AI Detection
Understanding and navigating AI content detection
33 posts
The New ChatGPT Watermark: What It Means for Your Writing
ChatGPT text is getting an invisible watermark in the EU. What it means if you write with ChatGPT, who can check for it, and what OpenAI's own tests show.
AI Is Writing UN Speeches Now, and They All Sound the Same
Nine world leaders delivered nearly the same sentence at the UN this year, and flagged AI speeches jumped from 6 to 30 in one session. What happened, and the lesson for anyone whose writing needs to sound like their own.
Why AI Writing Is Full of Em Dashes (and How to Fix It)
Why language models overuse em dashes, how the mark became the internet's favorite AI tell, and a three-pass method for fixing a draft.
What Is AI Slop? The Word Defining Content in 2026
A definitional guide to AI slop: where the term came from, the patterns platforms are penalizing in 2026, and how AI-assisted work stays out of the bucket.
LinkedIn's 2026 AI Slop Crackdown: Why Posts Lose Reach
LinkedIn is suppressing generic AI content with a claimed 94% detection accuracy and a member-facing slop report button. Here is what that means for anyone posting for work.
Substack Readers Can Now Scan Your Newsletter for AI
Substack's new Pangram-powered scan lets readers check any post, note, or comment for AI. How it works, why writers call it a witch hunt, and what to do before your next post.
SynthID Is Becoming the Default AI Watermark: What to Know
What Google's SynthID watermark is, which companies now use it, how to check content for it, and what its expansion into Search and Chrome means for AI-assisted publishing.
Does ChatGPT Watermark Text? What OpenAI Shipped in 2026
ChatGPT ships no text watermark in 2026, but OpenAI has one on the shelf and the EU AI Act just started the compliance clock. What that means for anyone who drafts with AI.
Pangram AI Detector Review 2026: The Tool Substack Uses
An honest review of Pangram, the AI detector powering Substack's reader-facing scans: accuracy claims, outside validation, pricing, blind spots, and the pre-publish self-check writers should adopt.
Claude's Watermark Punishes the Wrong People
Anthropic's new text watermark is easy to remove and easy to miss. The writers most likely to get flagged are the ones who did nothing wrong.
Claude Is Shipping an AI Text Detection API You Can Use Yourself
Anthropic confirmed it is shipping a text detection API that anyone can call to check for Claude's invisible watermark. Here is what that detector can actually establish, the limits Anthropic has published itself, and how WriteHuman is preparing for it.
Claude Is Now Adding Invisible Watermarks to AI-Generated Text
Anthropic is adding invisible watermarks to text generated by new Claude models. Here’s how the system works, what Anthropic says about its limits, and why users who don’t want to publish raw Claude output may choose to substantially rewrite it with tools like WriteHuman.