The Media Copilot: Building the newsroom AI playbook without turning journalism into slop

This March 2026 episode of The Media Copilot podcast, published on YouTube, features host Pete Pachal and Gina Chua, Executive Editor at Large at Semafor and Executive Director of the Tow-Knight Center for Journalism Futures at the CUNY Graduate School of Journalism. The conversation focuses on practical newsroom AI adoption — what Semafor has actually built and used, not what the industry is hypothetically planning.

Chua speaks from direct experience building internal AI tools at Semafor. The conversation is grounded in specific use cases: a copyediting agent that applies AP Style consistently, tools that search a story draft for relevant datasets, and workflows for summarizing conference transcripts into editorial products. These are operational details, not strategic vision, which makes the episode more useful for editors and reporters who want to understand what adoption looks like inside a publication that has moved past the pilot stage.

Who this video is for

Journalists, editors, and newsroom leaders who are evaluating or currently implementing AI tools in editorial workflows. The conversation assumes some familiarity with how newsrooms operate but does not require technical knowledge. It is also relevant to content strategists outside journalism who want to understand how high-credibility organizations are thinking about AI’s role in quality-sensitive writing contexts.

Key takeaways

1. Chua’s framing distinguishes between AI applied to process tasks and AI applied to editorial decisions. Copy editing, fact-checking support, and data retrieval are process tasks — they have clear success criteria and bounded scope. Deciding what to cover, how to frame a story, or what angle gives a reader the most useful understanding of a situation are editorial decisions that Semafor keeps with human journalists. The boundary between these two categories is where most newsroom AI debates actually happen.

2. A copyediting agent that applies AP Style consistently turns out to be one of the most tractable AI applications in a newsroom. Style guides are explicit, the rules are well-documented, and deviations are easy to verify. This gives AI something to optimize against that does not require the tool to exercise judgment it is not equipped to make.

3. The “slop” risk in the title refers specifically to the failure mode where AI generates plausible-sounding journalism that is vague, unverifiable, or optimized for surface readability rather than accuracy. Chua argues that the protection against this is not avoiding AI, but maintaining clear editorial process: AI assists specific steps, human editors verify the output against the same standards they apply to human-written drafts.

4. Semafor’s conference summarization product — nine thematic takeaways from hours of speaker content — illustrates a case where AI added genuine reader value rather than replacing editorial work. Producing that kind of synthesis at that scale would not have been feasible with a human-only workflow. The AI did not decide what the themes were; editors shaped the framing and verified the claims.

5. Chua is direct about what AI cannot do in a newsroom context: it cannot build source relationships, it cannot identify when a source is being evasive, and it cannot make the judgment calls that distinguish a story worth pursuing from one that is technically correct but editorially unimportant. Newsrooms that lose sight of this distinction are the ones most likely to produce AI-assisted content that erodes reader trust.

Worth watching if you are responsible for editorial decisions in a publication or content organization and you want a concrete, experience-based account of where AI tools are genuinely useful versus where they introduce quality risks that are difficult to detect before publication.