Substack just handed its writers a tool intended to fight “AI slop,” but it may have inadvertently triggered a human identity crisis. By partnering with a little-known startup called Pangram to offer built-in AI text detection, the platform has divided its creative heart. While writers feel like they’re being frisked for “bot-like” behavior at the door, another industry—the legal sector—is looking at the same technology and seeing a multi-million-dollar insurance policy for their reputation.
| Attribute | Details |
| :— | :— |
| Difficulty | Intermediate (Requires platform & legal compliance knowledge) |
| Time Required | 10–15 Minutes to audit current workflows |
| Tools Needed | Pangram, Substack Disclosure Tools, Copyleaks (Alternative) |
The Why: The War on “Slop” and the Death of Trust
We are entering the “Great Correction” of the generative AI era. After eighteen months of flooding the internet with synthetic text, the novelty has curdled into a phenomenon known as “AI slop”—low-value, repetitive content that clutters search results and feeds.
Substack launched Pangram’s detector to protect its brand as a haven for “pure” human thought. But for law firms, the stakes are different. They aren’t worried about being “boring”; they’re worried about being sued or losing client trust. If a $1,000-an-hour associate submits a brief that a client’s scanner flags as 90% AI-generated, that isn’t just a style issue—it’s a professional liability. This intersect of technology and liability is why we are seeing such a massive AI legal analysis shift in how firms audit their output.
How to Navigate the New AI Detection Landscape
- Scan Before You Publish: Use the “Scan for AI Text” feature located in the Substack post editor (via the three-dot menu). Do not wait for readers to do it.
- Audit for “False Positives”: Recognize that statistical classifiers like Pangram prioritize “perplexity” and “burstiness.” If your writing is highly structured, technical, or if you are a non-native English speaker, you are at a higher risk of being flagged as a bot.
- Use the “How I Made This” Disclosure: Instead of hiding AI usage, use Substack’s new disclosure feature. Transparency often kills the “gotcha” power of a detector. Many creators are finding that tools like Raptive Duet help them maintain this transparency while scaling their reach.
- Implement Firm-Wide Standards: If you are in a legal or corporate environment, establish a “human-in-the-loop” certification process. Don’t just ban AI; verify the human “finish” of the final work product.
💡 Pro-Tip: Detectors often flag “Listicle” styles and passive voice as AI-generated. To lower your “AI score” without changing your message, intentionally break the rhythm. Insert a personal anecdote, use a slang term, or vary your sentence lengths dramatically. These “human” imperfections are exactly what Pangram is looking for.
The “Buyer’s Perspective”: Pangram vs. the Field
Pangram is positioning itself as the high-end, “enterprise” answer to the detection problem. Unlike free tools that offer vague “percentage” scores, Pangram is marketing “protection” and “liability mitigation” specifically to law firms.
While competitors like GLTR or GPTZero target teachers and students, Pangram is following the money into the legal vertical. The value proposition is simple: authenticity is becoming a luxury good. However, there is a technical catch. No AI detector is 100% accurate. For a law firm, relying solely on a scanner to “prove” a lawyer wrote a brief could lead to devastating false accusations and damaged careers. This is why many organizations are turning to a dedicated AI Knowledge Hub to ground their generative tools in verified facts rather than probabilistic guesses.
The Scientific Side-Effect: Corrupting the Record
The Guardian recently noted that this “slop” doesn’t just annoy readers—it’s breaking science. Amateur birdwatchers using AI-photo enhancers are inadvertently “hallucinating” new features onto birds. They’re uploading photos of birds with the markings of different species or placing them in the wrong hemisphere. This isn’t just a bad photo; it’s data pollution that could skew biodiversity research for decades.
FAQ
Can Pangram really prove I didn’t write my own article?
No. It can only tell you how closely your writing matches the statistical patterns typical of large language models. It is a probabilistic guess, not a forensic fact.
Will using AI as a research tool trigger a “fail” on these scanners?
Usually, no. If you use AI to brainstorm and then write the content yourself, you’ll pass. If you copy-paste and “tweak” a few words, you will likely get flagged.
Why are writers citing GDPR in their protests?
Under Articles 21 and 22, individuals have rights regarding automated decision-making. Writers argue that if a platform uses an AI to secretly “label” or de-rank their work without a human review, it violates those protections.
The Bottom Line: Current AI detection technology is an “opinionated” tool that can successfully identify lazy writing, but it remains a blunt instrument that often punishes neurodiverse and technical writers.