January 6, 2026
Bad vibes, big clicks
65% of Hacker News Posts Have Negative Sentiment, and They Outperform
Grumpy posts win: HN’s bad vibes get more clicks, and the crowd’s not surprised
TLDR: A study finds 65% of Hacker News posts read as negative and these posts score 27% higher. Commenters split between “of course, all media skews negative” and “the sentiment model is flaky,” with jokers embracing HN’s grumpy reputation—making the debate itself proof that controversy drives attention.
Hacker News has a new king: bad vibes. A fresh study of 32,000 posts says nearly 65% of HN content reads as negative—and those posts score bigger, averaging 35.6 points versus 28. Translation: downer posts get a 27% bonus. The researcher ran six AI models to check tone (think “robots guessing if text sounds grumpy”), and they all agreed. Cue the comments. One user sighed, “Of course negativity drives discussion,” comparing it to mainstream media’s clickbait gloom. Another shrugged that every platform tilts negative, so HN isn’t special—just honest.
But the thread had spice. The skeptics roasted the sentiment tech as a black box that “made a frownie face,” arguing real tech talk is naturally critical, not cynical. Meanwhile, the self-aware crowd owned it: “BREAKING: HN is negative,” with a wink that they might be the problem. One commenter brought receipts, linking past “overwhelmingly negative” threads here and here. The running joke? HN wears “grumpy but correct” like a badge. Whether negativity causes engagement or hot topics get framed as negative, the community’s split between: “yeah, humans love drama” and “the model doesn’t get technical critique.” Either way, the vibe is clear: complain, explain, outperform.
Key Points
- •Negative-sentiment HN posts average 35.6 points versus a 28-point overall average, a 27% premium.
- •The dataset includes 32,000 posts and 340,000 comments, with nearly 65% classified as negative.
- •Negative sentiment skew persists across six models: DistilBERT, BERT Multi, RoBERTa, Llama 3.1 8B, Mistral 3.1 24B, and Gemma 3 12B.
- •DistilBERT results power the dashboard due to efficiency in a Cloudflare-based pipeline.
- •The study also examines decay curves, preferential attachment, survival probability, and early-engagement prediction; a preprint is on SSRN, with code, dataset, and a dashboard forthcoming.