Gyrusprint: The Measure Substack Couldn't Build

The views expressed here are solely my own. They do not represent the opinions, positions, or policies of any current or former employer, client, or affiliated organization.

Here's a question worth sitting with before you read another word: what if the tool built to catch AI fakery admits, in its own launch announcement, that it can't do the one thing it claims to do?

That's not a hypothetical. It's exactly what happened.

On July 21, 2026, Substack announced a partnership with Pangram to catch "Claudefishing"—passing off machine-written text as human. Yet, in the same announcement, they admitted the tool cannot measure whether genuine care or judgment shaped the words. This is the classic "tool-first" trap: deploying technology to solve a structural problem it is fundamentally blind to. By attempting to measure typing instead of thinking, they built a filter that penalizes the very writers they set out to protect. We must stop measuring the vehicle and start measuring the driver.

Substack CEO Chris Best's announcement makes two incompatible claims.

First, he notes that using AI does not guarantee low quality, nor does human writing guarantee high quality. Second, he offers Pangram as the diagnostic tool to separate the two.

But the tool only flags whether a machine produced the character string. It cannot detect the presence of an active mind.

This is not a minor oversight; it is a fatal structural flaw. You cannot build a diagnostic system for human judgment using a classifier that only counts statistical word patterns.

Substack identified a real problem, named it, and then shipped a feature that disqualified itself from solving it.

Consider two distinct workflows producing the same 800-word footprint.

Writer A uses a rigid, two-year-old template. She swaps in new nouns and statistics but contributes zero original thought.

Writer B collaborates with a generative model for three hours. She rejects four drafts, rewrites the conclusion to preserve the core thesis, and manually verifies every data point against primary sources.

An automated detector like Pangram will pass Writer A and flag Writer B.

The system rewards mindless manual execution and penalizes high-order curation. It measures the physical act of typing, not the intellectual act of decision-making.

Philosophy of technology has already mapped this distinction.

In a 2026 paper published in Philosophy & Technology, researcher Tiegue Vieira Rodrigues argues that AI-assisted writing does not eliminate authorship. Instead, it relocates it.

Rodrigues introduces the concept of the "apt curator," drawing on the work of philosopher Ernest Sosa. An act of curation is "apt" when its success stems from the human’s skill and judgment, not from mechanical accident.

Similarly, Luciano Floridi has written about "distant writing" in a 2025 paper for SSRN. The human's role shifts from a typist to an architect of possibilities.

When researchers at Northeastern and Stony Brook built a taxonomy of "AI slop" in 2025, they discovered that slop judgments consistently track coherence and relevance—not the technical origin of the words.

Without a human who cares in the loop, machine output can only resemble quality by accident. But with an active human architect, the technology is merely infrastructure.

The panic over "Claudefishing" assumes that information overload is a modern, model-driven crisis. It is not.

The economics of attention—the reality that an overabundance of information creates a poverty of attention—was formally named by Herbert Simon in 1971. He was diagnosing executives drowning in physical memos long before the internet.

Historian Ann Blair traced this identical complaint back to the mid-1500s. Less century after the printing press, scholars lamented that too many books were being published, most of them worthless.

In 2014, marketing consultant Mark Schaefer named this "content shock."

The root cause is never the tool; it is the human decision to prioritize volume over value. AI did not create the lazy publisher. It just accelerated their output.

By treating "AI involvement" as a proxy for low quality, Substack's strategy runs counter to empirical data.

A 2023 study in Science by Shakked Noy and Whitney Zhang demonstrated that ChatGPT reduced writing time for college-educated professionals by 40% while increasing output quality by 18%.

A separate 2025 study in the Quarterly Journal of Economics by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond analyzed over 5,000 customer-support agents. They found that AI assistance leveled the field, helping less experienced workers improve both speed and quality.

For writers working in a second language, AI is an equalizer. It bridges the gap between those who have deep ideas and those who happen to possess native fluency.

Furthermore, Substack’s proposed "How I made this" disclosure badge is a psychological trap.

Research published in Organizational Behavior and Human Decision Processes by Oliver Schilke and Martin Reimann found that disclosing AI use consistently erodes perceived trustworthiness, regardless of how the disclosure is framed.

Substack built an environment that actively penalizes transparency.

If statistical origin is the wrong metric, we need a name for what actually matters.

I call it Gyrusprint.

A gyrus is a physical ridge on the surface of the human brain. The pattern of these folds is entirely unique—more unique than a fingerprint. Not even identical twins share the same configuration.

A gyrusprint is the intellectual trace left in a piece of writing. It is the footprint of human judgment.

When you return to a writer week after week, you do not do so because you have verified they used a mechanical keyboard. You return because you trust their gyrusprint.

If Substack wanted to build a real diagnostic system, they would abandon statistical classifiers and look for the five structural signals of a gyrusprint:

These signals cannot be measured in a single, isolated post. A gyrusprint is a pattern that emerges over time.

The most efficient, lowest-cost, and most durable diagnostic tool has always been the audience itself.

Quality is adjudicated socially. Readers subscribe, stay, or leave based on the value they receive. This mechanism does not require a partnership with an external software vendor. It requires an audience with the freedom to choose—which was the exact operating model Substack was built on in the first place.

If you want to stabilize your content operations, stop looking for software to police your writers. Focus on the integrity of the system beneath the platform.

The platform is infrastructure. The gyrusprint is the product.

Juan Carlos Vásquez has spent ten years inside enterprise content operations, repairing content supply chains before scaling them. Fix to Flow is the discipline that work produced. The views here are his own.