Substack AI Detector Turns Every Writer Into a Suspect

On Tuesday, Substack turned AI suspicion into a product feature: the Substack AI detector will let readers scan posts, notes, replies, and comments to estimate whether text was written by AI or with AI assistance.

That timing matters because Substack is not framing this as a ban. It is framing it as a trust tool. The company is giving readers a way to investigate authorship themselves, rather than waiting for platform enforcement or creator disclosure, according to The Verge.

The deeper signal is sharper than the product announcement. Substack sells direct relationships between writers and readers. An AI scan button introduces a new layer of doubt inside that relationship. It may help readers spot low-effort AI material. It may also make polished writing feel like evidence in a trial.

Tuesday's Substack AI detector rollout turns trust into a clickable feature

Substack co-founder and CEO Chris Best gave the move a specific label: “Claudefishing.” In his framing, the problem is not simply AI use. The problem is a reader believing they are engaging with human thought when they are not.

“The core problem is not people using AI, or the quality of its output,” Best writes. “The problem is when there is a mismatch between a reader’s expectation and reality, especially when they unwittingly invest their attention in something with no human thought on the other end. That’s Claudefishing.”

That is an unusually direct statement for a platform that hosts writers who may use AI in very different ways. Substack is trying to draw a line between assistance and deception, but it is not pretending the line will always be clean.

The company’s own position leaves room for AI-assisted work. Best says “Pangram can only detect whether AI was used to make the text, not whether great human care went into creating it, nor whether AI tools were used as a source.” That caveat matters. A detector can flag likely machine involvement. It cannot judge effort, editorial intent, originality, or whether the final work reflects a writer’s real view.

XOOMAR analysis: This makes the Substack AI detector less like a moderation hammer and more like a trust signal. It gives readers more information, but it also pushes judgment onto them.

How Pangram scans posts, notes, replies, and comments after the launch

The tool is powered by Pangram, an outside AI detection company. It is rolling out on web and the iOS app, with Android coming “soon.” Readers can scan text longer than 100 words by opening the three-dot menu in the top-right corner of a post and selecting “Scan for AI text.”

The feature applies across Substack’s main social and publishing surfaces:

Substack is also adding creator-side tools. Writers can add a “How I make this” statement to explain their process to readers. They can scan drafts with Pangram before publishing. They will also have an option to report inaccurate results.

That structure is important. Substack is not automatically stamping every post with an AI label. A reader has to request the scan. That choice lowers the chance that the platform turns the detector into a universal public mark, at least for now, but it also means the tool’s impact will depend on reader behavior.

XOOMAR analysis: Outsourcing detection to Pangram gives Substack speed and specialist tooling, but it also means reader trust in the feature depends partly on Pangram’s accuracy and how clearly Substack explains the result.

The 100-word threshold exposes the limits behind AI probability scores

The hard number in the rollout is 100 words. Anything shorter cannot be scanned through the reader tool, based on the described feature. The other key rollout details are also concrete: the feature is available on web and iOS, with Android support coming later.

Substack’s own post adds a broader data point: Pangram estimates that AI-generated text is “as much as 40% on some platforms,” according to Substack’s announcement. Substack does not say that figure applies to its own service.

The bigger issue is how readers interpret the output. The tool provides an estimate, not a definitive verdict. That distinction may sound small, but it is the whole product risk. Percentages feel precise. Authorship is messier.

Best acknowledges the limitation directly. Pangram can detect whether AI was used to make the text, but not whether care, judgment, or real reporting shaped the final result. Substack also says the tool is “not perfect” in its own announcement.

That makes presentation critical. If the Substack AI detector returns a score that readers treat as proof, the feature could create more certainty than the underlying method deserves. If Substack frames the output as a prompt for judgment, it has a better chance of improving transparency without turning every scan into a verdict.

Substack's Tuesday move follows its fight against “fakeness,” not AI itself

Best’s post does not present AI as inherently bad. It says Substack uses AI internally for software, research, product features including clipping, translations, and more. The company is not trying to build a human-writing-only platform.

The target is deception at scale. Best writes:

“When readers have to wonder if what they’re reading is real, it undermines trust in authorship and threatens the livelihood of writers — including those who use AI tools thoughtfully to produce work they believe in,” Best writes. “Platforms that reward fakeness will create a race to the bottom.”

That sentence explains why Substack is moving now. The company’s paid-media model depends on readers believing there is a person worth paying behind the work. If subscribers begin to suspect that posts, comments, or replies are mostly synthetic, the value of access weakens.

This is also why the tool covers comments and replies, not just long-form posts. Substack’s trust problem is not limited to essays. A paid writer’s value often comes from the whole relationship around the publication: posts, discussions, reader responses, and the feeling that a human is present.

For XOOMAR readers tracking AI inside consumer products more broadly, this sits beside other product-level AI shifts we have covered, including Spotify AI Chatbot Takes Over the Premium Listening Flow and Physical AI Delivery Robots Hit the Costly Last 50 Feet. The common thread is control: who knows when AI is acting, and who gets to decide whether that is acceptable?

Writers and readers will use the scan button for different reasons

Readers get the clearest immediate benefit. They can test whether a piece of writing that feels personal, paid, or expert may have involved AI. That is especially relevant when a creator has not explained their process.

Writers get a more complicated bargain. The “How I make this” statement gives them a place to set expectations before readers start scanning. That could become a new norm for creators who use AI for drafting, editing, research support, or other parts of their workflow. The source material does not require such disclosure, but the product clearly makes process more visible.

Substack gets to signal quality control without banning AI-assisted writing. That is a pragmatic stance. A hard ban would conflict with Best’s own acknowledgment that some writers use AI thoughtfully and that Substack itself uses AI in product development.

Detection vendors also gain relevance. Pangram is no longer just a tool used outside the reading experience. It becomes part of Substack’s interface. That raises the stakes for how detection results are worded, challenged, and corrected.

XOOMAR analysis: The most important new behavior may not be scanning. It may be preemptive disclosure. Once readers can check, serious writers have a stronger incentive to explain their process before suspicion fills the gap.

Paid newsletters now face a sharper authorship test

The practical effect for creators is simple: if readers are paying for voice, judgment, reporting, or intimacy, unclear AI use becomes a business risk.

That does not mean every AI-assisted post loses value. Best explicitly leaves room for “people who use AI tools thoughtfully to produce work they believe in.” But the reader has to know what kind of work they are buying. A personal essay, an advice column, and an analysis newsletter each carry different expectations about human presence.

The new Substack AI detector may push creators toward clearer policies. A writer might explain whether AI is used for editing, summaries, idea generation, translation, or draft review. The source does not say Substack will require that detail. The platform is, however, giving creators a dedicated place to explain process.

For paid media, this is the key shift. Detection alone cannot preserve trust. Provenance, disclosure, and a recognizable editorial identity will matter more if AI assistance becomes common. Readers do not just pay for words. They pay for the belief that someone specific stands behind them.

The next decision point is whether Substack makes AI preference controls real

Substack says this is a starting point. Depending on feedback and interest, it is considering tools that would let users set preferences around AI content in Reply Rules, give creators more ways to express their individual value, give readers preference controls for recommendations, and improve systems that fight spam, bots, and scams as AI accelerates those problems.

That is where the real product test begins. The first version gives readers a scan button. The next version could shape what they see, what communities allow, and how creators describe their work.

The thesis to watch: the Substack AI detector will normalize AI authorship checks, but it will not settle the authorship debate by itself. Evidence that would strengthen that thesis includes wider use of “How I make this” statements, more reader preference controls, and clearer norms around AI-assisted writing. Evidence that would weaken it would be low reader adoption, frequent disputed scans, or creators treating the feature as noise.

The winning model will not be a magic detector. It will be disclosure plus human identity plus consequences for deceptive automation. Substack has started with the button. The harder part is what the platform, writers, and readers decide to do after someone clicks it.

Why It Matters

• Substack is making AI authorship suspicion a visible part of the reader experience.
• The tool could help identify low-effort AI content but may also increase distrust of legitimate writers.
• The rollout highlights the growing tension between AI assistance, transparency, and creator-reader trust.