AI & the future of media with The Atlantic CEO, Nicholas Thompson

Unknown Source July 24, 2025 52 min

artificial-intelligence generative-ai startup ai-infrastructure investment google openai apple

90 Companies

112 Key Quotes

5 Topics

🎯 Summary

Podcast Summary: The Four Horsemen Challenging Journalism in the Age of AI

This episode of the Exponential View podcast, featuring Azim Azal (Host) and Nick Thompson (CEO of Atlantic Media), provides a comprehensive analysis of the existential challenges facing the modern media and journalism industry, framed around “Four Horsemen.” The discussion moves from immediate traffic crises to long-term structural shifts in authority and the disruptive potential of Artificial Intelligence.

1. Main Narrative Arc and Key Discussion Points

The conversation centers on identifying and prioritizing the four major threats (or opportunities) to journalism’s business model: the Sovereign Creator, the Collapse of Search Traffic, the Weakening of Trust/Authority, and AI itself. Thompson ranks the collapse of search traffic as the most pressing existential threat, while acknowledging the decline in trust as the most important underlying issue. The latter half of the discussion pivots to the complex relationship between publishers and LLMs, focusing heavily on fair compensation models for training data.

2. Major Topics, Themes, and Subject Areas Covered

Media Business Models: The shift from advertising-supported attention to subscription-based conviction.
Traffic Dynamics: The decline of traditional referral sources (Search, Social) and the unexpected rise of generative AI (ChatGPT) as a referral source for some publishers.
Creator Economy: The power shift from institutional brands to individual, sovereign creators (e.g., Substack).
AI and Content Licensing: The ethics, economics, and leverage involved in LLMs training on copyrighted journalistic material.
Journalistic Quality: The trade-off between unfiltered individual expression and the quality derived from professional editing.

3. Technical Concepts, Methodologies, or Frameworks Discussed

LLM Crawling Intensity: The observation that LLM bots crawl media sites orders of magnitude more frequently than traditional search engines, imposing significant, uncompensated load.
Robots.txt: Mentioned as the traditional, often ineffective, method for blocking AI crawlers.
Cloudflare’s LLM Jail: A technical solution allowing publishers to actively track and block specific AI bots (like Perplexity, ClaudeBot) unless a value exchange is established.
Pro Rata Model (Bill Gross/Pro Rata): A proposed economic framework where LLMs attribute content contribution to their answers and distribute a percentage of their revenue proportionally to the original content creators.

4. Business Implications and Strategic Insights

Existential Threat of Search Decline: For The Atlantic, a 20% drop in search traffic (which accounts for 40% of total traffic) represents an 8% overall traffic hit, directly impacting ad revenue and subscription funnels.
Creator Retention Strategy: Traditional media must incentivize their top writers to stay by offering competitive financial rewards and the cultural/editorial support that institutions provide, countering the direct relationship and high earning potential offered by platforms like Substack.
Leverage in AI Negotiations: The ability to block LLM crawlers (via tools like Cloudflare’s) is crucial for gaining leverage in negotiations for fair compensation or traffic exchange.

5. Key Personalities, Experts, or Thought Leaders Mentioned

Nick Thompson (CEO, Atlantic Media): Primary expert offering the publisher’s perspective.
Azim Azal (Host, Exponential View): Provided data showing ChatGPT as his third-highest referral source.
Matthew Prince (Cloudflare): Mentioned for developing tools to manage and monetize LLM crawling activity.
Martin Gurri: Referenced for chronicling the weakening of institutional trust in media.
Bill Gross (Go2.com/Pro Rata): Highlighted as the originator of the cost-per-click model and the proponent of the pro rata licensing model for AI training data.

6. Predictions, Trends, or Future-Looking Statements

Fair Exchange is Inevitable: Thompson predicts that the current unfair extraction of data by LLMs will resolve into an economic exchange, likely involving a settlement percentage between 3% and 30% of derived value, similar to historical music industry royalty negotiations.
The Pro Rata Model’s Potential: The pro rata model, which links revenue sharing directly to content contribution in AI answers, is seen as the “best model” for long-term sustainability, provided the licensing entity (like Pro Rata) can sit in the necessary traffic choke point.

7. Practical Applications and Real-World Examples

The Atlantic’s Response: The Atlantic has proactively studied search traffic decline, sued an AI company, and made deals (notably with OpenAI) to secure value.
Substack’s Growth: Substack has scaled rapidly, now boasting five million paid subscriptions, demonstrating the market appetite for direct creator relationships.
Social Traffic Volatility: Social referral traffic is highly variable; Facebook traffic recently increased for The Atlantic, while X (Twitter) traffic has plummeted due to platform changes favoring internal video consumption over external links.

8. Controversies, Challenges, or Problems Highlighted

The “Original Sin”: LLMs built their foundational models by taking vast amounts of content without compensation, creating a massive liability/obligation that they are now reluctant to settle due to the difficulty of quantifying that initial value.
Trust Deficit: The challenge of maintaining authority when partisan actors can easily dismiss high-quality journalism as “false” to receptive audiences.
The Editing Trade-off: While most writers benefit significantly from professional

🏢 Companies Mentioned

The Guardian ✅ media

ClaudeBot ✅ tech

Google Discover ✅ tech

Google News ✅ tech

LinkedIn ✅ tech/social

Bing ✅ tech

Huawei ✅ tech

Applebot ✅ tech

United States Postal Service ✅ unknown

The Guardian ✅ unknown

New York ✅ unknown

And Substack ✅ unknown

The Free Press ✅ unknown

Bari Weiss ✅ unknown

Heather Cox Richardson ✅ unknown

💬 Key Insights

"Fact-checking is top-less, right? In part, like, this date problem, right? Because it doesn't understand—it has a hard time parsing, like, yeah, McCormick was—there are more stories that were written about McCormick when he was a Senate candidate than when he was a Senator."

Impact Score: 10

"He is a sitting senator since January, and human fact-checkers are there, and our LLM fact-checkers—and the LLM fact-checker didn't spot it because it thought that the big piece was, you know, the $80 billion..."

Impact Score: 10

"And so if some people actually say this will happen with AI—AI, right? Like, what could possibly bridge not just American Republicans and American Democrats, but like, what could bridge like nine states in China? Well, maybe if there's like an external catastrophe coming from AI."

Impact Score: 10

"I found it extremely difficult to find that balance because... if I am critical of something that Elon Musk might have done, the inbox fills up. And if I then write about something that I think he's done that's really impressive, SpaceX, the inbox fills up, and it's not the same people..."

Impact Score: 10

"Please read this essay and write kind of the most obnoxious but intelligent thing you can, right? And then that gets you a guide to like, 'Oh, wait, let me harden this argument or let me rephrase it.'"

Impact Score: 10

"My favorite example is I had a reporter; in my previous job as the editor of *Wired*, I had a reporter who was recruited by a venture capital firm... 'maybe like 30% of your stories will be positive stories of our portfolio companies, and you won't disclose it. You'll just go out and write stories about that.' And that—that's corrupt, right?"

Impact Score: 10

📊 Topics

#artificialintelligence 89 #generativeai 21 #startup 9 #aiinfrastructure 3 #investment 1