Hacker News banned AI comments: what changes for outreach

A comment three replies deep in a Hacker News thread says exactly what you monitor Hacker News for: someone naming your category, admitting they're evaluating options right now. You paste it into a model, ask for a reply, edit it for thirty

A comment three replies deep in a Hacker News thread says exactly what you monitor Hacker News for: someone naming your category, admitting they're evaluating options right now. You paste it into a model, ask for a reply, edit it for thirty seconds, and post it inside two minutes of the thread appearing. An hour later, a reply under yours says it reads like an LLM wrote it. Three more people pile on. Your comment gets flagged, then it's gone. You weren't wrong about the timing. You were wrong about what Hacker News's rules on AI-generated comments actually let you post once you found it.

That's not a vibe anymore. It's on the guidelines page.

HN wrote the unwritten rule down

Hacker News has always run on what longtime users call case law: the written guidelines are short, and a lot of actual enforcement lived in moderator judgment calls that never made it onto the page. The public guidelines now include one plainly: don't post generated text or AI-edited text, because the site is for conversation between humans.

Moderator dang confirmed in a discussion thread from March 2026 that the line had already been enforced as unwritten policy, and that this was the first time it moved onto the guidelines page itself. The thread wasn't a quiet edit nobody noticed.

The earlier Ask HN thread, asking whether replies that cite an AI's answer should be forbidden outright, predates the formal guideline by roughly three months. The community was already arguing about this before the moderators wrote anything down, which is the same order operations usually run on Hacker News: users flag a pattern, mods enforce it inconsistently for a while, and eventually it graduates to the page everyone quotes.

The rule targets what you post, not that you used AI

The instinct is to read this as an anti-AI stance. Dang has been specific that it isn't. In the same thread, he wrote that HN isn't asking people not to use AI, only asking people not to post AI-generated comments.

We aren't asking people to not use AI. What we're asking is not to post AI-generated comments to HN.

He drew a real distinction inside that, too: a non-native English speaker or a disabled user running their reply through a model for grammar isn't the target, and he said he doesn't want vulnerable users punished for trying to improve a comment they wrote themselves. Wholesale generation, where the model produces the argument and not just the phrasing, is the thing the rule is aimed at.

That last part is the one worth sitting with if you're using an AI reply generator for outreach anywhere, not just on HN. The rule doesn't care what tool touched your keyboard. It cares what the output reads like once it lands.

This lands hardest on the exact workflow you're running

If you monitor Hacker News comments for buying signals, the fast move has always been to draft off the thread, not off a template, because comments carry more real intent than posts. That part of the playbook doesn't change. What changes is the shortcut a lot of people were quietly taking at the drafting step: read the comment, hand it to a model, post what comes back with a light edit.

That shortcut was already the weaker move before the rule existed. The reply rate math on Hacker News has always rewarded specificity over speed, and a model asked to respond to a stranger's comment without your actual product knowledge tends to produce something generically on-topic and structurally uniform, which is exactly the tell that gets a comment flagged now. The rule mostly closes off the version of fast that was never working anyway.

The version that still works is the slower one: read the comment fully, note the specific thing they said, and write two or three sentences addressing that thing in your own words. If a model helps you check a technical detail before you post, that's closer to what dang described as fine. If a model wrote the reply and you copied it, that's the thing the guideline now names.

Show HN is a separate, still-open thread

Buried in the same discussion, dang mentioned the mods were "restricting Show HNs for now" as a related but separate move, and declined to give specifics, saying he didn't want it turned into "extra steps on some checklist." Nothing about that has landed in the written guidelines yet. For a founder planning a launch there, the actual submission rule has always been narrower than people assume: Show HN is for something you built and can run yourself, not a pitch. Whatever additional AI-related restriction is coming will most likely sit on top of that existing bar, not replace it. Treat it as a thing to watch before your next launch, not a settled rule to plan around yet.

Reddit runs on a version of the same logic, just enforced by different mechanics. Reddit's own reply norms rely on subreddit moderators reading the room rather than a platform-wide AI clause, but the underlying test is the same one HN just wrote down: does this read like a person who actually engaged with what was said, or does it read like something dropped in from outside the conversation. Neither platform is policing the tool in your hand. Both are policing whether the reply sounds like you were there.

● FAQ

Is it against Hacker News guidelines to post AI-generated comments?

Yes. HN's current guidelines state it directly: don't post generated text or AI-edited text, because HN is for conversation between humans. The rule was already enforced as informal moderator case law before it was written into the guidelines page in March 2026.

Does this mean you can't use AI at all on Hacker News?

No. Moderator dang has said directly that HN isn't asking people to stop using AI, only to stop posting AI-generated comments. Using a model to fix grammar or translate reads differently to him than pasting a model's output as your reply, though he's also said the line has real grey area and comes down to judgment calls.

How does Hacker News catch AI-generated comments if there's no detector?

It doesn't run a classifier over every comment. Enforcement is moderator judgment plus community reporting, the same way HN has always policed sock puppets and vote manipulation. That means a reply that merely reads like AI, even one a human actually wrote, carries the same practical risk as one that is.

Does the rule affect Show HN launches too?

Dang has mentioned restricting Show HN specifically, without detailing the mechanism yet in the guidelines. For founders launching there, that thread is still developing and worth watching separately from the comment rule.

What should I do instead of pasting an AI-drafted reply into a Hacker News thread?

Read the actual comment, use AI at most to help you think or check a fact, and write the reply yourself with specifics from the thread. It's slower than pasting. It's also the version that survives the guideline and the reader's eye at the same time.