The AI Slop Problem Nobody's Talking About | Substack CEO Interview

Overview

Chris Best (co-founder/CEO of Substack) explains why “AI slop” threatens the value of the public square—and describes Substack’s initial response: transparency tools rather than policing.

What “slop” is and why it matters

• Best defines “slop” broadly as content people don’t truly believe in—ranging from:
  • spam and clickbait
  • to generic, “soulless” material
• Over the past year, he’s increasingly felt that much online writing (including “deep work” presentations) appears:
  • copied, or copied-and-pasted
  • “AI-smelled”
  • or generated with little genuine intent
• A key reference point is Pangram’s AI-detection tooling, which (per its published stats) reportedly found very high proportions of likely AI-generated long-form writing on platforms like LinkedIn (Best cites ~40% for long-form there).
  • He stresses the stat isn’t perfect, but emphasizes that the direction of travel matters.
• The main harm isn’t only low quality—it’s erosion of trust:
  • Readers can’t tell whether writing is real or engineered.
  • They explore fewer new writers.
  • They engage less in comments.
  • Meaningful discourse declines.

Not anti-AI—anti-cynical, anti-intent-free use

Best emphasizes Substack’s pro-freedom stance: creators should be able to use tools.

But he draws a distinction between:

• Using AI as a paintbrush to express one’s own intent and thoughtfulness
• Gaming the system by generating large volumes of convincing output with no real belief or responsibility
  • (he references mass-generating viral posts “with no mistakes”)

He characterizes extreme “no-belief mass generation” as a kind of denial-of-service attack on the public square:

• it forces readers to constantly question authenticity
• it makes discovery harder

How Substack plans to respond: transparency via Pangram

Best argues Substack doesn’t have “all the answers,” and that the pace of change means waiting passively isn’t enough.

Substack’s first step is to integrate Pangram into the Substack app:

• Readers can scan long-form text to view an AI-generation likelihood estimate
• Best clarifies that Pangram-style results don’t judge whether a piece was:
  • “carefully crafted”
  • thoughtful or intentional
• Instead, it indicates whether it likely passed through an LLM during generation

The aim is transparency so that:

• readers and writers can interpret the signal themselves
• people can discuss what responsible use should look like
• the approach avoids restricting tools

Best frames this as creating space for cultural norms around “responsible AI use,” including the belief that responsible spaces should stand up to transparency.

Pangram’s concept: surfacing AI’s “style peaks”

Best says Pangram is compelling because LLMs tend to produce language with recognizable statistical patterns—a “peak” around how they express language.

He also notes that models can be fine-tuned toward different “human-ness,” which may reduce detection scores. This reinforces that the goal isn’t perfect certainty—it’s public visibility and conversation.

He connects this to his own writing workflow:

• He often starts with a raw “whisper transcript” to capture his voice.
• Then he iteratively works with LLMs to:
  • preserve clarity
  • remove typical LLM dramatization
• He distinguishes two modes:
  1. when the thesis is already clear
  2. when he’s still thinking through the idea—using AI more like a co-thinking tool across many drafts

Beyond text: measuring “ideas” and diversity, not just authorship

Best argues the deeper problem is that AI tends to pull ideas toward an “averaged central distribution.”

For example, AI may insert ethics/governance sections into AI-related documents even when not requested.

He proposes a “dream” tool: a Pangram-like approach for ideas, not just text—mapping:

• the diversity of human ideation
• versus the constrained distribution LLMs gravitate toward

He emphasizes that “alpha” (value/originality) often lives at the edges of distributions—where AI is less reliable—and that human creativity should guide models rather than be replaced by them.

The lasting value of human conversation

Best argues that even with superhuman AI, people still want:

• human connection
• shared reactions
• community discussion (comments, ripples, and “third perspectives”)

He also highlights the “stickiness” of great writing:

• it sparks ongoing conversations
• it changes how people see the world

He cites examples of Substack posts that drove discourse (e.g., his Open Engine discussion about memory). He also suggests rewarding “the opposite of slop” via signals that highlight:

• genuinely net-new
• coherent
• conviction-producing ideas

He references writers like Paul Graham and Naval as examples whose posts “stick” and change readers’ thinking.

Additional concern: video may face similar trust/betrayal issues

Best worries that if AI-generated video becomes indistinguishable from real human effort, norms will shift again—mirroring the writing authenticity dilemma.

Key risk: mismatch of expectations

• If an “interaction” is actually an AI avatar or synthetic video while the audience believed it was human, it can feel like betrayal (like catfishing).

He suggests new cultural norms will need to define:

• when AI video use is acceptable
• how transparency should work

What Substack should reward: “featured discourse”

Best proposes that Substack highlight posts/topics that are advancing discourse—an “anti-Pangram” idea:

• not “is this AI?”
• but “is this genuinely excellent, spiky, variance-rich, and discussion-sparking?”

He frames this as building a positive loop:

• thoughtful long-form work stays discoverable and trusted
• low-effort clickbait doesn’t drown out creators
• regardless of whether that clickbait is AI-assisted or not

Closing hope and call to action

Best hopes Substack’s transparency tool will spark broader conversation about anti-slop norms:

• how to defend human, thoughtful authorship
• what “value” should mean

He ends with a light promotional note encouraging viewers to subscribe to his Substack.

Presenters / contributors

• Chris Best (Co-founder and CEO, Substack)