Chris BestCo-founder & CEO

AI slop consists of generic, clickbait, and soulless content that nobody truly believes in. Substack exists to act as the foil to this trend, offering a platform where writers can create meaningful work they care about and earn a living doing so.

Over the past year, many people have felt inundated by automated content online. I noticed a growing feeling of frustration as a consumer, which led me to talk to Pangram Labs and discover that a large portion of professional writing is now machine-generated.

Pangram's measurements showed that approximately forty percent of long-form writing on LinkedIn was fully generated by artificial intelligence. While not all AI writing is low quality, this massive shift indicates that platforms must start offering transparency options.

When the internet is flooded with automated text, readers have to constantly wonder if what they are reading is real. This makes it incredibly difficult to find new writers or discover original ideas, effectively shutting down constructive public discourse.

I have no interest in policing the specific tools that creators use. The line is crossed when someone uses AI to game systems and generate thousands of posts without reviewing them, skipping the step of actually caring about or understanding the content.

We integrated Pangram into the Substack app to allow readers to scan long-form text and check if it passed through an LLM. Writers can also add a statement explaining their creative process, encouraging open conversation rather than implementing rigid bans.

Claudefishing is a catfishing analogy where someone uses AI to respond to comments or write articles while presenting themselves as entirely human. It feels like a betrayal because it violates the expectation of a mutual investment of human attention.

Pangram works by identifying specific statistical distributions and word choices that large language models naturally favor. While it can be bypassed by advanced prompt engineering, it provides a transparent signal that helps writers and readers discuss authorship.

I record a ten to fifteen minute voice transcript to get all my initial thoughts and personal style on the screen. The large language model takes this raw context and cleans up my verbal ticks while keeping the core thesis bold and intact.

Large language models are cautious by nature and tend to water down original theses. I have to push the model to respect my original vision rather than smoothing it out, ensuring that the final draft remains as sharp and distinct as my thoughts.

I use the model as a tennis wall to bounce ideas back and forth. I tell the AI that the thesis is not finished and list the directions I want to avoid. We iterate through seventeen drafts, using the tool to accelerate my thinking.

Reinforcement learning and safety training create a gravitational sync around certain concepts. For example, if you ask an AI to write about artificial intelligence, it will always include a section on ethics and governance even if you did not request it.

AI models represent an averaged out, central distribution of concepts, whereas human ideation has high variance and operates at the edges. The true value and human connection are found at the edges of the distribution, which AI struggle to replicate.

As media becomes cheap and infinite, human attention becomes the only non-inflationary resource that cannot be artificially multiplied. The challenge for platforms is ensuring that when a human spends attention, they are receiving genuine value.

As the cost of generating standard text drops to zero, the value shifts entirely to things that cannot be automated, like authentic human experiences and relationships. Much like a live string quartet, the proof of human effort becomes the premium asset.

The Odyssey has kept humanity debating and sharing perspectives for three thousand years. A great piece of writing is defined by its ability to spark long-term human conversations and build connections that ripple outward across generations.

Substack could build tools that highlight concepts and themes that are driving meaningful, high-context conversations. Highlighting these spiky, high-variance ideas acts as the opposite of slop detection, showing readers that original thought is valued here.

Video currently serves as a guarantor of human effort because it requires physical presence. However, as AI video avatars become indistinguishable from real humans, we will need to develop new cultural norms and expectations to prevent deceptive interactions.