Do People Want to Get Their News from Chatbots? | by Nick Hagar ...
The existential threat that large language models (LLMs) seem to pose for news organizations is hard to ignore. If chatbots can answer user news-related queries directly — without sending them to news websites — they could crater what remains of outlets’ search traffic. While LLMs aren’t yet reliable sources of news and information (hallucinations are still an issue), products like SearchGPT and Google’s AI Overviews are gradually paving the way for this shift.
The recent surge in news publisher lawsuits, licensing deals, and scraping restrictions suggests that the industry is bracing for a paradigm shift that further weakens newsrooms relying on digital engagement to survive.
Do People Want to Get Their News from Chatbots?
But while tech companies have a clear incentive to push this transformation, it’s far less obvious whether news audiences actually want to use AI chatbots as a news source. Do people really ask OpenAI’s ChatGPT for the latest election headlines? Are they using Google’s Gemini to check scores from last night’s NBA games? In the real world, are LLMs truly becoming a go-to source for reading news? Much of the impact of these chat interfaces on news and journalism will ultimately hinge on whether there is an appetite from users and audiences.
The analysis uncovered that users are engaging in complex tasks like summarization, translation, and analysis when interacting with LLMs for news content. Some users incorporate news articles into broader conversations, turning AI into a medium for “conversational news.” These behaviors point to ways that LLMs can assist readers in processing, interpreting, and personalizing news content to meet their individual needs.
Overall, the study shed light on the evolving role of LLMs in news consumption and the potential for AI chatbots to enhance the news experience for users.
Data Analysis on News Queries
To explore this question of how (and whether) people use LLMs for news, data from Wildchat, a dataset collected by researchers at the Allen Institute for Artificial Intelligence, was analyzed. The dataset contains interactions between users from various countries and OpenAI models GPT-3.5 and GPT-4.
When journalists have analyzed the Wildchat dataset, they found a variety of conversations. Users engaged in topics ranging from sex and homework help to journalism queries and article drafting. The dataset revealed a wide range of use cases for ChatGPT, highlighting diverse user behaviors.
Identifying News Queries
Manual annotation and targeted searches were conducted to identify news queries from the dataset. The analysis revealed that news queries are rare, accounting for a small percentage of interactions. However, the diversity within the identified news queries was significant, showing various ways users approach LLMs for news-related information.
