The Show HN Sequel Window: Ship in a Day or Skip It
On October 31 at 19:56 UTC, Andrej Karpathy posted HackerTube, a YouTube built from the videos Hacker News commenters keep recommending. By November 4 it had 2,864 points and 634 comments. Over the next 84 hours, 23 people posted their own version: HackerTube for books, for Wikipedia, for repos, for Amazon products, for games. If you are thinking of riding a viral post with a Show HN of your own, those 23 posts are a clean test of the timing, because they all copied the same idea in the same week.
The first sequel, 20 hours in, got 1,843 points. The second, 13 hours after that, got 339. The other 19 straight copies got a median of 2 points, the same median as all 522 Show HN posts submitted since HackerTube went up.
What happened after HackerTube
I pulled every story that mentions HackerTube or copies its "X, but for Y" title from the public Hacker News Algolia API on November 4, along with every Show HN posted since October 31. Points and comments are as of that morning. I left out two reposts of HackerTube itself, a duplicate submission, an article about querying the data and the Ask HN complaint, which leaves 23 sequels.
Post | Hours after HackerTube | Points | Comments |
|---|---|---|---|
HackerBooks (books) | 20 | 1,843 | 304 |
Hacker News, "HackerBooks, but for all the posts on HN", and HackerTok | 22 to 24 | 1 to 6 | 1 |
Wikihack (Wikipedia articles) | 33 | 339 | 104 |
HackerHack (a joke, see below) | 35 | 182 | 25 |
HackerTube for Apple TV | 39 | 8 | 4 |
Ask HN: "Isn't anyone worried about this flood of HackerTube-but-for apps?" | 41 | 11 | 11 |
HackerHype, HackerZon, HackerWords, HackerGames, Hnrepos, HackerProducts and others (12 posts) | 36 to 50 | 2 to 19 | 0 to 9 |
HackerTV (HackerTube's videos as 24/7 TV channels) | 53 | 333 | 68 |
Theirshelf, HackerCinema, HackerEverything and one more | 56 to 84 | 1 to 4 | 0 |
Set aside the first two, HackerHack and HackerTV, and the other 19 straight copies earned 80 points between them. HackerBooks alone earned 23 times that.
Move within the first day
The early slots belonged to HackerBooks at hour 20 and Wikihack at hour 33. Once hour 36 passed, fresh nouns alone topped out at 19 points, and by hour 41 the flood itself had sparked an Ask HN. Early timing alone wasn't a guarantee, though: the other day-one copy was a gag pointing at the HN front page and managed only 6 points.
Today a day is enough to get one of these built. Karpathy noted that HackerTube took roughly two hours using Claude Code and Opus 5.5, with him mostly chatting to Claude while it wrote and deployed the code. The HackerBooks maker said Claude built theirs, and HackerTV's creator said a few hours. The real constraint is noticing the original quickly. HackerTube landed on a Saturday evening UTC, so anyone checking HN weekly only saw it after the window shut.
Put the original's name in the title
Both early winners named their source right in the title: "HackerTube, but for books mentioned on HN" and "HackerTube, but for Wikipedia articles mentioned on HN." Readers who had just upvoted HackerTube knew exactly what they were clicking. HackerTV's title omitted the name, but its first line linked HackerTube and credited it with the data.
Naming the original alone wasn't enough. Nearly every late entry copied the "HackerTube, but for" formula and still flopped. Books were the strong second subject because HN commenters have argued about books for close to 20 years, and HackerBooks counted about 16,500 of them. Products, words and repositories didn't draw the same interest. The same rule from our post on marketing an AI product applies: the title has to name a job the reader already wants done.
Say what the AI did, then write the post yourself
The HackerBooks post said it plainly: "Claude wrote almost all of it, including the data pipeline and the site. I mostly steered and made calls, bought the domain name, etc." It got 1,843 points anyway. HackerTV said it was built with Claude Code and got 333. Saying it up front did neither of them any visible harm.
The complaints in these threads were about the volume and about AI-written prose. A comment on HackerBooks that drew 11 replies quoted a line from the post and said, "Please don't put generated text in HN posts." Under Wikihack, one commenter said Show HN had entered its "fast food slop age." HackerHack's whole post was "No AI used for this btw." So say in one sentence what the AI built, and write the post's text yourself. Our post on the tells that make replies sound like AI covers what readers notice first.

Change the format, not the noun
Once HackerBooks and Wikihack were up, swapping in a fresh subject stopped earning points. The two late posts that still did well both reworked the idea instead of copying it.
HackerHack appeared under two hours after Wikihack as "HackerBooks but for Posts Posted on HN," linking to the HN front page itself — a parody of the whole trend. It pulled 182 points, with commenters applauding it as the one Show HN that wasn't AI slop. But that punchline only works once.
HackerTV arrived at hour 53 and earned 333 by doing something genuinely different: it took HackerTube's existing videos and rebuilt them as browsable channels, including a late-night channel that only airs from 11pm to 5am. The late flops merely swapped what appeared in the list; HackerTV changed what you could do with it.
So when two sequels are already on the board, a third noun swap is dead on arrival. Either reinvent how the original gets used, or don't bother.
The decision, when a thread in your space goes viral
Check the clock. If the original is under a day old and nobody has posted a good sequel, you can still be first.
Pick a subject your readers already care about. Books worked on HN because people there have been recommending books for years.
Put the original's name in your title and link it in the first line of the post.
Say what the AI built in one sentence, and write the rest of the post yourself.
If two sequels are already ranking, don't post a third with a new noun. Change the format, or go to the comments.
If you're late, the comments are still open
The HackerBooks thread had 304 comments, and a few came from people who had built something close. One linked a ranking of the most-requested tools on HN since 2008. Another pointed out that a similar site existed back in 2021. Under Wikihack, someone linked the HN data sets on BigQuery and ClickHouse for anyone who wanted to build their own. Those replies reach readers who clicked in for exactly this kind of project, and you don't have to race anyone to post one.
That's the same approach as our Hacker News marketing playbook. A useful comment in a thread people are already reading is in front of them today, while a new post has to climb out of /newest first. Answer what the thread is asking, say that you built your tool, and link it only if it answers the question.
How Social Cue helps
None of this works if you notice the thread too late. The scarce resource in this playbook is attention: spotting that a conversation in your niche is heating up, judging whether a sequel or a comment is the right move, and acting before the window closes.
Social Cue is a Claude Code plugin that keeps watch for you. You register your product as a brand with search tags for its category, the problems it solves and the tools people compare it to, and Claude reads Hacker News, Reddit, X, Threads and Product Hunt in a dedicated Chrome window the way a person would. It scores each thread from 1 to 10 for fit and drafts a reply in your voice. Nothing goes out until you approve it. It runs on the Claude subscription you already pay for, so there is no extra AI bill and no platform API deal that can be repriced or revoked.
Discovery, scoring and drafted replies are free. Pro is $19 a month. Claude submits the replies you approved and handles product posts, Show HN included.