Substack rolls out AI detection tool to flag machine-written posts, notes and comments

Substack is adding an AI detection feature that will let readers see whether a post, note, reply or comment may have been generated or heavily assisted by artificial intelligence. The company says the tool is designed to make authorship clearer at a time when online publishing is increasingly crowded with synthetic text and blurred human-AI boundaries.

The new feature, built with AI detection firm Pangram, is now rolling out on the web and iOS, with Android support expected soon. Readers can use it on content longer than 100 words, while writers will also get ways to describe their process and challenge results they believe are wrong.

What Substack is changing

Substack’s latest update is less about blocking AI entirely and more about labeling it. The platform says the tool will estimate how likely a piece of text is to have been written with AI assistance, giving readers another signal before they decide whether to spend time on a post.

That matters because Substack has become a home for independent writers, niche publications and personality-driven newsletters, all of which depend heavily on trust. The company is effectively acknowledging that readers are now encountering more content that may have been drafted, edited or even mass-produced with the help of generative tools.

The company’s terminology is notable. Substack has framed the issue around what it calls “Claudefishing,” a label for content that presents itself as human-made while being driven in part by AI. The broader concern is not just whether AI is used, but whether readers are being misled about the amount of human thought behind what they are reading.

How the AI detector works

The detector is powered by Pangram, a company focused on identifying machine-generated text. On Substack, readers will be able to open the three-dot menu on a post and select a “Scan for AI text” option.

Only text longer than 100 words can be analyzed. Substack says the feature is being introduced across multiple parts of the platform, including posts, notes, replies and comments, giving users a way to inspect content throughout the social and publishing ecosystem the company has built.

Where readers will see it

The rollout currently covers the web and iPhone app. Android users will have to wait a little longer, with Substack saying support is coming soon.

That phased launch suggests the company wants to test the feature in places where its reading and publishing experience is most active, then expand once the system is stable enough for broader use.

Why Substack says it is doing this

Substack’s leadership is presenting the feature as a trust-and-transparency measure, not an anti-AI crusade. In the company’s view, the real danger is not the use of AI itself, but confusion about what kind of work a reader is actually consuming.

Chris Best, Substack’s co-founder and chief executive, said the platform wants to reduce the gap between what readers assume they are reading and what was actually created. He argued that the problem begins when someone invests attention in a piece of writing without realizing there may be little or no human thinking behind it.

Best also suggested that a publishing environment that rewards imitation and deception can become corrosive for the people who make a living writing. In his view, if platforms reward “fakeness,” they push creators toward a bottom-of-the-barrel race to appear real while offering less genuine value.

The company’s stated objective is straightforward: help readers make an informed choice about whether a post is worth their time. That framing reflects a broader shift across the internet, where authenticity has become a commercial and editorial concern rather than just a philosophical one.

What creators can do with the new system

Substack is not only giving readers a detection tool; it is also giving writers more ways to explain themselves.

Creators will be able to add a new “How I make this” statement that can describe their workflow, including whether they use AI tools at any stage. Writers can also scan their own drafts with Pangram before publishing, which may help them understand how the detector treats their text and adjust if needed.

In addition, Substack says authors will be able to flag results they think are inaccurate. That is an important safeguard, because even sophisticated detection systems can make mistakes, especially when prose is highly edited, formulaic or stylistically similar to machine-generated writing.

Why the writer disclosure matters

The disclosure feature may become just as consequential as the detector itself. By allowing authors to explain how their work is made, Substack is signaling that readers may want context, not just a yes-or-no label.

For journalists, essayists and independent newsletter operators, that context could include anything from brainstorming with AI to using it for copyedits, outlines or research summaries. Substack appears to be trying to create room for those distinctions instead of reducing the conversation to a binary human-versus-machine debate.

How reliable is AI text detection?

Substack is being careful not to oversell what Pangram can do. Best noted that the system can identify whether AI may have been involved in producing a text, but it cannot measure the quality of human judgment, originality or editorial care behind it.

That caveat is critical. A piece of writing can be fully human and still be poor, deceptive or spammy. It can also be polished with AI help and still be thoughtful, original and useful. The detector is therefore a signal, not a verdict.

It also cannot determine whether AI was used as a source rather than as a writing tool. That limitation matters in an era when creators might use AI for research, summarization, drafting or translation, all of which can leave different traces in the final copy.

In practical terms, that means the feature will probably work best as a transparency aid, not as a decisive enforcement mechanism. Readers should treat it as one clue among many rather than a final judgment on the credibility of a post.

How does this fit into the wider AI-authorship debate?

Substack’s move arrives as publishers, social platforms and creators struggle with a new question: what counts as authentic writing when AI tools are increasingly integrated into the creative process?

The answer is different depending on who you ask. Some readers care mostly about usefulness, regardless of whether a human, an assistant or a model helped produce the text. Others want a clear boundary between independent writing and generated content, especially if they are subscribing to someone for a personal voice or expert judgment.

Platforms are being pulled between those two expectations. If they allow AI use but say nothing about it, they risk eroding trust. If they police AI too aggressively, they may alienate creators who use the tools responsibly and transparently.

Substack’s solution is to surface the issue rather than settle it. The company is betting that more information will create more confidence, even if it does not resolve every edge case.

What this means for readers and writers

For readers, the most immediate effect is a new layer of context. A Substack post may now come with an opportunity to ask whether its voice, style or structure feels human-authored in a meaningful way.

For writers, the update could encourage more disclosure and, potentially, more pressure to justify how they work. Some creators may welcome that. Others may worry that AI detection tools could misread stylistic choices or discourage experimentation.

• Readers gain a quick way to inspect potentially AI-generated writing.
• Writers gain a disclosure tool to explain their process.
• Substack gains a visible trust feature as AI content spreads online.
• The detector may help with transparency but cannot prove intent or quality.

There is also a reputational layer. Substack has long marketed itself as a platform for direct relationships between writers and audiences. If readers begin to suspect large amounts of synthetic content, that relationship becomes harder to maintain. The company’s new tool is an attempt to preserve that bond before skepticism spreads further.

Why this matters for the newsletter economy

The newsletter business depends on personal voice, niche expertise and a sense of direct connection. Unlike larger social platforms, where content can feel anonymous and disposable, Substack’s appeal rests on the idea that readers are following a person or publication they trust.

That trust becomes fragile when it is unclear whether a newsletter is genuinely written by the person whose name appears on it. AI has made it cheap and easy to produce large volumes of plausible text, which has raised fears about spam, impersonation and low-effort publishing at scale.

By introducing detection and disclosure tools, Substack is responding to a market-wide problem: the value of written content is increasingly tied to provenance. Readers want to know not just what they are reading, but who made it and how.

That could become a competitive advantage if Substack can make its platform feel more transparent than open web publishing elsewhere. But it could also expose how difficult the problem is to solve. Detection alone will not prevent abuse, and disclosure alone will not guarantee honesty.

Timeline of Substack’s AI transparency push

The company’s latest update fits into a broader pattern of platforms adding rules, labels and warnings around generative AI. The details below summarize the rollout described by Substack.

What to watch next

The key question now is whether Substack users will actually rely on the detector and whether creators will embrace the transparency tools. If the feature becomes part of the reading habit, it could change how newsletter audiences evaluate trust. If it is ignored, it may become just another buried platform setting.

Another issue is accuracy. Pangram’s results may be persuasive for some users, but any detection system in the AI era will inevitably face false positives and false negatives. How often writers dispute results, and how Substack responds, could shape the feature’s credibility.

Finally, the rollout may influence other publishing platforms to follow suit. As AI-generated and AI-assisted writing becomes more common, platforms will face pressure to explain how they distinguish between creative assistance, editorial work and deception.

For now, Substack’s message is clear: when readers cannot tell whether a person really wrote something, the platform believes trust starts to erode. The company is betting that more disclosure will help keep that from happening.

And in a digital landscape where readers are increasingly asked to sort authentic voices from synthetic ones, that could become one of the most important product decisions a publishing platform can make.

Quick facts

• Tool name: Pangram-powered AI scan
• Available on: Web and iOS now, Android soon
• What it scans: Posts, notes, replies and comments
• Length threshold: Over 100 words
• Writer feature: “How I make this” disclosure statement