How to spot AI slop on Substack

The clearest signs of AI slop on Substack are no original reporting and no named sources, generic claims with nothing you can check, an anonymous pen name or an AI-generated headshot, a firehose of posts on unrelated topics, and generic stock or AI header images. Look for several of these together, not one. If you want a feed where every writer is a verified journalist and none of this happens, that is what we built Newsreel for.

First, the fair part. Substack is full of real reporters, essayists, and subject-matter experts who do careful, first-person work you cannot get anywhere else. Nothing below says otherwise. The problem is that AI-generated filler now sits in the same lists and the same inboxes, and at a glance it can look like the real thing. An April 2026 analysis found that more than a quarter of the writing in Substack's top technology newsletters was fully or partly AI-generated, the highest share of any category studied. The skill worth having is telling a real voice from a machine that is padding a word count, not writing off Substack wholesale.

We make Newsreel, so we have kept this straight: these are the tells reporters and researchers actually use, and none of them require a tool.

The content tells

Read the writing itself before you look at anything else. AI slop tends to share a shape.

• No original reporting. It is a roundup of other roundups. Nobody was called, nobody was there, and nothing is sourced back to a person or a document.
• No named sources and no checkable specifics. Claims stay general on purpose. There are no names, no dates you can look up, no numbers tied to a report you could open yourself.
• SEO-bait headlines. Titles built for a search engine rather than a reader, stuffed with the exact phrase someone would type into Google.
• The house style of a language model. Em dashes on every line, the rule of three in sentence after sentence, the "it is not just X, it is Y" construction, stock openers like "in an era of," pet words like "tapestry" and "testament to," and small claims inflated into sweeping significance.

The author tells

Then look at who is behind it. A real writer leaves a trail. A content farm tries not to.

• No real bio or verifiable credentials. There is no way to confirm the person exists, worked where they claim, or wrote anything before this.
• An anonymous or generic pen name with no accountable human attached to it.
• An AI-generated headshot. A too-perfect, oddly lit portrait that does not appear anywhere else on the web when you reverse image search it.
• A firehose of posts on unrelated topics. Crypto on Monday, parenting on Tuesday, geopolitics on Wednesday, all in the same flat voice and at a volume no single human could report.
• Subscriber and like counts that look inflated next to almost no real comments or discussion.

The image tells

The header image is often the quickest giveaway. AI newsletters lean on generic stock art or AI-generated illustrations that do not match any real scene the writer describes. Real reporting tends to show specific pictures: a named place, a real person, a screenshot of an actual document. If the art is a glossy abstract nothing that could sit on top of any article ever written, treat it as one more mark against.

Why one tell is not enough

A newsletter that uses an em dash or a clean, numbered structure is not proof of AI. Many careful human writers do both. What you are looking for is a cluster of tells together with the absence of any firsthand reporting.

This matters, because the reflex to shout "AI" at any tidy sentence is its own problem. Plenty of good writers were using em dashes and tight structure long before language models existed. Judge the whole picture: the content, the author, the images, and above all whether anyone actually reported anything.

What about Substack's own AI detector?

In July 2026 Substack added a built-in scan, worked up with an outside detection company, that estimates how much of a post reads as AI-written. You open a post in the Substack Reader, tap the three-dot menu, and choose to scan the text. It is available for posts longer than 100 words published from July 21, 2026 onward.

Use it as one weak input, not a verdict. The rollout drew sharp pushback from writers who called it a witch hunt, and text detectors in general are unreliable: their accuracy drops on edited or paraphrased writing, and they produce false positives, which is unfair to the human who gets flagged. Substack also lets writers add a short "how I make this" note disclosing whether and how they used AI, which is worth reading when it is there. But the manual read you just did, the content, the author, the images, and the sourcing, is more dependable than any score.

Frequently asked questions

Is Substack full of AI slop?

No. Substack has many excellent human writers doing real reporting and real commentary. The problem is that AI-generated filler newsletters now sit next to them, and the two can look alike at a glance. The goal is to tell them apart, not to write off the platform.

What are the clearest signs a Substack post is AI slop?

No original reporting or named sources, generic claims with nothing checkable, an anonymous pen name or AI-generated headshot, a firehose of posts on unrelated topics, and generic stock or AI header images. Look for a cluster of these, not a single one.

Does Substack's AI detector tell me if a newsletter is written by AI?

Substack added a built-in scan in July 2026 that estimates how much of a post reads as AI-written, but detectors are unreliable and have drawn strong pushback from writers. Treat any score as one weak input and rely on the manual read of the writing, the author, and the sourcing.

Does an em dash mean a newsletter was written by AI?

No. Plenty of careful human writers use em dashes and clean structure. One stylistic habit proves nothing. What matters is a cluster of tells together with the absence of any firsthand reporting or named sources.