Podcast Summary

The Digiday Podcast

Episode: Why The Guardian’s first reader-facing AI product isn’t a chatbot
Date: March 31, 2026
Host(s): Kamika McCoy (Senior Marketing Reporter), Tim Peterson (Executive Editor, Video and Audio)
Guest: Chris Moran (Head of Editorial Innovation, The Guardian)

Overview: Episode Theme and Purpose

This episode explores why The Guardian’s first reader-facing AI product is not a chatbot, but instead a feature called "Storylines." The discussion delves deeply into the reasoning, guiding principles, product development journey, editorial philosophy, and practical implementation of this AI-driven module. Central to the episode is a look at the Guardian’s deliberate, principles-led approach to integrating AI into its products, ensuring alignment with journalistic values and the needs of both staff and readers.

Key Discussion Points and Insights

1. The Evolution of AI Use at The Guardian

• Early Consideration (04:04)

  • The Guardian began thinking about AI products soon after ChatGPT launched.
  • The process started with establishing guiding principles before developing any products.

• "Principles First" Approach (04:26)

  • Three foundational principles:
    1. Benefit readers
    2. Benefit staff and support the mission
    3. Respect copyright and content creators
  • “We went principles first, first of all, and I think that matters to us more than anything else.” — Chris Moran (04:04)

• Cautious Adoption (06:01)

  • Focus on understanding external threats and protecting the Guardian’s value.
  • Licensing and copyright concerns led to a policy of "grounding," not "training" LLMs on Guardian content.

2. Internal vs. Reader-Facing AI Prototyping

• Chatbot Experimentation (07:21–09:46)

  • Initial focus was on internal AI chatbots for staff, not readers.
  • The chatbot could summarize articles and assist journalists, but there were reservations about releasing this to the public:
    • Risk of inaccurate summaries
    • Loss of editorial accountability
    • "If you do point chatgpt or another LLM at a Guardian archive is what it spits out. Guardian journalism, ultimately, I don't think it is." — Chris Moran (09:10)

• Editorial Philosophy: Static vs. “Liquid Content” (09:55)

  • Importance of static, consistent content for readers to create a shared experience and maintain a Guardian point of view.
  • “When we produce an article, everybody who comes to the Guardian sees the same thing... that’s...the beginning of a whole other series of things that fall out of that, like community.” — Chris Moran (09:55)

3. The Genesis of "Storylines"

• Moving Toward Reader-Facing Tools (13:00–14:02)

  • Sparked by a Google workshop and funding from the Google News Initiative.
  • The idea: Improve "tag pages" (topic pages) to organize and contextualize Guardian’s vast archives.
  • Existing tag pages were just “walls” of reverse chronological articles and failed to realize their potential as curated, contextual destinations.

• Technical Approach (15:37–17:52)

  • The AI analyzes the most recent 200 articles for a topic and surfaces three main “storylines” as narrative threads.
  • Only the three storyline titles are machine-generated; summaries and selections are curated from existing editorial material.
  • Selected four most useful news pieces for each storyline, with additional recommendations for opinions, deep reads, multimedia, etc.
  • "What I think the AI enables us to do in a way that is automated that we couldn't do manually, is to decode this page by using the idea of narratives." — Chris Moran (15:37)

4. Philosophy: Not Another AI Summary Button

• Rejection of AI Summaries as a Destination (18:16)
  • Many publishers offer AI-generated summaries, but The Guardian sees limited value, arguing that their article structure already summarizes key points at multiple levels.
  • “Offering a reader an opt-in summary on top of all of those three things...we never felt the need to really pursue.” — Chris Moran (18:16)

5. Editorial Oversight and Guardrails

• Curatorial and Editorial Control (19:32–21:28)

  • LLM only sees headlines and trail text, not body copy, to ensure recommendations reflect editorial intent.
  • AI doesn't parse the full article, reducing risk of error or irrelevant associations (e.g., ICE as Immigration, not “ice” the substance).

• Evaluation and Feedback Loop (22:19)

  • 20 senior editors reviewed AI storylines and recommendations, providing detailed feedback on accuracy, language use, and context.
  • Adjustments made in prompts and design—e.g., reducing the number of storylines from four to three to limit AI "reaching" for tenuous connections.

• Risk Management (25:40–27:38)

  • Strict A/B testing on only 10 topic/tag pages
  • Quick “red button” kill-switch for any problematic results
  • Avoidance of high-risk topic types (e.g., “rogues galleries” of individuals)
  • Storylines are generated for all users, not personalized, and undergo regular editorial review

Notable Quotes & Memorable Moments

• On Foundational Principles (04:24):
  "Anything we do should be for the benefit of our readers … for the benefit of our staff and our mission … with respect to copyright and content creators."
  — Chris Moran

• On Past Hype Cycles (08:01):
  "What hangs across the top of all of this … is the pivot to video. … The idea that what the Guardian offers right now is valueless in the face of [AI], I think is wrong."
  — Chris Moran

• On Editorial Accountability (09:10):
  "Just because it's pointing at Guardian journalism doesn't mean it's going to be accurate … If you do point chatgpt or another LLM at a Guardian archive is what it spits out. Guardian journalism, ultimately, I don't think it is."
  — Chris Moran

• About Tag Pages and Their Opportunity (14:02):
  “What excited me about those pages was this could be a way of leveraging our archive and actually showing it to human beings in a way that was useful.”
  — Chris Moran

• Explaining How Storylines Works (15:37):
  “The technology … generates from a list of the most recent 200 articles on this tag what it thinks the three big storylines are right across those articles.”
  — Chris Moran

• On Limiting AI’s Scope (21:10):
  “We are not showing the LLM, the articles, the body copy at all. We’re only showing it the headlines and the trails. And that’s very, very intentional.”
  — Chris Moran

• On Human Editorial Involvement (22:19):
  “A significant portion of the work that went into this was rooted in an evaluation period, with 20 of our most senior editors evaluating multiple instances of these.”
  — Chris Moran

• On Managing Editorial Risk (25:55):
  “Part of this is us understanding how comfortable we can get with some level of risk and whether or not we can guardrail it enough.”
  — Chris Moran

Timestamps for Key Segments

• 00:41 – Publisher perspectives on AI at industry events
• 03:38 – The Guardian’s principles-first approach to AI
• 06:01 – Copyright, licensing, and AI’s legal challenges
• 07:21 – Internal AI product experimentation
• 09:55 – Editorial philosophy: static vs. dynamic content
• 13:00 – Transition to reader-facing AI products
• 15:37 – Live walkthrough: Storylines feature and its workings
• 18:16 – Why not just AI summaries?
• 19:32 – Editorial oversight in machine learning applications
• 22:19 – Human editor feedback and iteration on storylines
• 25:40 – Testing, risk, and controls in deployment
• 27:36 – Regular reviews, scaling approach, and risk avoidance

Episode Tone and Concluding Thoughts

The episode is thoughtful, measured, and at times self-deprecating, illustrating the Guardian’s cautious optimism and pragmatism about AI. There is a strong undercurrent of editorial integrity, transparency, and a willingness to learn from past industry mistakes. Both hosts and the guest emphasize that while the AI-powered storylines product is not revolutionary on its own, the process and the principles behind it set a foundation for future, values-driven AI innovation in journalism.