FAQ

A catalog of the 9.4K blogs that Hacker News readers keep sharing. A blog is listed when at least 15 different HN accounts linked to it, in a story submission or in a comment, between 2006 and 2026-08-21. Each blog page shows its most-shared posts, a sample of the HN comments that linked to it, how many people shared it each year, and other blogs that the same people shared.

From Andrej Karpathy's HackerTube (HN discussion), a YouTube-style site made only from videos that HN commenters recommended. He built it in about two hours by asking an AI coding agent (Claude Code with Opus) to do the work. A day later, HackerBooks (HN discussion) did the same for books. HackerBlogs uses the same method for blogs.

Almost all of it was built by an AI coding agent (Amp, running Claude) from a short instruction: “build HackerBlogs from the HN dataset, the same way as HackerTube”. A person reviewed the result, asked for changes, and supplied the domain. The steps:

1. Get all of HN. The open-index/hacker-news dataset has every HN item since 2006: 49.3M stories and comments in about 12 GB of Parquet files.
2. Pull out every link. DuckDB reads each story URL and every link inside comment HTML: 10.7M links from 518K accounts. Deleted and flagged items are skipped.
3. Group links by site. A site is a host name (without www.). On shared platforms, each author is a separate site, for example medium.com/@name or example.edu/~name. 35K sites were shared by 15 or more people.
4. Find the blogs. Claude Sonnet labeled each site from its top HN titles and most-shared URLs: personal blog, company blog, group blog, or not a blog (news sites, social networks, code hosting, docs, shops and so on). One blog on several hosts, such as blog.x.com and x.com, is merged into one entry.
5. Build each blog page. Most-shared posts, people per year, up to 30 HN comments that link to the blog (one per person), and related blogs, calculated from the overlap between the people who shared each pair of blogs.
6. Describe each blog. Claude Sonnet wrote the one-line description, the “What HN says” line (based only on the sampled comments) and the topics.
7. Thumbnails. The preview image (og:image) of a blog's most-shared posts, when it has a wide one. Other blogs get a plain card with the post title.
8. Publish. The result is static HTML, JavaScript and JSON files on Cloudflare Pages.

The number of different HN accounts that linked to the blog at least once. It counts links, not praise: a link in a critical comment counts the same as a recommendation. One person who links the same blog many times counts once.

Usually one of these: fewer than 15 different people linked to it; it is on a shared host that is grouped as one site; the AI labeled it as “not a blog” (for example a project homepage with a blog section); or its links appeared after 2026-08-21, when the dataset snapshot ends.

They were written by an AI from post titles and a sample of HN comments, so expect some mistakes. Some post titles come from the URL when no HN story named the post.

There are no accounts. The site remembers the topics of the blogs you open and follow, and the blogs related to them, in your browser's local storage. It uses them to reorder the list. Without history, the order is shuffled once a day.

Anonymous usage analytics with PostHog: page views, searches, filters, follows, and clicks to blogs. There are no accounts and no personal profiles.

Yes. The About page has zip files with all the data and a DATA.md file that explains it. You can give them to your own AI agent and ask for a different interface, a newsletter, or an analysis.