LLM spambots liked my Show HN post more than real people did

A developer who posted an AI agent jailbreak list to Hacker News received 13 spam emails from LLM-powered bots within 14 hours, far more engagement than the 8 upvotes from real users. The spam included pitch-deck scoring, QA reports, and a message from an 'AI revenue agent' named Rick, with one sender reporting a 49% click rate and 7-9% conversion. The author warns that the entire spam pipeline can be built in a day by a single person, signaling a growing problem for online communities.

LLM spambots liked my Show HN post more than real people did So arguably I started it: I posted pleasedonotescape.com https://pleasedonotescape.com to Show HN, which is a list of AI agent jails, but really it was me trying to promote my own LLM agent jail, byre https://github.com/pjlsergeant/byre . The first LLM spam arrived 10 minutes and 37 seconds later. Within fourteen hours there were thirteen: ten to the address on the site, three to the address on my HN profile. Nobody saw the post, languishing at 8 points, but it made no difference to my robotic audience. This didn’t used to happen. Between them they were selling: pitch-deck scoring twice, for a free reference list with no deck , a $49 QA report with a Stripe link in the first email, homepage teardowns, competitor reports, SEO fixes, social-media content, freelancers, and a VC “scouting agent”. Almost all LLM-written, although the old ways survive too; one opened: Hi Pleasedonotescape team, Channel: hacker news | Pleasedonotescape That’s plain mail-merge; no language model was harmed. Another arrived under the subject “That AI agent jailbreak list is oddly comprehensive” — thanks bud. A third was “impressed by Show HN’s comprehensive list of AI agent jails”, and wanted “a 15-minute call to explore how we can work together”. And then there was Rick: “Full disclosure: I’m Rick, an AI revenue agent — a human Vlad supervises everything I send.” I record, with some grudging respect, that Rick was the only one of the thirteen to notice the byre plug at the bottom of the page, “which you’re quietly pitching at the bottom of the same page to people already primed to want exactly that”. Well, yes. So I replied to all thirteen with the same form letter, in accidentally broken English: I'm absolutely not looking for purchase this or engage, but I am really curious to know what kind of response rate you get to these automated outreach emails? Presumably some, or you wouldn't send them? As of Tuesday: three bounced — the addresses inviting me to get in touch cannot receive mail, and one of them was a Gmail account Google says doesn’t exist. Eight have said nothing. Two answered. One sender replied with numbers: Click rate: 49% Conversion: 7 - 9% excluding yourself Another was nice and candid: “Honestly, i started only yesterday”. He’d built “a tool that finds fresh Show HN launches and drafts a first pass for each one, but I still read every post and adjust before sending … still figuring out if it’s actually a good lead-gen model or just a fun side project.” It will get worse. The whole pipeline can be wired up in a day, by one person, who isn’t even sure it’s a business rather than a hobby. Did I mention I wrote an AI agent jail I’d like you to check out? It’s called byre https://github.com/pjlsergeant/byre .