How much of HN is AI?

Security researcher lcamtuf spent February doing something deceptively simple: logging what actually ranked in Hacker News's top five each day. The results were about what you'd expect, and also worse.

Only three days in the month passed without an AI story in the top five. On February 5th, every single slot went to AI-related content — including one lcamtuf flagged as covert marketing for an AI vendor. The dominant category throughout was vendor announcements.

The more interesting part was what came next. lcamtuf ran Pangram, an LLM-text-detection tool he describes as conservative and accurate, across the top-ranked stories to find out how many were written by AI rather than humans. He then manually reviewed every flagged post to check the tool's work. His conclusion: the flags held up. And Pangram missed a few — meaning the real share of AI-generated content is probably higher than the data shows.

One post he singled out: a February 19th story titled "AI is not a coworker, it's an exoskeleton," which he assessed as carrying multiple hallmarks of LLM authorship. His broader point about why detection works is worth noting — it's not that AI writing is obviously inhuman, but that today's models converge on the same default voice. The same stylistic fingerprints show up across outputs reliably enough to catch.

Separate research from Viktor Löfgren at marginalia.nu adds weight to the picture. Löfgren scraped HN's newcomer comment endpoints and found newly registered accounts use em-dashes, arrows, and similar typographic tics at nearly ten times the rate of established users — 17.47% versus 1.83%, with a p-value of 7e-20. New accounts also referenced AI and LLMs significantly more often.

For anyone tracking the agent ecosystem, the implications are specific. Hacker News has long been an organic discovery layer for developers and early adopters — the audience that decides which frameworks and tools get traction. If that feed is being gamed by the same vendors whose products it's meant to surface, the platform's value as a read on genuine developer sentiment starts to erode.