Substack Now Lets Readers Check If a Newsletter Was Written by AI

Substack has rolled out a built-in tool that allows readers to scan posts, notes, comments, and replies for signs of AI-generated or AI-assisted text.

The feature, powered by detection company Pangram, launched in late July 2026 and is available on the web and iOS app (with Android support coming soon).

The move comes after growing evidence that AI writing has already penetrated the newsletter ecosystem. In April 2026, technology journalist Taylor Lorenz partnered with Pangram to analyze thousands of recent posts from Substack’s top bestsellers across categories.

The study found that 28% of writing in top Technology newsletters was fully or partially AI-generated — the highest rate among categories examined. Philosophy (23%) and Health (22%) also showed elevated levels. Some individual publications appeared to publish entirely AI-written issues.

---

Against “Claudefishing”

Substack CEO Chris Best announced the partnership in a post titled “Against Claudefishing.” He coined the term to describe the mismatch that occurs when readers invest attention in content they believe is human-authored, only to discover it was generated by AI with little or no human thought behind it.

“The core problem is not people using AI, or the quality of its output,” Best wrote. “Not everything made with AI is slop, and not all slop is made with AI.

The problem is when there is a mismatch between a reader’s expectation and reality.” He argued that unchecked AI text pollutes the human commons of the internet, erodes trust in authorship, and ultimately threatens the livelihoods of writers — including those who use AI tools thoughtfully.

Pangram estimates that as much as 40% of text on some social platforms is now AI-generated. Substack positions itself as an exception it wants to protect: a place built on direct relationships between writers and readers.

---

How the Tool Works

Readers can open a post or note in the Substack Reader (or the iOS app), tap the three-dot menu, and select “Scan for AI text.” The system returns an estimate of how much of the content appears human-written versus AI-assisted. Scans are available only for text longer than 100 words published on or after July 21, 2026, and the result is visible solely to the person who requested it.

The tool is deliberately non-punitive. Substack is not banning AI assistance. Instead, it is giving both sides of the relationship more information.

Writers received a particularly elegant addition: an optional “How I make this” statement. Authors can briefly disclose whether and how they used generative AI — research, outlining, editing, full drafting, or not at all — and set expectations with their audience upfront. They can also run Pangram on their own drafts before publishing and report (or remove) any scans of their work they believe are inaccurate.

Pangram itself is a classifier trained on large volumes of human and AI text. Independent evaluations have credited it with high accuracy on fully generated content, though no detector is perfect, and false positives remain a concern—especially for non-native English writers.

---

Transparency Over Prohibition

The design choice is notable. Rather than trying to police process or declare certain workflows illegitimate, Substack is betting on disclosure. Readers who care about human authorship can check. Writers who use AI productively can explain their method without stigma. Those who produce fully automated newsletters will find it harder to pass the work off as personal.

In an environment where AI-generated text is rapidly becoming ambient, Substack’s approach treats the real problem as broken expectations rather than the technology itself. Whether the tool meaningfully restores trust—or simply makes the presence of AI more visible — will depend on how widely readers use the scan button and how honestly writers fill in the “How I make this” box.

For now, the platform has given its audience a new way to ask a simple question: Was a human actually behind these words?

---

Also worth reading:

• Patreon Fully Blocks AI Training Bots from Scraping Creator Content
• The First Exit in Clipping?
• GoPro on the Brink: Once a $4 Billion Icon, Now Facing Bankruptcy Warnings and Layoffs
• How People Are Really Using AI in 2026: Therapy, Tamagotchis, and Tarot

---

Thank you!