Substack has launched an AI-detection feature that estimates what share of a newsletter's text was written by AI, making that estimate visible to readers — a move that puts every AI-assisted writer on the platform under new scrutiny.
Key takeaways
Substack's new tool estimates the percentage of AI-generated text in individual newsletters and surfaces that information to readers.
The feature signals a platform-level shift toward AI-content transparency, following similar moves by YouTube and Patreon.
Writers who rely on AI drafting tools without disclosure now face a reputational risk if the detector flags their work.
AI-detection tools are probabilistic, not definitive — false positives remain a known limitation across the industry.
The move mirrors a broader trend: platforms are increasingly building AI-content signals into their reader-facing interfaces, not just their moderation back-ends.
Substack's new tool surfaces an AI-writing estimate directly to newsletter readers.
According to TechCrunch, the feature gives readers a percentage-based estimate of how much of a given issue was AI-authored — a figure derived from detection models that analyze writing patterns. Substack hasn't disclosed which underlying detection technology it uses, and the company hasn't announced whether high AI-percentage scores will trigger any monetization or distribution penalties.
That uncertainty matters. Writers who use AI tools for research summaries, headline drafts, or structural outlines — rather than wholesale generation — may find themselves flagged at rates that don't reflect their actual process. AI-detection tools have a documented false-positive problem: highly structured, edited prose can read as machine-generated even when a human wrote every word.
The platform transparency arms race
Substack's move fits a clear pattern. YouTube updated its monetization policies to explicitly bar low-effort AI-generated content from ad revenue, and Patreon partnered with Cloudflare to block AI training scrapers. Each of these represents a different pressure point — YouTube hitting the wallet, Patreon protecting the data pipeline, and Substack now targeting reader trust directly.
The reader-facing framing is the sharpest edge here. A monetization penalty happens behind the scenes; a visible AI percentage label on a newsletter shapes how subscribers perceive the writer before they read a single sentence. For writers who've built audiences on voice and authenticity, that label — even an inaccurate one — is a credibility risk.
What this means in practice for AI-assisted writers
For creators who use AI tools as part of their writing workflow, the calculation shifts. Lightly edited AI drafts that weren't disclosed before now carry a visible flag. Writers who use AI for structural scaffolding but rewrite heavily may still get flagged, with no mechanism on the platform to contest or contextualize the score.
The smarter response is preemptive disclosure. Writers who openly describe their AI-assisted process — in a byline note or an introductory paragraph — can reframe the detection score as confirmation rather than accusation. That transparency play also tends to perform better with audiences who've grown skeptical of undisclosed AI use, a skepticism that's accelerating as detection tools become standard infrastructure.
It's also worth watching whether Substack's feature feeds into its recommendation or discovery algorithms. If high AI-percentage scores quietly suppress distribution — even without a formal policy — writers would have no way to know. The platform hasn't addressed that question publicly.
For creators working across text and image generation, the dynamic is familiar: tools that make production faster also make provenance more visible. The AI image generation space has grappled with watermarking and provenance debates for years; text is now catching up. The difference is that image watermarks are often invisible to end users, while Substack's percentage label is front and center — a deliberate choice to make AI use a reader-visible attribute rather than a back-end moderation signal.
How accurate Substack's detection proves to be in practice will determine whether the feature builds trust or just manufactures anxiety. Either way, the era of undetected AI-assisted newsletters on the platform is effectively over.
