Napoleon Dynamite and Why I Am Leaving Substack

Walking away from enshittification to a new home on Ghost

My newsletter statistics started changing last August—at least that is when I noticed. In short, while my subscribed readers kept reading, I was losing what I called accidental tourists—people who found my work through Notes and chats. At first, I blamed myself, reviewing what and when I posted: opening paragraphs, keywords, SEO text, tags—just throwing darts. But as I wrote in a “guest article” by ChatGPT a few weeks ago, I realized Substack was the villain.

Substack had changed how it connects readers to writers through Notes. While the algorithm is inscrutable, the strategy is clear. Substack presents as working hard to be like Netflix, connecting readers to writers not by stated interests, but through textual analysis of what they have just read.

Just like Netflix

You could see this as good—Substack simply searching out more of what you like. But what if you are a Napoleon Dynamite? This very offbeat 2004 film was Netflix’s Achilles heel in the mid-2000s: its algorithm could not reliably predict who would like the film. This was back when Netflix was mailing DVDs, so viewers could wait days to receive a movie they hated—not good for business. Netflix put up a million-dollar prize, ostensibly to improve recommendation accuracy by 10 percent, but the real drivers were outlier films like Napoleon Dynamite that resisted prediction.

Netflix got its 10 percent, but did not solve Napoleon Dynamite. So, Netflix threw in the towel and changed strategy. As The Guardian documents, Netflix’s turn toward what critics call “algorithm movies” reflects a deliberate retreat from cultural risk. Content is engineered to be blandly legible, emotionally predictable, and globally portable, optimized away from serendipity and toward completion rates and churn reduction. Films and series are built from familiar genre components, designed to offend few and satisfy many just enough to keep viewers watching. The algorithm is risk-averse, favoring content that avoids polarization and rewards seamless engagement.

A similar pattern is now showing up on Substack. As the platform leans more heavily on feeds, recommendations, and promotions, it naturally favors writing that generates steady, predictable engagement. Work with a strong, unusual voice or a sharp point of view tends to produce mixed reactions, which makes it harder for the system to promote. Content that feels familiar, easy to place, and broadly appealing travels better because it keeps the metrics moving in the right direction.

In both cases, the solution to the Napoleon Dynamite problem is not a better understanding of user preferences, but avoidance of works that expose the algorithm’s limits—a shift invisible on the screen but decisive in what gets shown and supported. Netflix declares victory by redefinition, not by solving the problem. Goodbye, future Napoleons.

I am Napoleon Dynamite

“I had not expected where your thinking would lead you, but I like where you went. Let us see if it plays out.”
— Peggy Carter, comment on my piece A Turning Point for Marjorie Taylor Greene

My work asks readers to slow down, follow an argument, and sit with ideas that do not resolve immediately or confirm prior beliefs. I mix registers freely—religious framing alongside academic philosophy, political history alongside strategic advice—sometimes four modes in a single piece. I build original conceptual frameworks rather than recycling standard progressive commentary. And I challenge assumptions within progressive worldviews I broadly share, which means my conclusions often surprise even sympathetic readers.

That approach may interest readers who stay with the work, but it confounds Substack’s algorithm. Substack’s recommendation system is designed to anticipate what a reader will find immediately satisfying and move seamlessly from one recommendation to the next. Writing that unfolds through several conceptual turns before arriving at its conclusion does not send clear early signals. The system cannot easily infer whether a given reader will experience the piece as rewarding.

This mirrors the Napoleon Dynamite problem: mixed signals reduce recommendation confidence, regardless of a work’s quality. My subscribers keep opening my emails, but writing that generates sharp, uneven engagement is treated as a risk by a system optimized for predictability. Peggy captured what I always aim for: surprise, intellectual challenge, and useful new perspectives. Those qualities strengthen direct relationships between writers and appreciative readers, but are anathema for an algorithm designed to avoid uncertainty.

Progressive Conjecture is reader-supported.
If this essay was valuable, you may make a one-time contribution here.

No obligation — for readers who want to support independent writing.