A significant number of journalists at the Sacramento Bee are expressing concerns over the deployment of generative artificial intelligence in news reporting. Out of the 40 journalists, more than 30 are voicing their discontent regarding this technology’s role in news content generation.
Members of the Sacramento Bee News Guild—a union representing reporters at the Bee—have opted to withhold their bylines from stories produced by the newly-introduced Content Scaling Agent (CSA), an AI-powered tool.
According to Ariane Lange, Vice Chair of the Sac Bee News Guild, this tool was developed by the Bee’s parent organization, McClatchy Media, to boost the quantity of content generated by the newspaper.
“For instance, if a reporter writes a story highlighting a change in the Sacramento City Unified School District, the AI might create a slightly altered version of that piece, tailored more towards parents,” Lange explained.
Lange, an investigative reporter at the Sacramento Bee since 2021, expressed her discomfort with an AI system manipulating her narratives in unexpected ways. “It feels wrong to modify a published story into a diluted version just to attract clicks from a particular demographic,” she said. “It feels a bit exploitative.”
While the Bee has published stories created by AI, these articles are clearly marked as such, indicating they are based on a reporter’s original work. Lange emphasized that they want to preserve the integrity of reporters’ bylines to maintain the credibility of the CSA.
“The public trusts reporters,” Lange said. “They are relying on our credibility to bolster the reputation of this AI tool, which many at the Sacramento Bee don’t support.”
All news articles, including those generated by AI, must be reviewed by human editors, as mandated by the Bee’s AI policy.
Lange pointed out that the Guild’s primary concern is the overall quality of the newspaper, beyond just AI-generated pieces. “This initiative will likely compromise the quality of the stories we are most passionate about,” she remarked. “Our editors will be overwhelmed, juggling these new AI-generated stories alongside their usual workload. We recognize that this approach prioritizes quantity at the expense of quality.”
Lange noted that the tool is “sandboxed,” meaning it is designed not to retrieve information from sources outside its designated URL. Chris Fusco, Executive Editor at the Sacramento Bee, did not provide details about how the tool was developed, only referring to the details outlined in the AI policy.
The headline of a recent Sacramento Bee story that was created using generative AI and edited by a Sacramento Bee editor. There is no reporter byline.Screenshot of Sacramento Bee website
“Our policy asserts that we do not mislead our readers, ensuring transparency in our use of bots and automated technology in our reporting and products,” the policy states.
Although the policy does not elaborate on how these tools are built or trained, it mentions that the organization collaborated with technologists to define the parameters for story templates.
“For those familiar with the phenomenon of AI ‘hallucination,’ where systems produce fabricated facts, rest assured: our automated content does not utilize that kind of technology.”
When harnessing generative AI, particularly with large language models, there is a risk of responses appearing factual while lacking substance, particularly in information gaps.
While McClatchy’s model may not hallucinate, Lange noted that it has already made errors during its initial rollout. “We are aware that it generates mistakes, as reporters have been pressured into adjusting stories sourced from their work,” she revealed. “While it may not hallucinate as much as a system like ChatGPT, which draws from a vast array of online information, it is still not human and does not discern fact from fiction.”
McClatchy Media has yet to respond to requests for comment.
AI and journalism: the ethics of generation
CalMatters has been refining its AI tools for journalism for over ten years. The initiative Digital Democracy aims to monitor bills and illuminate the legislative processes they undergo.
Neil Chase, the CEO of CalMatters, shared that their organization employs AI to cover topics that often go unnoticed by journalists. “On one hand, generative AI is being utilized to produce considerable amounts of misinformation, which is detrimental to journalism. On the other hand, there are applications of generative AI that enable journalistic practices that might not otherwise be feasible,” Chase noted. “At CalMatters, we utilize AI tools for data analysis about the state legislature, instead of generating articles directly.”
Chase explained that their Digital Democracy initiative analyzes data to discover potential story ideas—such as inconsistencies or varying behaviors—that reporters might observe when covering the Capitol. “These findings serve as leads, akin to jotted notes in a reporter’s notebook, meant for further investigation,” he elaborated. “We provide these leads to reporters statewide, not just within CalMatters.”
He emphasized that CalMatters does not produce any of their articles via generative AI, primarily due to the depth of reporting they undertake. In situations where resources are constrained, AI can assist reporters in covering aspects they might overlook. For example, small-town newsrooms have created tools to monitor police scanners and alert them to significant incidents they may have missed.
“That seems like a beneficial tool,” Chase remarked. “It’s very different from publishing content that directly impacts audience trust.”
Chase foresees a future role for AI in journalism and has been impressed by recent advancements in the field. “While we cannot use these tools indiscriminately, we also cannot dismiss the possibilities, suggesting that AI will never achieve ethical reasoning or perform specific tasks,” he commented.
However, Djordje Padejski, a professor at Stanford and the Walter Cronkite School at Arizona State University, argues that current technology does not possess the capability for ethical decision-making particularly in journalism. “My concise answer is no,” Padejski asserted. “Not in the manner required by journalism.”
He explained that language models are engineered to predict the statistically likely next word, which can create a seemingly realistic yet sometimes inaccurate output. “They optimize for probability, coherence, and pattern recognition, but do not prioritize truth. They aren’t designed to understand nuances like harm or public interest,” Padejski elaborated.
He added that the accuracy threshold for AI systems fails to meet the rigorous standards expected in journalism. “In computer science, an accuracy rate above 80% is celebrated,” he noted. “However, journalism demands a verification-based, evidence-driven, and document-focused approach. AI operates on a statistically-patterned, synthetic framework.”
Padejski observed that even human summarization of news stories can lead to important details being misrepresented. This concern resonates with Lange as she navigates the implications of using the Bee’s content scaling agent.
As an investigative journalist at the Bee covering traffic fatalities, Lange has documented every individual lost to traffic incidents on city streets since 2024. In instances of these tragedies, she often interacts with grieving families on their most challenging days. “I strive to honor the lives of their loved ones and ensure public officials are held accountable for preventing traffic fatalities,” Lange asserted. “However, the thought of my stories being processed by a chatbot feels revolting.”
CapRadio relies on the support of its listeners for credible news reporting. As a nonprofit entity, your donations play a vital role in sustaining the journalism that brings important stories to light for our audience. If you believe in our mission, please consider donating today.
