What a 2.6M-post Hacker News title study suggests about length, format and words
Hacker News is a social news site where readers submit links and text posts, and the community votes them up or down. The front page is competitive, and a title is often the only thing a reader sees before deciding whether to click. This page walks through a historical analysis of Hacker News submission titles from 2006 through the end of 2019, drawn from a public dataset of more than 2.6 million titles. The findings are descriptive, not a formula: they describe what happened in that sample, and they are a starting point for thinking about a title rather than a guarantee of performance.
Why Hacker News titles matter
Hacker News was founded by Paul Graham and works in a way that will feel familiar to Reddit users: people submit content, the community votes, and the most upvoted items tend to reach the front page. A front-page placement can send a large volume of visitors to a linked site. The competition is real. Around a thousand posts are submitted on a typical day, and the readership is known for caring about content quality. That combination makes the title a high-leverage piece of text: it has to earn a click from a skeptical audience without misrepresenting the link.
The analysis behind this page used a public dataset of Hacker News submissions from 2006 to the end of 2019. The dataset covers more than 2.6 million titles. The figures below are historical observations from that sample; Hacker News voting behavior, submission volume and audience composition can change, so treat the numbers as a dated reference rather than a current benchmark.
General findings from the dataset
Two baseline facts shape everything else. First, every submission starts with one point. Second, some users can downvote, so a post can end up at zero. The distribution of scores is heavily skewed toward the bottom: almost 66% of all posts never get more than two points, which means most submissions are not upvoted by anyone else at all. The median score is therefore a poor way to compare title strategies, because it sits at the floor for almost every group.
Submission volume per quarter has been fairly stable over the last six years of the dataset. Title length in the dataset is centered on seven words. That low number is not surprising: Hacker News titles cannot exceed 80 characters, which puts a hard ceiling on how much a title can say.
Title length: flat medians, shorter winners
Plotting the median score against the number of words in a title produces a flat line: regardless of length, the median score is 2 or 3. That is a direct consequence of the 66% floor described above. Most posts do not perform, so the median cannot distinguish between a one-word title and a ten-word title.
The picture changes when you discard posts that did not perform and look only at posts with more than five points. Among that higher-scoring group, shorter titles are associated with better scores. The practical reading is not 'short titles always win' but 'a short title does not hurt, and among posts that already have traction, brevity is common.'
A few one-word titles in the dataset illustrate the point. '2048', the release of the game, reached 2,903 points. 'Hyperloop' and 'Atom' also appear as short, high-scoring titles. These are recognizable names with built-in curiosity; the word count is a feature of the subject, not a trick that can be copied onto any link.
Title formats and categories
Some Hacker News submissions follow recognizable title formats. The analysis grouped them into five categories, shown in the legal directory table below. A post can belong to more than one category: a title such as 'Scale (YC S16) is hiring engineers to build infrastructure for AI' counts as both a YC post and a hiring post. That overlap matters when reading the category results, because the same submission can lift more than one group.
Posts about YC companies appear to perform better than the other categories. When the comparison is restricted to posts with more than five upvotes, hiring-related posts come out on top. Both observations are conditional on the sample and the threshold; they describe which formats were associated with higher scores in this historical data, not which format will work for a new submission.
Words associated with higher scores
To look for word-level patterns, each title was split into individual words with stop words such as 'a', 'the' and 'an' removed. Each remaining word was paired with the score of the post it came from, and the pairs were aggregated to find the words with the best associated scores. The result is a scatter plot of words, and a handful of terms stand out.
'Zwiebelfreunde' appeared in seven submission titles, and the median score of those submissions was 121 points. 'Profit/Month' appears because of posts discussing the history of FipLab. 'Mathics' is an open-source alternative to Mathematica. 'Jmap' is an open API standard for modern mail clients. These are topic-specific terms tied to particular stories, not general-purpose title ingredients. A word that scores well because it appears in a few well-received posts is not evidence that inserting the same word into an unrelated title will help.
The 'perfect' title and why it is a joke
Combining the observed signals produces a candidate title: short, mentioning a YC startup, mentioning hiring, and including JMAP or LinuxInsides. The literal result of that combination is the single word 'JMAP', which is not a useful title for most submissions. The more playful composite, 'JMAP (YC S10) Linux Inside is hiring', is a statistical curiosity rather than a recommendation. It stacks several correlated features into one string, and the correlations come from a historical sample in which those features happened to co-occur with high scores.
A better use of the findings is as a checklist. Before submitting, ask: is the title short enough to read at a glance? Does it accurately describe the link? If it uses a recognized format such as Ask HN or Show HN, is the format appropriate to the content? Does it avoid borrowing a high-scoring word that has nothing to do with the submission? Those questions are more useful than chasing a composite title.
Bonus observations
Two side analyses are worth noting. The first ranks the 25 programming languages that appear most often in Hacker News post titles. C is absent from that ranking because it was treated as a stop word during processing, which is a processing artifact rather than a statement about C's popularity. The second tracks how often particular words appear in titles over time, which gives a rough view of which topics rise and fall in the submission stream. Both are descriptive views of the same historical dataset.
| Title format | Category |
|---|---|
| Ask HN:<question>? | People asking HN readers a particular question. |
| Ask HN: who is hiring? | Opportunity for companies to post job listing. |
| Show HN:<project name> | Showing to HN readers something you recently build: Open-source project, side-project or company. |
| <something> YC YYYY <something>? | A post related to a company that did Y-Combinator. |
| Classic post |
A post can overlap multiple categories; for example, a title containing both a YC reference and a hiring phrase counts in both groups. The table is a historical classification of title formats in the 2006–2019 dataset.
Frequently asked questions
Does a shorter Hacker News title always get more points?
No. In the historical dataset, the median score was 2 or 3 regardless of title length, because most posts never get more than two points. Shorter titles were associated with better scores only among posts that already had more than five points. Length is not a guarantee of performance.
Which Hacker News title format performed best in the analysis?
Posts about YC companies appeared to perform better than other categories in the overall sample. When the comparison was limited to posts with more than five upvotes, hiring-related posts came out on top. A single submission can belong to both groups, so the categories are not mutually exclusive.
Should I put a high-scoring word like 'zwiebelfreunde' in my title?
No. Words such as 'zwiebelfreunde', 'Profit/Month', 'Mathics' and 'Jmap' scored well because they appeared in a small number of well-received posts about specific topics. The association is with those stories, not with the word itself. Inserting an unrelated high-scoring word does not reproduce the result.
Is there a character limit on Hacker News titles?
Yes. In the dataset, Hacker News titles could not exceed 80 characters, which is one reason the title-length distribution centers on seven words. The limit is a platform rule, not a finding of the analysis.