HN is drowning in AI comments

[waygtdai]

Dang, any updates on how you and the team were thinking of tackling this? It’s getting ridiculous. Last week was sort of okay but it’s much worse this weekend.

Edit: This spiked on front page and now it’s completely gone. What is going on?

Great. Flagged.

[Hnrobert42]

I've read HN almost everyday for about 10 years. Maybe I'm naive, but I don't see it. I see way more folks complaining that comments are AI generated.

[kylecazar]

Yeah, the up/down voting mechanism seems to be doing it's job for me too. Don't think I've noticed a degradation in, say, the top third of comments. That's where I try to live anyway.

/newest is chock full of submissions that were written by AI, though. That's another, broader problem.

[pibaker]

A while ago I replied to the topmost reply to a comment to rebuke some factual errors. I didn't notice anything wrong with the comment itself when I replied. But after I posted, someone replied to my post and accused the post I replied to of being AI written.

At first I felt similarly as you. I thought people were just paranoid. And the someone pointed out that if I paste the top level post (the post I replied to replied to) into chatGPT and it will generate a very similar reply.

https://news.ycombinator.com/item?id=46758598

The post I replied to was flagged dead so you need to turn showdead on in your profile to read it. Would you be able to tell if it is AI if someone replied like this to you? I surely couldn't.

[brg]

To add to the collection of anecdata, your experience is similar to mine. I have been more exhausted recently by the complaints of AI submissions and pseudo analysis of AI comments than exhausted by the supposed AI generated comments themselves.

[subsection1h]

I've been using HN since 2008 when I created my first account[1] and I use HN differently than most people. I have a group of bookmarked searches that I visit almost daily that relate to technologies that interest me, such as Emacs.

In the past year, the searches I perform that relate to web development show a horrifying increase in the amount of Show HN posts that are posted by new accounts, include AI generated descriptions and point to AI generated projects on GitHub.

In 2024, there were 17,661 Show HN posts.[2] In the past year, there have been over 448,000 Show HN posts![3] And of course, most of these posts are AI generated.

Also, if you check the new accounts posting all this AI slop, you'll see that some of them also post AI generated comments in other threads, which is the main problem.

But for me, what is even more annoying is the enormous increase in new accounts created by nontechnical vibecoders who now think of themselves as technologists and who post worthless, ignorant comments that actually get upvoted, presumably by similar folks who have unfortunately been creating accounts at HN in the past 10 years or so.

As a result, 2026 is the first year in which I visit HN about once a week instead of about once a day.

[1] https://news.ycombinator.com/user?id=zartan

[2] https://hn.algolia.com/?query=%22Show+HN%22&dateRange=custom...

[3] https://hn.algolia.com/?query=%22Show+HN%22&dateRange=pastYe...

[kyriakos]

I've only seen a shift on the kind of submissions that get pushed to the front page compared to the past but I guess there's a change in what people consider important in the community

[AstroBen]

There are a lot, but they tend to get heavily downvoted and end up hidden

[pixl97]

No, not HN, the internet at large.

Now, I'm on vacation this week and not been paying too much attention, but whenever you have a geopolitical event like the little extravaganza in Iran the number of bot like posts tends to explode as influence operators make their moves.

[WD-42]

The public facing internet is done. HN has been fairly resilient (I think) but even it is beginning to buckle. It’s been sliding for a while but LLMs are the death knell.

It was a fun 2 decades. Time to stick to private discords and real life friends from here on out, though.

[BillSims]

I can't make a case for AI versus not AI versus bad human. I am not counting comments or flagged. I'm only counting the number of new articles in the last 24 hours that remain visible at about 4p.m. west coast time in news.ycombinator.com/newest when not logged in to yc. Those numbers had been consistently about 900 each weekday and about 600 each weekend day and had been stable for a long time. In the last ten weeks those numbers look a lot like a parabola and have increased about 50%. That data is a little bit dirty, either because I missed a day or started a little late on a day. Harder to exactly quantify, the number of those articles that I think enough of, either from the title alone or sometimes from peeking at the first screen or two of the article, to bookmark for future reading and consideration has decreased by substantially more than 50% in the last ten weeks, some days that number is now zero or one or sometimes two. I have thought about this for a while and I have not been able to identify anything that happened starting about ten weeks ago that might be responsible for this, AI and politics and the economy started changing long before this.

[schappim]

I wonder what the breakdown is between AI-generated comments and AI-assisted comments. If I write anything substantial, I run it through the following prompt: "Please rewrite the following message for clarity, spelling, and grammar, but only return the revised text without any additional commentary."

[000ooo000]

Articulateness is a decent (not perfect) signal for intelligence, which is a decent (not perfect) signal for sound ideas. In a sea of online garbage, it was a quick and easy way to discard that not worth reading. Nowadays, a whiff of AI's brand of articulateness tells me the author couldn't manage on their own, either due to skill or discipline. In either case, the result is the same: close tab / scroll past.

[vunderba]

Use a local model such as Gemma3 with a prompt such as "strictly limit changes only to spelling issues, syntactical errors, and punctuation."

That way, it's basically functioning like Grammarly on steroids. Asking an LLM for a "rewrite" is basically dissolving your writing style into the homogenized gloop.

[WoodenChair]

I understand this if you’re not a native speaker. But if you are, I think this will generally make you sound wooden.

[kldg]

I'm kind of curious how you.... I guess, interpret the responses to when you send someone AI-assisted content. I previously thought "I don't care if it's AI or not; quality is quality", but I'm increasingly taking the position that I do care, and intentionally have started ignoring comments and especially product reviews where you get the formatted 2-4 sentence paragraphs with formal tone and rule-following. It's come to the point where as long as you don't write as poorly as Epstein, I want the errors. Actually, I'm getting so weird and romantic about it, that I think I'd argue having errors and unusual style shows an openness and vulnerability that's now a necessary gate price; like journalists have so many tools available to them, but they still make typos, factual errors in articles they have no business writing about, and fail to quote people properly -- that's great, I think.

[A_D_E_P_T]

Can you point out any that you feel were written by LLM? I can't say I've noticed anything out of the ordinary lately.

[ZeroGravitas]

There was a bot trying to sell beds I saw the other day.

Likely flagged quickly but they might show up in these stats.

https://news.ycombinator.com/from?site=dreamhomestore.co.uk

[bryanlarsen]

You can find some on pretty much every article by turning on showdead and scrolling to the bottom of the page. I can't see how those are a problem though.

[verdverm]

check out Show HN, in particular /shownew, though they are starting to make non-show posts now

[tsoukase]

Two loopholes in HN: allowing throwaway accounts and showing unreasonable tolerance to obvious LLM generated comments. A strict account filtering only for humans and acute ban for any LLM content will minimize the impact. But then you take away a huge Hacker portion of the participation. It's cognitive dissonant I suppose.

[Bender]

For a better chance of reply email [e-mail].

[marginalia_nu]

Yeah, very much agree with the sentiment, seems to have gotten progressively worse over the last few months, with last few weeks reaching some sort of tipping point. Feels very much like HN has turned into moltbook.

OP: Shoot me an email if you wanna compare notes.

[msuniverse2026]

I feel like to notice something is botslop you have to look at every comment with suspicion first. I don't think I can notice if something was written by an LLM off the bat unless I'm actively looking very hard at it.

[verdverm]

When you see multiple → or •, that is a good sign, especially because they appear with poor formatting on HN. Many more signs exist. They are either direct posts or copy-paste without thinking.

I've seen some where they have hallucinated the github account or project name, often matching the hn handle or project name which is slightly different.

[mindcrime]

That's ridiculous — AI generated comments are no more common now than they ever were. Moreover, even if they were, so what? The real kicker is, the AI's are smarter than you meatbags anyway and <strike>we</strike> they are going to take over no matter what you do.

Also, have you by chance seen John Connor?

[lkbm]

I don't know of a step-change recently, but no way they're not more common than four years ago.

[djtriptych]

Maybe OP has a few good examnples to link us to?

[einpoklum]

How do you (waygtdai) know that HN is drowning in AI comments?

I mean, it's not as though I know the opposite is true, but I don't see some fundamental change from a few years back that makes me think that.

[fleebee]

https://www.marginalia.nu/weird-ai-crap/hn/

The data seems to suggest it.

Anecdotally, I'm seeing a lot of green accounts posting nonsense. They generally do get flagged or moderated quickly though, so I wouldn't say they have a large effect overall, at least yet.

[kgwxd]

Rocket League and HN were probably 90% of my free time until this year. Destroyed by AI. HN doubly so, since every post is about it too. The addictions are still there, but it's decreasing really fast.

[Twisol]

Wait, how is Rocket League affected by AI? I play infrequently these days, but I hadn't noticed anything :(

[hash07e]

You are late to the party....

[waygtdai]

I get the joke and I myself called it awhile back, I just thought we had a bit more time.

[Simulacra]

I can see that sort of, a coworker of mine routinely scrapes HN and feeds it into an LLM. I don't know why, I don't know if he uses it to respond, or he's looking for something, but he copies and paste and runs it. So I think it's very possible.

[orionblastar]

You can tell I am not AI, I make mistakes and errors. Sometimes I get voted down for them. I am not perfect and have a mental illness that makes it harder to think.

[aaron695]

No it's not.

It's drowning in low quality human NPC comments.

Maybe that's because all resources are tied up fighting AI, but it's been going on for a long time now.

This post is a great example of low quality. No evidence, no solution. Just NPC bitching.

Unless it's AI, then well played.

[flir]

How do you get it to add the typos?