15% of Hacker News’ front page is now AI-written.
We’ve checked 14,671 articles from the Hacker News front page over the last 176 days. Every linked story gets run through an AI detector and logged. The number keeps going up.
— The headline figure
of front-page articles in Oct '26 were flagged as AI-generated by Pangram.
— Key findings
Three things we learned.
You can reproduce all of it from the open data. Pangram, the detector we use, reports 99.98% accuracy on its own benchmarks.
The first month we tracked sat at 6%. The latest reads 15%. This counts the front page only, not the web at large.
Open source / docs run the highest AI share of any source type, across 1,705 articles. Personal blogs and open-source docs sit far lower.
Posts written straight to HN stay almost entirely human. The AI turns up in the articles people link to, not in HN itself.
— Month by month
The AI-written share, every month we’ve tracked.
— By source type
Where the AI is coming from.
— Methodology
How it works.
A script checks the Hacker News front page every 15 minutes. For each of the top 30 stories it opens the linked page and pulls out the article text. That text goes to Pangram, an AI-content detector. Pangram reports 99.98% accuracy on its own benchmarks.
Pangram rates each article in chunks of about 225 words. We average the chunks into one score, then drop it into a bucket:
— Caveats
What this doesn’t tell you.
- No detector is perfect. Pangram is accurate, but it still gets things wrong. Most misses are false positives on text that’s been heavily edited or translated.
- A high score means the text reads the way AI models write. It can’t tell you who actually wrote it.
- The HN front page is whatever its users vote up. This measures that, not the web at large.
- For each article we store the URL, the score, and Pangram’s chunk-by-chunk breakdown. The breakdown is what powers the segment view.