In short
Substack has launched an AI detection feature with Pangram that estimates how much of a post, note, reply or comment was written by a human versus AI. The company says the goal is transparency, not punishment, as it tries to preserve trust in its newsletter ecosystem.
- Substack now lets users scan qualifying posts, notes, replies and comments for estimated AI use.
- The feature is powered by Pangram and is intended to promote transparency, not penalize AI-assisted writing.
- Writers can add an optional AI Author’s note and can challenge or remove mistaken scans.
- The rollout reflects a wider platform trend toward labeling synthetic content more clearly.
- The move could improve trust over time, but it may also expose how many newsletters rely on AI.
Substack has introduced a new AI-detection feature that lets readers see whether newsletters, notes, replies and comments on the platform were written by a person, generated by AI, or produced through a mix of both. The rollout matters because it could reshape trust on one of the web’s biggest independent publishing platforms at a moment when AI-assisted writing is becoming increasingly common.
The company says the tool, built with AI detection provider Pangram, is intended to increase transparency rather than punish creators who use AI in some part of their workflow. It also gives publishers a way to review drafts, challenge false positives and disclose their process with an optional “AI Author’s note.”
For Substack, the feature is both a product update and a statement about the kind of publishing ecosystem it wants to support. In the short term, the new label could unsettle some creators and raise questions about how much machine help is already being used on the platform. In the longer term, Substack is betting that clearer disclosure will make readers more confident that the newsletters they pay for are honest about how they are made.
What Substack launched and why it matters
Substack’s new feature provides an estimate of how much text in a post may be human-written versus AI-generated, and it does so directly inside the app. The system can scan posts, notes, replies and comments longer than 100 characters, giving readers a quick signal about the composition of the content they are seeing.
That may sound technical, but the stakes are practical. Substack has built much of its brand around the idea that independent writers can publish directly to readers without the gatekeeping of traditional media. If readers begin to believe that a meaningful share of those publications are not actually authored by the people behind the mastheads, trust in the entire ecosystem could weaken.
At the same time, AI-generated text is now commonplace across media, marketing and social platforms. Substack is joining a broader industry trend toward labeling synthetic content more explicitly, rather than pretending the distinction does not matter.
How the new AI detection feature works
The feature relies on Pangram, a company focused on identifying AI-written text. Inside Substack’s app, users can scan eligible content and receive an estimate of how much of that material appears to have been written by a human and how much appears to have come from AI.
Substack said the feature is not designed as a ban or a penalty system. Instead, the company wants it to act as a transparency layer, encouraging writers to describe their process more openly.
What content can be scanned?
Substack says the tool will apply to posts, notes, replies and comments over 100 characters. That means it is not limited to full newsletters; it also extends into the social layer of the platform, where quick replies and shorter updates can also shape a writer’s reputation.
Writers will also be able to add an optional AI Author’s note. Substack says that note is meant to help creators disclose how they used AI, if they chose to use it at all.
Can publishers challenge the results?
Yes. Substack says publishers can use Pangram on drafts before publication and can report scans they believe are mistaken. They can also request removal of scans from their own work if they think the system flagged content incorrectly.
That safeguard matters because AI detection is not perfect. Even companies that build these tools acknowledge that automated judgments can produce false positives, especially when writing is formulaic, highly edited or produced with the help of language tools that do not fully generate text from scratch.
Why Substack is taking this direction now
Substack is moving in the same direction as other major platforms that have begun labeling AI-generated media more clearly. Social networks now routinely tag AI-made images and videos, while the music industry has started experimenting with disclosures and penalties for synthetic songs.
The change reflects a broader shift in how technology companies are thinking about authenticity. A few years ago, the question was whether AI could produce content that looked human. Now the question is whether platforms can preserve trust once that content becomes hard to distinguish from the real thing.
For Substack, the pressure is especially high because it sits at the intersection of publishing, paid subscriptions and personal brands. A reader who subscribes to a newsletter is not simply buying access to information. They are often buying the voice, judgment and perspective of a specific writer.
If that writer is leaning heavily on AI, readers may want to know. Some may not care as long as the final article is accurate and useful. Others may see undisclosed AI use as a breach of the relationship they believed they had with the author.
What Chris Best says Substack is trying to protect
Substack chief executive Chris Best framed the feature as a way to preserve the part of writing that still belongs to humans. In a conversation with Pangram founder Max Spero, Best said the company sees this kind of detection as a productive use of AI because it helps clarify authorship rather than obscure it.
Best said that when he used to pitch Substack to writers, he described the service as handling everything except the hardest part: having a real idea that is worth reading, caring about and sharing. In his view, software should take care of the rest, while the human author should still be responsible for the core intellectual work.
That comment gets at a central tension in digital publishing. AI can now draft quickly, summarize information and smooth out prose. But Substack is suggesting that the value readers pay for is not just polished output. It is originality, judgment and a point of view that still feels human.
Why the move could help and hurt Substack at the same time
The feature may strengthen Substack’s credibility in the long run, but it could also create short-term discomfort. If the tool reveals that many newsletters rely on AI to some degree, some readers may question whether the platform’s independent publishing culture is as authentic as they assumed.
That could hurt the perception of high-quality writing on the platform, at least temporarily. It could also pressure writers who use AI lightly for editing, outlining or polishing to disclose more than they have in the past.
Still, Substack appears to believe that greater honesty is a better business strategy than ambiguity. If readers feel they can trust the platform’s disclosures, they may be more willing to subscribe, share and pay for content.
| Key element | What Substack announced | Why it matters |
|---|---|---|
| Detection partner | Pangram | Provides the AI-human content estimate |
| Content covered | Posts, notes, replies and comments over 100 characters | Extends transparency beyond full newsletters |
| Purpose | Disclosure and transparency, not punishment | Aims to reduce mistrust without banning AI use |
| Publisher controls | Draft scanning, error reporting and removal requests | Gives writers a way to manage false positives |
| Disclosure option | AI Author’s note | Lets creators explain how AI was used |
How does this compare with other platforms?
Substack’s approach resembles a growing set of platform rules that distinguish between synthetic and human-made content instead of treating all AI use the same way. Social media companies frequently label AI-generated images and video, while streaming platforms and music services have begun testing similar signals for AI-created songs.
The idea is not necessarily to eliminate AI from the creative process. Rather, it is to give audiences a better sense of what they are consuming and to let them decide how much weight to place on that disclosure.
Substack’s approach is somewhat distinct because it is aimed at writing, a medium where AI assistance is often subtle. A newsletter may not be entirely machine-written to be meaningfully affected by AI. Even light use in drafting or rewriting can alter tone, voice and originality.
Why written content is harder to label
Written work is harder to classify than images or videos because AI assistance can range from small edits to full generation. A writer might use a model to brainstorm headlines, summarize interviews or polish grammar without surrendering authorship in any obvious way.
That ambiguity is precisely what makes a tool like Pangram attractive to platforms and controversial among creators. The same feature that promises transparency can also feel like surveillance if users think the system will overstate machine involvement.
Who is most likely to be affected?
Newsletter writers who rely on AI for significant portions of their work are the most obvious group likely to feel the impact. But the feature could also affect editors, media brands, marketing writers and creators who use Substack as a publishing and community platform rather than just a simple email tool.
Readers are affected too. The label may change how they interpret a piece before they even begin reading it, much as a nutrition label can influence how people think about a packaged product.
- Independent writers may feel pressure to disclose their process more clearly.
- Readers may become more skeptical of polished but impersonal newsletters.
- Publishers may use the tool as a drafting aid before hitting publish.
- Platforms may continue to expand AI labels across more content types.
What this means for the future of newsletter publishing
Substack’s move suggests that the next phase of the AI content debate will not be about whether AI is present, but about how openly it is acknowledged. For many publications, the line between human and machine authorship is already blurry. The question is whether publishers will treat that blur as a problem to hide or a reality to explain.
If Substack’s tool is widely adopted and trusted, it could become part of a new standard for digital publishing: not “AI or not AI,” but “how much, and for what purpose?” That would represent a meaningful shift from the early hype cycle, when many companies were eager to emphasize speed and scale above all else.
There are also reputational risks. A platform that markets itself as a home for writers wants readers to believe they are hearing from people with original ideas and strong editorial judgment. If detection tools reveal otherwise, Substack may need to balance transparency with reassurance.
That balance will likely define how this feature is received. Readers want honesty. Writers want flexibility. Platforms want trust. Substack is now trying to satisfy all three at once.
Timeline of the rollout
The company’s announcement marks the latest step in a broader industry move toward AI labeling. Here is a concise look at the rollout and the surrounding context.
| When | Event | Significance |
|---|---|---|
| Recent years | AI tools become common in writing and editing workflows | Raises questions about authorship and disclosure |
| Before this week | Platforms begin labeling AI-generated images, video and music | Sets precedent for transparency labels |
| This week | Substack announces Pangram integration | Brings AI detection to newsletters, replies and comments |
| Going forward | Writers can add AI Author’s notes and challenge mistakes | Creates a disclosure-first framework rather than a ban |
What readers should watch next
The most important question is not whether AI appears in Substack content at all. It is how often the detection system flags legitimate writing, whether creators embrace the disclosure tools and whether readers actually use the labels when deciding what to read and pay for.
If the system is accurate and widely adopted, it could help normalize honest disclosure across the platform. If it is noisy or inconsistent, it may deepen skepticism without improving trust.
Either way, Substack has made a clear choice: in the age of AI-assisted publishing, opacity is no longer the default. The company wants readers to see not just the finished product, but also the likely role of the software behind it.
Frequently asked questions
What did Substack announce about AI content?
Substack announced an AI detection feature that estimates how much of a newsletter post, note, reply or comment was written by a human versus AI. The company says the goal is transparency and clearer disclosure, not banning or punishing AI-assisted writing.
How does Substack’s AI detection tool work?
It works through an integration with Pangram, an AI writing detection company. Users can scan eligible content in the Substack app, and publishers can also test drafts before publishing, report mistakes and request removal of scans they believe are inaccurate.
Does Substack forbid AI-written newsletters now?
No, Substack does not prohibit AI-assisted writing with this feature. The company says it wants writers to disclose their process more openly, including through an optional AI Author’s note, rather than face penalties for using AI in some parts of their workflow.
Why is Substack adding AI detection now?
Substack is adding AI detection now because AI-generated text is becoming more common and platforms across the internet are moving toward clearer labeling. The company appears to believe that better disclosure will help preserve reader trust in its newsletter ecosystem.
Can writers dispute a false AI detection result?
Yes, writers can challenge suspected mistakes. Substack says publishers can run Pangram on drafts before publication, report scans they think are wrong and ask for scans to be removed from their own work if the system flags content incorrectly.
