The Washington Post Launches AI-Customized Podcast
The Washington Post has introduced "Your Personal Podcast," an audio offering that utilizes artificial intelligence to create customized content for its users. This initiative integrates algorithmic personalization with portable audio.
Podcast Functionality and Features
The podcast's content is automatically personalized based on a user's reading history of Post articles. Listeners have options to adjust the topic mix of their podcast and change the computer-generated voices of its hosts. The Washington Post has stated on its help page that the podcast is in an early beta phase and is not a traditional editorial podcast.
Bailey Kattleman, head of product and design at the Post, described the product as an "AI-powered audio briefing experience." An upcoming release is planned to enable listeners to interact with the podcast by asking follow-up questions.
Reported Concerns and Official Responses
Reports from Semafor indicated that Post staffers cited errors, including misattribution or invention of quotes and the insertion of commentary that interpreted sources' statements as the paper's own stance. The newspaper's app includes a note advising listeners to verify information against the source material.
The Washington Post Guild, representing newsroom employees, issued a statement to NPR expressing concern regarding the new product and its rollout. The Guild alleged that the podcast undermines the Post's mission and the work of its journalists, questioning why the technology would be held to different standards than traditional journalism's correction practices.
Kattleman stated that the new product is not intended to replace traditional podcasts, noting that they maintain a "unique and enduring role" at the Post. She outlined the technical process: a large language model (LLM) converts a story into an audio script, a second LLM vets the script for accuracy, and then a synthesized voice narrates the episode. Kattleman affirmed that all content is based on Washington Post journalism.
Industry Context and Publisher Motivations
Andrew Deck of Harvard University's Nieman Lab noted that the Post's venture is significant among U.S. legacy publishers, though AI-generated podcasts have precedents. Examples include the BBC's My Club Daily, an AI-generated soccer podcast, and a Swiss public broadcaster's use of voice clones in 2023. News outlets have also long offered automated text-to-speech conversion for articles.
Publishers are exploring AI podcasts due to potential cost-effectiveness, as it may reduce the need for resources and personnel involved in traditional podcast production. Gabriel Soto, senior director of research at Edison Research, suggested that a successful AI virtual podcast could become valuable intellectual property. Deck indicated that if successful, the Post could significantly scale its audio journalism offerings without commensurate labor investment.
Innovation and Audience Reception
Deck identified the level of customization as an innovation, allowing for a podcast tailored to individual listeners beyond what a human team could manually produce. The Post provides voice options, such as "Charlie and Lucy" and "Bert and Ernie." Kattleman noted the aim is to offer flexibility and appeal to younger, on-the-go listeners, acknowledging that audience preferences for AI and journalism vary.
Soto reported that 1 in 5 podcast consumers have listened to an AI-narrated podcast. However, he also stated that many podcast listeners prefer human connection, accepting AI tools for content creation but not for execution or hosting. Deck observed that younger audiences may show familiarity and comfort with algorithmic curation, similar to platforms like TikTok.
Potential Downsides and Concerns
Potential consequences of AI podcasts include job displacement within the voice performance industry, as AI automation could reduce the demand for human voice actors. Concerns have also been raised regarding the potential for AI-driven news personalization to create echo chambers by omitting context or skepticism that a journalist would typically provide. Deck highlighted the issue of generative AI models' propensity to "hallucinate" or produce false information, often with high confidence.
Blurring the distinction between human and AI voices could also impact trust, a critical factor for news organizations. Soto questioned the audience's reaction when expecting human-generated content and encountering AI instead.