Substack co-founder Chris Best on AI slop, detection, and what still counts as thinking.

Source: Nate’s Newsletter Date: 2026-07-22 URL: https://natesnewsletter.substack.com/p/ai-detection-ideas-not-words

Summary

Substack co-founder Chris Best argues that AI-detection tools like Pangram measure surface-level word patterns, not whether genuine thought produced a piece of writing — a human can type a borrowed idea with zero originality, while AI-assisted writing can still carry a considered, novel perspective. He proposes an “Ideas Graph” concept as a better signal: mapping conceptual relationships and novelty rather than scanning for AI-typical phrasing.

Implications

This feeds the agent-layer convergence and AI governance/policy threads: as AI writing tools become ambient, the locus of trust shifts from “was AI used” to “did a publisher take responsibility for what they put out.” It’s a preview of the accountability framing likely to shape platform policy and detection-tool marketing (Pangram, GPTZero, etc.) as binary AI/human classifiers keep proving unreliable.