Substack is adding an AI-writing detection feature that lets readers check whether posts, comments, replies, and other text may have been written with AI. The tool is powered by Pangram and is designed to give users an estimate of how much of a piece appears human-written versus AI-generated.
The feature is not just about catching writers. It points to a larger issue across online publishing: readers increasingly want to know whether a person actually wrote what they are reading.
How the Feature Works
Substack’s new tool allows users to scan longer pieces of text for AI involvement. The feature has been described around a clear user action: “Scan for AI text.” It can apply to posts and social-style interactions on the platform, including replies and comments.
Pangram has also published a plain-language explanation of how its detector works, which is useful context for readers who want to understand what these tools can and cannot prove.
Substack is also adding a creator-facing feature called “How I make this,” which gives writers a place to explain their process. That may matter as much as the detector itself. Readers do not always object to AI assistance, but they usually want honesty about how the work was made.
Why This Is a Trust Issue
AI-written content is not automatically low quality. A writer might use AI to outline, edit, summarize notes, or clean up structure. The problem starts when content is presented as personal, expert, or deeply human while being mostly generated by a tool.
That gap between expectation and reality is where trust breaks. Newsletter readers often pay because they value a specific writer’s judgment, voice, research, or lived experience. If that work is quietly replaced by automated text, the relationship changes.
This is especially important on platforms like Substack because the product is built around direct reader support. People are not only reading information. They are choosing whose work deserves attention and money.
The Limits of AI Detection
AI detectors can be useful, but they are not perfect. They can misread heavily edited writing, formulaic human writing, or text from non-native speakers. They can also miss AI-assisted work that has been rewritten carefully.
That means detection should be treated as a signal, not a final judgment. A platform can use it to encourage transparency, but it should be careful about turning uncertain scores into public accusations.
For readers, the most practical approach is to combine tool signals with normal judgment. Look at whether the writing has original reporting, personal experience, clear sourcing, useful examples, and a consistent point of view.
Why Platforms Are Acting Now
Substack is not alone in dealing with this problem. Social platforms, video sites, and publishing tools are all trying to respond to low-effort AI content, often called “AI slop.” The challenge is separating useful AI-assisted work from mass-produced filler.
Platforms do not want to block all AI use. Many creators use AI in practical ways. But platforms also know that reader trust can fall quickly if feeds become crowded with generic automated posts.
Tech Block Central has also covered how AI is changing work and content production in AI and jobs and AI coding tools.
The Practical Takeaway
Substack’s AI detector is less about banning AI and more about disclosure. Readers can handle AI-assisted writing when the process is clear. What they dislike is being sold human insight and receiving generic machine-made content instead.
The future of online publishing may depend less on whether writers use AI and more on whether they are honest about it. Would an AI label make you more or less likely to trust a newsletter?