a weather bot that would automatically produce publishable weather alerts to quickly bring lifesaving information to the paper’s digital audience. “The project was pitched to replace a manual process which involves a journalist sitting on the [National Hurricane Center] webpage and constantly refreshing for updates every three hours,” according to the case study.
- The newsroom of WFMZ-TV in Allentown, Pennsylvania, parsed through thousands of emails to train an AI tool aimed at automatically sorting high-quality pitches into a coverage planner, according to an October 2023 report on the initiative.
- For the Brainerd Dispatch in Minnesota, participation in the initiative resulted in a natural language generation tool that could produce basic news stories based on police blotters. The development team encountered challenges with the tool’s ability to digest a variety of formats from different agencies, and the scope of the tool was narrowed to create stories based on blotters from three law enforcement agencies, according to the case study.
“We chose projects that didn’t just solve one newsroom’s need,” Rinehart says, noting that several newsrooms asked for help parsing information from police blotters.
“We thought, well, if we built a tool, could we build it with best practices in mind?” she adds. The police blotter tool they built aims to avoid publicly excoriating people arrested for but not convicted of crimes by excluding names, race and ethnicity from the basic news stories.
Development teams from Northwestern University, the University of Michigan and Stanford University helped create the AI tools and were the “superpower” of the initiative, according to the October 2023 report.
Meanwhile, observations from Schaetz and Schjøtt, and their subsequent paper, offered a neutral perspective: “In the context of the initiative, some of the promised value of AI was nevertheless challenged by the participating newsrooms,” write Schaetz and Schjøtt. “Specifically, the expectation that the developed tools would free up resources and streamline processes was questioned given the remaining need for human oversight.”
For the AP program managers, the researchers offered perspectives that would have been otherwise unavailable.
“We just wanted that feedback loop of, what does an outside observer think of this and think of how we’re working?” Rinehart says. “It’s nice to have that sort of slight distance to see, how did I come off? How was the research received?”
As for AI as a panacea — a cure-all that will allow journalists to do more in-depth work — Rinehart notes that AI tools did free up some time for reporters in the newsrooms the initiative worked with, but it’s also not a question the initiative has rigorously assessed.
“I think right now it’s marketing terminology or phraseology,” Rinehart says of the idea that AI can give journalists more time to do journalism. “That hasn’t really been proven.”
But, she adds, smaller newsrooms are more likely to see efficiency gains because they produce relatively few stories — several dozen per week compared with thousands per day from a global news outlet like the AP.
Rinehart points to two other small newsrooms that have found value in AI tools: The Baltimore Times, which has used AI to generate images for stories, and The Haitian Times, which has used AI to find expert sources for their articles and do first pass edits.
“If something truly fills a need,” Rinehart says, “people will use it.”
A researcher-newsroom collaboration at the BBC
At the time Schaetz and Schjøtt were observing the development of the AP’s Local News AI initiative, other researchers were embedded at the BBC to explore the needs of journalists experimenting with AI tools while working at a massive news organization with an international audience of 450 million people.
From 2020 to 2023, researchers studied how the BBC was using AI — and in June 2023 produced a primer on the risks of using AI in journalism, aimed at reporters and news leaders across the industry.
“Impact-oriented research in applied disciplines like journalism has a long history but grappling with AI and other emerging technologies will require robust academic-industry collaborations that span disciplines,” they write in their January 2025 paper, “Action research at the BBC: Interrogating artificial intelligence with journalists to generate actionable insights for the newsroom,” published in Journalism.
The authors also interviewed more than a dozen BBC journalists on their attitudes toward AI, and convened workshops with technologists, journalists and academic researchers to discuss the use of AI in journalism.
Through the interviews and other informal conversations, the authors find that “most journalists engage with AI systems as ‘users’ of pre-made tools and end up as ‘observers’ of AI in the news industry, rather than as contributors to directing the future shape of their profession or practitioners with agency in determining how newsroom technology is developed, negotiated or in fact resisted,” they write. “A fundamental barrier to their understanding of AI which can in turn lead to disengagement, is how invisible and abstract AI seems to them.”
One way newsrooms can help is by educating journalists with concrete definitions of AI, says one of the authors, Bronwyn Jones, research lead in journalism and AI at the Responsible Innovation Centre for Public Media Futures at the BBC.
“There should be delineation between certain different types of AI, or at least an attempt to provide enough of an understanding of the purposes and underpinnings of different types of AI to get people to grip with those basics before moving on,” Jones says, speaking in her capacity as an academic researcher. “So, I think the kinds of professional literacy efforts that you see across organizations now are very much needed.”
Jones is also a former BBC reporter and is a translational fellow at the UK-based Bridging Responsible AI Divides. She says newsrooms that work with academic researchers may avoid becoming pigeonholed in how they use AI, based on expectations or biases around the technology.
“Journalism can be quite inward looking,” Jones says. “[The field of] journalism studies itself is basically a collaboration of many different disciplines focused around one profession, or one institution in society. So, it’s actually really well placed to coalesce those different insights from different places. From what I’ve noticed in organizations like the BBC is that they really appreciate the translation of knowledge. But it’s got to be in certain formats. It’s got to be quick and it’s got to be accessible.”
That’s a big part of what the authors call translational work — talking to journalists “using language and repertoire familiar to them in order to convey unfamiliar concepts or ways of thinking necessary to understand each other’s worlds,” they write. In the case of the study with the BBC, the authors found common ground with journalists by emphasizing their interest in improving public interest reporting through mutual respect and constructive criticism.
For researchers who want to collaborate with reporters on AI projects or observe how journalists are using AI tools, it’s essential to get buy-in from newsroom leaders, so reporters are given time away from their regular duties to engage with the technology.
“They see it as a break, a moment that they can reflect and engage in things — not just on the hamster wheel, if you like, of production,” Jones says.
Expert Commentary