New research published in Communication Report Readers have found that they interpret the involvement of artificial intelligence in newswriting in a variety of, often inaccurate ways. News articles with various byline descriptions (noting that the article was written with or written by AI) provided a wide range of explanations for what it means. Most people didn't see AI acting independently. Instead, they built a story in their minds to explain how AI and human writers worked together.
As AI technology becomes more integrated into journalism, it becomes increasingly important to understand how people interpret the role of AI. Generated Artificial Intelligence refers to a tool that allows you to create human-like text, images, or audio based on prompts or data. In journalism, this means that AI is often used to summarise information, generate headlines, and write complete articles based on structured data.
Since 2014, some newsrooms have used AI to automate financial and sports stories. However, the release of more advanced tools, such as ChatGPT, in late 2022, has expanded the possibilities and makes AI far more visible in everyday news production. For example, in 2023, a large UK media company hired work that included AI support and pointed it out on their byline. However, readers are not always informed exactly how AI has contributed. This can cause confusion and doubt.
The researchers behind the new research wanted to know how people understand the bystanders referring to AI, and whether their interpretations are influenced by their familiarity with the media and attitudes towards artificial intelligence. They were particularly interested in whether people could accurately guess what AI did during the writing of the news article, based on the wording of the signature. This is important because trust in journalism relies on transparency. This relies on previous controversy that shows sports are being accused of using AI-generated content without disclosure, but shows that unclear authors can undermine credibility.
To explore these questions, the research team designed an online study that included 269 adult participants. This sample closely reflected the US population in terms of age, gender and ethnicity. Participants were recruited and paid for that time through Prolific, an online platform often used for social science research. After giving consent, participants completed a short survey measuring media literacy and general attitudes towards artificial intelligence. Each person was then randomly assigned and was able to read a slightly edited Associated Press article about healthy stories. This article was the same for everyone except for the one line at the top, the signature type.
This byline was varied in five different ways. Some stories said they were written by “staff writers,” while others said they were written “by staff writers using AI tools,” “AI support,” “AI collaboration,” or simply “by AI.” After reading the article, participants were asked to explain what the signatures meant and how they interpreted the role of AI in writing the article.
The answers indicated that readers tried to understand the meaning of byline, even if they were not completely clear. This act of constructing meaning from limited information is known as “sensemaking.” This is a process that people are using what they already know or believe in understanding new or vague. In this case, people relied on personal experiences, assumptions about journalism, and existing knowledge of AI.
Many participants hypothesized that AI helped in some way, no matter how precisely they couldn't say it. AI wrote most of the articles and thought that human editors were intervening to make things clean. Others believed that humans wrote most of the articles, but used AI for smaller tasks, such as checking facts and suggesting better language.
Imagine that a journalist entered several keywords, and AI generated articles by compiling text from the internet. Another described the collaborative effort in which AI gathered background information and human writers evaluated its accuracy. These mental models (often called “folk theory”) show how readers try to fill gaps when information is missing or vague.
Interestingly, even when the signature said the signature was written “by AI,” many participants still assumed that humans were involved in one way or another. This suggests that most people do not view AI as a completely independent writer. Instead, they believe that human surveillance is necessary for guidance, directing, or final editing.
Some participants expressed skepticism or frustration towards the bystander. When they said that the article was written by a “staff writer” but not the name was included, some assumed this was an attempt to hide the fact that AI actually wrote it. Others said the quality of the writing was poor, and this was attributed to the involvement of AI, even when the article was described as being written by humans. In both cases, negative judgments were derived if the specified author was absent. This finding supports previous research showing that readers expect authors to be transparent, and when those expectations are not met, they may feel distrustful of content.
To further understand what influenced these interpretations, researchers grouped participants based on media literacy and general attitudes towards AI. Media literacy refers to how well people understand the media they consume, such as how news is produced.
Researchers found that participants with high media literacy were more likely to believe that AI had mostly written. Those with low media literacy were more likely to assume that a person wrote an article or that the work was a collaboration with a human. Surprisingly, previous attitudes towards AI had no significant impact on how participants interpreted their signatures.
This suggests that when trying to understand who wrote the story, it can be more important than how many people feel about artificial intelligence about media. It also shows that simply including phrases like “AI Asistance” is not sufficient to clearly understand the role of AI to readers. In this study, people often misinterpret or overthink these statements, and the lack of standard language regarding AI involvement only adds to the confusion.
This study has several limitations. The researchers did not include the appointed authors in any of the signer's terms, so participants may have responded negatively because they missed seeing the actual person's name. It is also worth noting that the articles used in this study are based on scientific reports. Responses to AI involvement may be stronger for topics such as politics and opinion writing. Future research will explore how these findings apply to other types of journalism and how people respond when articles include full disclosure or transparency statements about the use of AI.
Despite these limitations, the investigation raises important questions for news organizations. As AI becomes more common in the newsroom, it's not enough to say that stories were created “with AI.” Readers want to know what AI did exactly. Have you written your first draft, summarise your data, suggest edits, or spellcheck the final copy? Without this clarity, readers need to speculate, and those speculations often leaned towards doubt and confusion.
Researchers argue that greater transparency is needed not only as an issue of ethics, but also as a way to maintain confidence in journalism. According to the Association of Expert Journalists' guidelines, journalists are expected to explain their processes and decisions to the public. This expectation should extend to AI use. Like human sources, AI contributions must be clearly cited and explained.
The study, “Who Written It? A Newsreader's AI/Human Byline Sensation,” was written by Steve Bien-Aimé, Mu Wu, Alyssa Appelman, and Haiyan Jia.
