How to search Hacker News effectively
Hacker News has no search of its own. The box at the bottom of news.ycombinator.com hands your query to the Algolia HN Search index — a keyword index over every story and comment since 2007. It is fast and complete, but it answers only one question: which posts contain these words? The question you usually have is what people actually concluded. grephn sits on the same index and adds filtering and AI thread compression on top.
Rank by points, not relevance
Algolia's default ordering over-weights exact title matches. For research questions, popularity is a better proxy for quality: a 2018 thread with 900 points usually beats a 3-point match from last week.
- Sort by points to find the canonical thread on a topic.
- Sort by date when recency matters — releases, incidents, model launches.
- Search comments, not stories when you want opinions and war stories. Most of HN's value lives in comments.
Narrow the time window aggressively
A bare keyword search spans 18 years. Almost every question has an implicit horizon, and setting it removes more noise than any other single filter: 24 hours for a morning scan, a week for a release you half-remember, a year for tooling comparisons, all time for "has anyone solved this".
Filter by author and by source
- Author — a domain expert's comment history is a curated reading list you didn't have to build.
- Source — restricting to one publication turns HN into a filtered feed for that outlet, ranked by how the crowd received each post.
Search the subject, not the keyword
Keyword search fails on categories. No single word retrieves "distributed systems posts" — you would have to OR together Raft, Paxos, consensus, sharding and CAP, and still miss the good ones. grephn's Subject lens matches on topic instead of string: Programming Languages, Infrastructure, Databases, Artificial Intelligence, Theoretical Computer Science, Hardware.
Compress the thread instead of reading it
The cost of HN research is reading, not finding. A 500-comment thread holds maybe six distinct positions plus one buried correction that invalidates the top comment. grephn's compression returns the positions, the strongest counter-argument, and the consensus, with comment scores attached. See worked examples of compressed threads.
A practical workflow
- Start broad in time, narrow in topic.
- Sort by points to find the canonical thread, then by date to see what changed.
- Switch to comments when links stop being useful.
- Compress anything over ~100 comments rather than skimming.
- Bookmark as you go — the second search costs as much as the first.