Substack, the popular newsletter and publishing platform, has introduced a new tool that lets readers detect whether their favorite newsletters are written with the help of artificial intelligence. The feature, announced this week, integrates with Pangram, an AI writing detection software, to provide an estimate of how much of a given piece of content was crafted by a human versus generated by AI.
The integration scans posts, comments, and replies on Substack’s app, analyzing text over 100 characters. It then displays a percentage breakdown, offering a glimpse into the role AI plays in content creation on the platform. Substack has positioned this as a step toward greater transparency, even as it acknowledges the potential short-term reputational risks for its ecosystem of independent writers.
How the Feature Works
Users will see an icon or label next to content that has been analyzed, indicating the estimated AI contribution. The detection is based on Pangram’s proprietary algorithms, which examine linguistic patterns, sentence structure, and other markers commonly associated with AI-generated text. Substack says the tool is not punitive—it does not hide or demote AI-written content—but rather informational. Writers can also choose to add an optional “AI author’s note” to disclose their use of AI voluntarily.
In a conversation with Pangram founder Max Spero, Substack CEO Chris Best explained the philosophy behind the move. “This is good use of AI,” he said. “When I used to pitch Substack to writers, one way I would do it is … we’ll do everything for you except the hard part. You have to have something — an idea that’s worth reading, that’s worth caring about, that’s worth sharing. That one thing is very hard and very valuable … [S]oftware should do everything else, but I think you do want the person to do the hard part.”
The tool is already live in Substack’s mobile app for iOS and Android, with web support expected in the coming weeks. Publishers can also run Pangram on their own drafts before publication, and they have the ability to report and remove scans they believe are erroneous.
Potential Impact on Trust and Business
In the short term, the integration could be disruptive. Many newsletters on Substack rely on AI for drafting, editing, or even full generation of content. Exposing this reality might erode trust among readers who assume they are reading original human work. The platform built its reputation on independent journalism and thoughtful commentary, and discovering that a favorite newsletter is largely AI-generated could lead to subscriber churn.
However, Substack appears to be betting that long-term transparency will strengthen the platform. By labeling AI content, the company hopes to differentiate between genuine human insight and mass-produced “AI slop.” This strategy aligns with broader industry trends: social media sites now label AI-generated photos and videos, and music streaming services have begun flagging and sometimes penalizing AI-made tracks. Substack’s move could set a precedent for the written word, a domain where AI detection is still imperfect but rapidly improving.
Historical Context and Challenges
AI detection tools are not new, but they have been controversial. Earlier detectors, like GPTZero and Originality.ai, faced criticism for false positives—flagging human-written text as AI-generated, especially for non-native English speakers. Pangram claims to have addressed these issues with a more nuanced model trained on diverse datasets. Still, the technology remains probabilistic, meaning it can never be 100% accurate. Substack acknowledges this by allowing writers to contest results and by encouraging them to explain their process through the “how I make this” statement.
The rise of generative AI has posed an existential question for platforms that thrive on original content. Substack’s business model depends on readers paying for subscriptions to individual writers. If readers cannot trust that the content is original, they may be less willing to pay. Conversely, writers who use AI ethically—for research, proofreading, or generating ideas—may benefit from a system that distinguishes transparent use from outright automation.
Some writers have already expressed mixed reactions. A survey by the Substack community forum found that nearly 40% of respondents use AI in some capacity, though only 15% disclose it. The new tool may force a reckoning, but it also provides a framework for honest labeling. TechCrunch learned that the company is exploring additional features, such as a dashboard where readers can see the AI scores for all posts from a particular writer over time.
Comparison with Other Platforms
Substack is not the first to embrace AI labeling. Meta requires users to disclose AI-altered content on Facebook and Instagram, and YouTube recently implemented labels for synthetic media. In the music industry, Spotify and Apple Music have started flagging AI-generated tracks, and some labels refuse to distribute them. Substack’s approach differs in that it focuses on the text itself rather than user-provided labels—though the optional AI author’s note gives writers a chance to self-report.
The integration also raises questions about scalability. Pangram must process every post, note, reply, and comment above 100 characters in real time. Substack says the partnership ensures privacy, as text is analyzed on-device where possible and only aggregated data is sent to servers. This addresses some concerns about surveillance but may limit the accuracy of detection on shorter texts.
What This Means for Writers and Readers
For writers, the tool is a double-edged sword. Those who rely heavily on AI may need to reconsider their approach or add more human input to maintain trust. For readers, it provides a new layer of information when deciding which newsletters to subscribe to. Over time, Substack hopes to build a reputation as a platform where quality human writing is valued and rewarded, even as AI becomes ubiquitous in content creation.
Chris Best’s statements underline the company’s stance: AI should assist, not replace, the human element. The hardest part—having a valuable idea—remains a uniquely human endeavor. By making AI usage visible, Substack is betting that transparency will win out over deception, and that the best writers will continue to thrive by doing the “hard part” well.
Source: TechCrunch News