Substack: How writers are reacting to Substack's AI transparency tools

What the article is about

In July 2026, Substack integrated Pangram, an AI detection service, into the platform. Writers can now scan drafts before publishing, and readers can estimate how much of a post was written by humans versus AI. Writers also have the option to add a “how I make this” statement — a short disclosure describing their creative process.

This article, published by the Substack editorial team on July 24, 2026, collects and reflects on how writers responded to these tools in the days after launch. It is not an analysis of AI detection accuracy; it is a record of how a professional community is navigating questions of transparency, authenticity, and trust that the writing industry has been circling for years.

Context

The rollout came after Substack introduced Pangram for readers to scan posts with more than 100 words. Publishers can disable scanning on individual posts, and the feature is positioned as encouraging disclosure rather than enforcement. Substack is explicit that it permits AI-assisted writing; the tools are framed as helping readers understand process rather than certifying human authorship.

Key takeaways

The reactions documented in the article divide roughly into two camps. One group of writers welcomed the tools as overdue: they argued that readers who pay for newsletters are entitled to know how those newsletters are produced, and that the “how I make this” feature creates a useful norm of transparency even for writers who use no AI at all. Commentators like Gergely Orosz and Brie Wolfson were among those who saw the feature as affirming what distinguishes quality writing from automated output.

The second group raised concerns that cut deeper than the tools themselves. Some writers noted that the detection feature creates scrutiny of process rather than attention to content — a dynamic that may penalize writers who use AI for research or editing without it affecting the quality of what they publish. Others worried about false positives, particularly for writers whose voice has been unconsciously shaped by exposure to AI-generated text.

The “how I make this” statements that writers shared reflect a wide range of approaches: some disclose using Claude for structural editing, others describe zero AI involvement, and a few offered wry, humorous summaries that treat the question of AI involvement as absurd or beside the point.

What it does not cover

The article does not assess detection accuracy or discuss the technical limitations of AI detection tools. It also does not address how Substack’s approach compares to policies at other publishing platforms.

Who it is useful for

Writers who publish regularly on Substack and have not yet formed a position on AI disclosure will find this article a useful map of the positions being staked out. It is also relevant for editors and content strategists at publications that are developing their own AI-use policies, since the “how I make this” format offers a practical model for voluntary disclosure.