The San Francisco Chronicle. The Texas Tribune. Time magazine. The Washington Post.
Dozens of America’s most well-known newsrooms are experimenting with chatbots to help readers pick restaurants, learn more about political candidates and dive deeper into articles. But new research from Poynter and the University of Minnesota shows people may not be ready for it.
Nearly half of Americans say they don’t want news from generative artificial intelligence, according to a wide-ranging study from Poynter and the University of Minnesota on how audiences feel about AI. Twenty percent of people say publishers shouldn’t use AI at all.
Benjamin Toff, associate professor at the Hubbard School of Journalism and director of the Minnesota Journalism Center, presented research at the second Summit on AI, Ethics and Journalism, led by Poynter and The Associated Press in New York City last week.
“The data suggests if you build it, do not expect overwhelming demand for it,” said Toff, who has been studying news audiences — and avoiders — for nearly a decade.
[Click here to check out the data referenced in this article, which was presented exclusively to news leaders at the Poynter and AP Summit on AI, Ethics and Journalism. When using the research, please cite Poynter and the University of Minnesota and link to this story.]
The research builds on a series of focus groups Toff led last year.
Data journalist Meredith Broussard, associate professor at the Arthur L. Carter Journalism Institute of New York University and author of “More Than a Glitch: Confronting Race, Gender, and Ability Bias in Tech”, was more direct during her keynote speech at the summit.
“Anybody really like using a chatbot? No. I can’t stand it. So, guess what? Your users feel like that, too,” she said. “They’re not excited about interacting with a chatbot on your site.”
Still, surveying audiences about chatbots is tricky, as most people haven’t actually experienced a news-based AI product — just customer service bots. And data from the study show that most people aren’t using generative AI as much as the researchers thought.
But there’s still value in journalists experimenting with technology, which is encroaching further into our digital lives every day. Hearst has been at the forefront of chatbot experimentation, with its restaurant-recommending Chowbot debuting in early 2024.
“We are basing this off of 30 years of high-quality reporting, high-quality editing, that Google might push to the side or not value in their model,” said Ryan Serpico, deputy director of newsroom AI and automation at Hearst Newspapers, of the value of trying to compete with Gemini, ChatGPT, Claude and other massive AI companies.
“We need to be experimenting with more formats of information delivery,” said Christina Bruno, digital growth strategist at Spotlight PA. “I think chatbots are one way of doing that.”
From left, Garance Burke, of The Associated Press; Lam Thuy Vo, of Documented; Phoebe Connelly, of The Washington Post; and Nikita Roy, of the “Newsroom Robots” podcast, discuss how to be an ethical leader of AI initiatives. (Courtesy: Keith Hughes)
Audience-facing AI platforms are more well-received and impactful abroad, according to Nikita Roy, founder of the “Newsroom Robots” podcast and Knight Fellow at the International Center for Journalists.
An EU election chatbot developed by Swedish publisher Aftonbladet answered more than 150,000 questions from the audience, and a content marketing virtual assistant deployed by Ringier Axel Springer in Poland to boost German tourism generated 33,000 unique travel plans for users.
“Experiments are great,” Broussard said. “But you’ve got to pay attention to the results of the experiment.”
According to the research, Americans remain broadly wary of AI in the news:
Fewer Americans — especially young people — are using AI than many news leaders thought. Nearly half of those from 18 to 29 years old haven’t used or heard of generative AI tools like ChatGPT. Large swaths of news audiences think publishers are using AI to write articles or create photos. More than a quarter of those surveyed think newsrooms are using AI to write the text of stories often or all the time. But they have low confidence in how journalists actually use AI. More than half of Americans have little or no confidence in newsrooms writing articles or creating photos with AI. And a high demand for disclosure and specific ethics guidelines for AI use remains. Of the most engaged news consumers — those with high levels of news literacy — more than 90% want disclosures for AI-generated text and photo editing.
There’s a clear disconnect between what journalists are actually doing with generative AI and what audiences think they’re using it for.
“It’s not surprising,” said Zuri Berry, digital strategy editor at The Baltimore Banner. “It also serves as a confirmation of our current approach to AI, which entails disclosures, human review and verification and limitations on some tools that undermine our trust and credibility with readers.”
Chloe Lee Rowlands, web manager and data journalist at Bay City News Foundation, said that newsrooms must go beyond simple, begrudging disclosures and practice “radical transparency.” That means being intentional, specific and — in the case of cartoonist Joe Dworetzky — personal.
Dworetzky, like photojournalist Ray Saint Germain and others at Bay City News, wrote about his experience incorporating generative AI into his work. The sprawling piece chronicles his experience with ChatGPT’s DALL-E and Adobe Firefly, and learning how to prompt the programs to get what he actually wants.
“I probably would’ve been a bit wary to task someone with that,” Rowlands said during a panel on local news at the summit. “But for him to come to me and be so interested in exploring that, I thought was a really ripe opportunity to explore those ethical considerations, but also emotional and personal considerations.”
MediaWise, a digital media literacy initiative based at Poynter, debuted at the summit an AI literacy toolkit developed with AP aimed at bridging the digital divide between news consumers and journalists. The toolkit offers practical resources for newsrooms looking to engage audiences with AI literacy, along with examples of how some newsrooms are already communicating about AI.
Serpico said he plans to pitch an AI literacy series to his bosses this week.
“I’m personally interested in creating a series of videos that demonstrate how we’re using AI not to generate slop at scale,” he said, “but to transcribe the audio of eight school board meetings that take place at the same time each week so that our education reporters can more effectively get a grasp of the important topics that parents in those communities need to know about.”
Ryan Serpico, deputy director of newsroom AI and automation at Hearst Newspapers, shared lessons from Hearst’s Chowbot and Kamala Harris News Assistant at the second Summit on AI, Ethics and Journalism. (Courtesy: Keith Hughes)
The research also highlights the need to ask the audience for feedback while experimenting with generative AI, something Vania André, publisher and editor-in-chief of the Haitian Times, has done regularly. The Haitian Times planned to build a chatbot to surface immigration information but scrapped the idea after surveying readers, who were more interested in business news.
“But it’s a sobering moment where, again, that echo chamber, right?” she said. “You would have thought this would have been something that everybody wanted and it was just the complete opposite.”
Toff often cites journalists writing for news junkies as a component of news avoidance. Audiences can feel like they turned on a random episode of “Game of Thrones” and are expected to know the characters and why storylines are important.
Generative AI — in the form of automated summaries, clickable explainers or even chatbots — holds promise in bringing casual readers closer to the story.
“These tools can make packaging that information, all of that context, all of that background information that a lot of people just don’t have, and make that news more accessible to people,” Toff said. “I think there’s a real opportunity there to reach more and more of the public who right now feel pretty disconnected from news and its value.”
MediaWise interactive learning designer Sean Marcus contributed to this story.
