Substack Fights Back Against Ghost-Written Content With New AI Scanning Feature

Substack is taking a direct swing at the growing problem of AI-generated content flooding newsletter platforms, rolling out a built-in detection tool that gives readers the ability to scrutinize what they're actually consuming. The feature can analyze posts, notes, replies, and comments, returning an estimate of how much of a given piece of text may have been written by AI or produced with significant AI assistance. It represents one of the more concrete platform-level responses to a problem that has quietly undermined trust across content ecosystems for the past two years.

The detection capability is powered by Pangram, an AI detection company that has been building models specifically designed to identify the statistical fingerprints left behind by large language models. Readers who want to run a scan simply need to open the three-dot menu in the top-right corner of any post longer than 100 words and select the 'Scan for AI text' option. The rollout is currently live on the web platform and the iOS application, with an Android version described as coming 'soon' in the company's official blog post.

Substack has built its reputation on the idea of authentic, individual voices — writers building direct relationships with paying subscribers who trust that what they're reading reflects genuine human thought and effort. The rise of AI writing tools has complicated that premise considerably, with some publishers reportedly using generative tools to churn out content at scale while still charging subscription fees. This new tool is a direct acknowledgment that the platform's core value proposition could erode if readers begin to doubt the authenticity of what they're paying for.

The move puts Substack in a growing cohort of platforms attempting to surface AI provenance information to end users rather than leaving detection entirely up to individual readers or third-party browser extensions. Turnitin and GPTZero have operated in the academic space for some time, but integrating detection natively into a content platform signals a shift toward systemic accountability. Whether Pangram's underlying models are accurate enough to avoid damaging false positives — flagging genuine human writing as AI-generated — will be a critical test of whether this tool earns reader confidence or simply creates new controversy.

Substack co-founder and CEO Chris Best has previously spoken about the importance of the creator-reader relationship as foundational to the platform's business model. This detection feature appears consistent with that philosophy, essentially handing readers a new instrument to hold writers accountable. The longer-term question is whether detection alone is sufficient, or whether Substack will eventually need to introduce disclosure requirements or labeling policies to more formally address AI content on the platform.