TLDR: A study analyzing over 40,000 news articles found a significant increase in AI-generated content since late 2022, particularly in local and college news outlets. AI is often used for article introductions, boosts word richness and readability, but reduces formality and descriptive detail, leading to more uniform writing styles. Major news outlets show less AI adoption, possibly due to stricter policies.
The landscape of news production is undergoing a significant transformation with the increasing integration of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs). A recent study titled “Echoes of Automation: The Increasing Use of LLMs in Newsmaking” by Abolfazl Ansari, Delvin Ce Zhang, Nafis Irtiza Tripto, and Dongwon Lee, delves into the extent and impact of AI-generated content across various news media. This research sheds light on how AI is subtly reshaping journalistic practices and the content we consume.
The study analyzed over 40,000 news articles published between early 2020 and late 2024, sourced from major, local, and college news outlets. To identify AI-generated text, the researchers employed three advanced AI-text detectors: Binoculars, Fast-DetectGPT, and GPTZero. A key aspect of their methodology involved comparing content from the “pre-GPT era” (before November 2022, when ChatGPT-3.5 was publicly released) with the “post-GPT era,” providing a clear temporal perspective on AI adoption.
The Growing Presence of AI in News
A striking finding from the research is the substantial increase in GenAI use in news articles, especially after the public release of ChatGPT-3.5. This surge was most pronounced in local and college news agencies. Local news, for instance, experienced a remarkable tenfold increase in fully AI-generated articles compared to the pre-GPT era. College newspapers also saw a significant rise, though to a slightly lesser extent. In contrast, major news agencies appeared to be the least affected, which the researchers suggest might be due to stricter editorial oversight and explicit organizational policies regarding AI use.
When examining different media formats, the study found that text-centric modalities like newspapers and radio outlets showed the largest increase in AI-written content. While broadcast and TV content also displayed upward trends, they were less significant. Interestingly, some AI-flagged instances in broadcast and radio datasets were identified as machine-generated templates, such as weather forecasts or subscription advertisements, indicating an existing level of automation beyond just GenAI tools.
A fascinating aspect of the analysis involved a sentence-level examination of AI-generated opinion articles. The study revealed that authors tend to rely on LLMs more heavily for the initial sections of their articles. The probability of AI generation was highest in the first 40% of sentences and gradually decreased towards the end, with the final 20% showing the lowest average generation probabilities. This suggests that while AI might help kickstart the writing process, human authors are more likely to craft the conclusions themselves.
Linguistic Shifts Caused by AI
The research also explored the linguistic impact of GenAI on news content, comparing stylistic and lexical features between AI-generated and human-written texts. The findings indicate that AI use alters several linguistic patterns, often enhancing certain qualities, particularly in smaller, under-resourced news categories.
For example, AI-generated content showed an increase in word richness and readability. While human-authored content from major news agencies typically had greater word richness than local outlets, the use of generative models significantly narrowed this gap. However, AI-generated texts also tended to be less formal and exhibited reduced descriptive details, particularly in named entities like persons and nationalities. On a lexical level, AI-generated content showed a notable increase in the use of modifiers (adjectives, adverbs), functional words (adpositions, determiners, auxiliaries), and structural elements (nouns, punctuation). These stylistic and linguistic shifts were more evident in smaller media outlets, such as local news agencies, where the disparity between human-written and AI-generated articles was more pronounced.
Also Read:
- Study Reveals Large Language Models Exhibit Bias Towards AI-Generated Content, Raising Concerns for Human Contributions
- Educators and Experts Deliberate on the Evolving Role of Artificial Intelligence in Classrooms
Looking Ahead
The study provides compelling empirical evidence of the growing presence of AI-modified content in news agencies and media outlets. The trends observed, particularly the surge in AI use since late 2022, highlight the increasing role of GenAI in news production. As a Gen AI news and article publisher, understanding these dynamics is crucial. The findings underscore the importance of developing responsible integration policies that prioritize transparency, preserve editorial identity, and leverage AI’s strengths without compromising content quality, truthfulness, and media diversity. For more details, you can read the full research paper here.
