Impact of the Epstein Investigation on Media and AI Reporting

In July 2025, the Justice Department made a controversial decision to refrain from releasing any further files related to its investigation of child sex trafficker Jeffrey Epstein. The announcement triggered a significant backlash, drawing criticism from unexpected segments of the online community.

Right-wing commentators and influencers swiftly voiced their discontent, pointing fingers at President Donald Trump and his administration for not honoring their campaign commitment to release federal documents. Political podcasters who had previously supported Trump were particularly incensed, with prominent figures like Joe Rogan and Andrew Schulz publicly urging the government to reconsider its stance.

The New York Times has closely monitored the escalating dissatisfaction within the GOP base, which culminated in the near-unanimous passage of the Epstein Files Transparency Act by Congress last November. A report generated by artificial intelligence, sent directly to journalists’ email inboxes, proved crucial in the Times’ coverage. Zach Seward, the editorial director for AI initiatives at the Times, indicated that this was one of the initial signs of conservative media’s shift away from the administration. (Seward previously worked as an associate editor at Nieman Lab.)

The Manosphere Report

Internally dubbed the “Manosphere Report,” this AI tool leverages large language models (LLMs) to transcribe and summarize new episodes from a diverse array of podcasts.

“The Manosphere Report provided us with swift clarity indicating that this situation wasn’t well-received within the President’s base,” Seward explained. “We saw a direct correlation between this information and our editorial decisions.”

The term “manosphere” encompasses online communities that adhere to rigid, often patriarchal definitions of masculinity and commonly hold misogynistic and anti-feminist views. These communities frequently intersect with MAGA and far-right social media ecosystems. Following Trump’s reelection, the Times prioritized in-depth coverage of the manosphere.

“Monitoring influencers, particularly conservative young male figures, became essential for adequately covering this administration,” Seward remarked. “The strong interest within the newsroom made automating this process logical.”

Launched a year ago, the Manosphere Report currently tracks approximately 80 carefully selected podcasts covering politics, public health, and internet culture. This includes right-leaning shows such as The Ben Shapiro Show and The Clay Travis & Buck Sexton Show, as well as liberal programs like MeidasTouch, which has a predominantly male audience.

Whenever a podcast airs a new episode, the tool automatically downloads it, transcribes the content, and summarizes the findings. Every 24 hours, these summaries are compiled into a comprehensive overview highlighting shared themes and notable trends. Journalists receive this report via email every morning at 8 a.m. ET, and the Times is exploring how to utilize this system for other reporting areas.

Seward stated that these emails signal evolving sentiments and shifts in rhetoric across the manosphere. Ultimately, it is up to Times journalists to delve deeper into stories related to the leads indicated in the reports.

“We would never solely rely on AI-generated summaries,” Seward clarified. “Reporters actively engage with the actual podcasts, using the report as a tip-off, or motivation to investigate further.”

Case Studies and Broader Applications

For instance, when actress Sydney Sweeney’s advertisement for American Eagle sparked a culture war debate last summer, Times journalists identified, partly through these reports, that right-wing podcast hosts were instrumental in igniting the backlash. Analyzing the commentary revealed that these podcasters had fabricated a narrative of progressive outrage against Sweeney, despite only a small number of posts about the ad on social media at the time.

The Times is not alone in employing LLMs to sift through the vast amounts of audio and video content necessary for comprehensive journalism. Various local news outlets have also utilized LLMs to monitor livestreams of school board and town hall meetings through emailed summaries. Last year, colleague Neel reported on “Roganbot,” a tool developed by AI consulting firm Verso that generates searchable transcripts of The Joe Rogan Experience podcast, highlighting potentially controversial or false statements for verification.

The Manosphere Report was initiated by the Times’ AI Initiatives Team, a small division formed in 2024. While many major newsrooms have explored AI for reader-facing chatbots or article assistance, this team has concentrated on harnessing generative AI for data analysis and investigative reporting. They have also developed additional tools to facilitate basic functions such as transcription and summarization.

“The Manosphere Report evolved from one of our existing tools known as Cheatsheet,” said Seward. This tool was conceived from a script written by Dylan Freedman, a machine learning engineer. Investigative journalist Jesse Drucker approached Freedman with a request to search 10,000 individuals who had registered for a tax incentive in Puerto Rico.

“He said, ‘I simply can’t Google 10,000 names—but a machine could,’” Seward recounted. Freedman leveraged LLMs to automate the process, evaluating the results to pinpoint individuals whose financial histories warranted further investigation. This resulted in a piece published in May 2024 that unveiled the extensive abuse of the tax program.

“That was our first breakthrough,” Seward stated. Following that success, the team began experimenting with LLMs for various complex datasets and file analyses. Currently, many of these functions are housed in a single spreadsheet-based tool. Reporters can upload datasets into Cheatsheet, where they can execute preset scripts and prompts referred to as “recipes.” Some recipes, such as transcribing extensive video footage and summarizing it, form the foundation of the Manosphere Report.

Though still in beta, Cheatsheet has already engaged around 300 users in the newsroom, with 50 being highly active. Seward revealed that at least one new project is initiated in Cheatsheet daily. The tool has been employed to investigate election-interference groups, transcribe and translate Syrian prison records, and find topical instances of Trump’s commentary about January 6. At times, Cheatsheet has facilitated deeper historical analyses of podcasts.

Last spring, the Times conducted an investigation into Dr. Mehmet Oz’s medical claims while a television personality after his appointment by Trump to lead the Centers for Medicare and Medicaid Services. Through Cheatsheet, reporters analyzed statements made by Oz across 2,500 media appearances, episodes of The Dr. Oz Show, and social media posts, revealing his financial ties to products he had promoted, including those lacking scientific evidence for health benefits.

In February, Seward confirmed that Cheatsheet will be made available to every journalist in the Times newsroom. Staff will receive optional training sessions this year to enhance their skills in utilizing the tool.

Conclusion

The approach adopted by the Times demonstrates a forward-thinking strategy in combining traditional journalism with cutting-edge AI technology. Rather than viewing AI solely as a tool for content creation, it is being harnessed to enhance investigative capabilities. This evolution reflects the Times’ commitment to staying ahead in the fast-paced media landscape, ensuring they have the resources necessary to deliver comprehensive reporting.