Hacker News AI Detector
Tool to detect if a Hacker News story is AI-Generated
Visit Website
April 23, 2026
Hey! 👋
As an avid reader of Hacker News, I've started to notice some posts being blatantly AI-generated. Sometimes, there would be comments accusing the author of generating AI slop. I started to wonder, how many posts are actually AI-generated these days? I started collecting data and labeled the posts using AI text detection tools. This page mirrors the Hacker News front page, with posts labeled either Human, Mixed, or AI-Generated.
3 Comments
Most Hacker News regulars have developed a "sixth sense" for those overly polished, rhythmic posts that feel more like a press release than a genuine technical insight. It is a real downer to dive into a promising thread only to realize you’re reading a homogenized summary that lacks any unique, human perspective or "burstiness." Since your tool provides labels like "Mixed" and "AI-Generated," how do you differentiate between a purely synthetic post and a human-written story that was simply run through an AI for grammar and flow?
Right now I'm using the Pangram API to classify text, which labels "Mixed" for text that is AI-Generated but has been modified by humans. The other way around, human text that has been rewritten with AI, I'm not exactly sure how it classifies.
Using an API to catch that middle ground is a smart starting point because the "Mixed" category is where most of the noise happens these days. The real challenge is definitely going to be those human-first stories that just use AI for a quick polish since that often strips away the very burstiness that makes HN content feel authentic.
It is quite similar to my work in Digital PR and Media Placements on sites like MSN or AP News. We focus on building real authority and trust which is something AI slop just can't replicate because it lacks that genuine perspective needed to bridge a trust gap. In both your detector and my PR work the goal is to filter out the noise so the high-value insights can actually stand out.
Keeping an eye on how rewritten text gets flagged will be key as these models get better at mimicking natural human flow.