AI Generates 9% of US News Articles
About 9% of 186,000 articles from 1,500 US newspapers are AI-generated, with higher rates at small outlets covering weather and tech. Opinion pieces are 6.4x more likely to be AI-written, while tools like BART-large models are being used to summarize news and draft breaking stories from reporter notes.
The use of AI in newsrooms has a history stretching back nearly a decade. Bloomberg was an early adopter with its "Cyborg" program for financial reports, and The Washington Post utilized its own AI, "Heliograf," to cover the 2016 Rio Olympics. The Associated Press now uses AI to automatically generate roughly 40,000 stories a year, particularly for corporate earnings reports. Globally, the adoption of AI in news is widespread, with 76% of news organizations experimenting with or actively using AI tools as of 2023. This integration is projected to deepen significantly, with a 2024 forecast from the Reuters Institute predicting that 85% of newsrooms will have fully integrated AI by 2027. Beyond article generation, the most frequent applications of AI in journalism are for language processing tasks such as transcription, translation, and copy-editing. News organizations also use AI to analyze audience data, personalize news feeds, and moderate comments. Semafor, in partnership with Microsoft and OpenAI, built a tool called Signals to help its journalists find and research stories from sources in multiple languages. The "BART-large" model is a specific type of AI architecture developed by Meta AI that excels at text summarization. It is a sequence-to-sequence model that combines a bidirectional encoder like BERT with an autoregressive decoder like GPT. To improve its performance on news content, it was specifically fine-tuned on the CNN/DailyMail dataset, which consists of news articles paired with human-written summaries. Despite its capabilities, AI-generated content has demonstrated a 14% higher error rate in fact-checking tests compared to articles written by humans. This aligns with audience perception, as a 2024 Pew Research survey found 61% of people found AI-written articles to be less trustworthy. As a result, many outlets like Germany's EXPRESS.de require that every piece of content from their AI "digital colleague" be reviewed by a human editor. Concerns over AI's impact on journalism jobs are significant, with one 2023 study predicting AI automation could displace 20,000 journalism roles by 2025. Some journalists also worry about the technology's potential to amplify bias and the lack of disclosure policies at many organizations. A survey of UK journalists revealed high levels of concern about AI's potential impact on public trust and the originality of journalistic work. Major news outlets are now building their own proprietary AI systems. The New York Times has rolled out an internal tool called "Echo" and is training journalists on its use. The Washington Post created "Ask the Post AI," an experimental chatbot trained on the paper's archives that can answer reader questions on topics like climate change.
