We're talking Google, Apple and AI: 19 Years of HN data analyzed
Article URL: https://www.riteshkhanna.com/blog/hacker-news-data-analysis Comments URL: https://news.ycombinator.com/item?id=47617823 Points: 1 # Comments: 1
47.5M
Total Items
6M
Stories
41.4M
Comments
19
Years of Data
Hacker News has been the tech industry’s collective consciousness since 2007. Every startup launch, every security breach, every paradigm shift in programming. It all flows through this one orange page. I queried a complete public archive of 47.5 million items spanning nearly two decades, hosted on Hugging Face and analyzed with DuckDB, and ran every query I could think of. What follows is what the data says about us.
The Growth Curve
From 23,000 stories in 2007 to nearly 400,000 per year by the 2020s, Hacker News has grown 17x in story volume alone. But the real story is in the comments: they’ve grown from 70K to over 4 million per year, a 60x increase. The ratio of comments to stories has shifted from roughly 3:1 to over 10:1. The community didn’t just grow; it became dramatically more conversational.
The pandemic year of 2020 saw the biggest single-year jump in story submissions (428K, a 15% spike), while comments peaked at 4.2 million in 2023. There’s no sign of a plateau. Even in 2025, stories hold steady at 394K and comments at 3.5 million.
What Hacker News Talks About
The single most dramatic trend in the dataset is the AI explosion. In 2008, the word “AI” appeared in about 200 story titles. By 2022, that had grown to around 6,600, respectable growth but nothing earth-shattering. Then ChatGPT launched. In 2023, AI mentions spiked to 29,434, a 4.5x jump in a single year. By 2025, the count has reached 36,625, completely dwarfing every other topic on the platform. Nothing else even comes close.
Crypto tells a more interesting story, a boom-and-bust narrative with two distinct peaks. The first came during Bitcoin’s 2013–2014 breakout (4,700 mentions), and the second during the ICO and DeFi crazes of 2017–2018 (12,142 mentions at peak). The 2021 NFT wave pushed it to 9,794 before a sharp decline. By 2025, crypto mentions have returned to roughly 2,100, below where they were in 2015. The hype cycles came and went, but the sustained interest didn’t materialize the way it did for AI.
Click the topics above to explore. Note how “Startups” peaked in 2012–2013 at nearly 10,000 mentions and has been declining ever since, possibly a sign that HN’s identity has shifted from a startup community to a broader tech one.
The Platform Wars in Links
Medium’s arc on Hacker News is a cautionary tale of platform dependency. Submissions from medium.com rose from zero in 2011 to a plateau of 18,000+ per year from 2016 through 2020, making it one of the most-linked domains on the entire site. Then it cratered. By 2024, Medium links had fallen to just 2,763, an 85% decline from the peak. The culprit? Aggressive paywalls, the pivot to the Partner Program, and a general erosion of trust among the developer community.
Substack filled much of the gap, rising from 140 links in 2018 to 7,494 in 2025. But the real constant is GitHub. It has been the top linked domain by volume since 2015 and keeps growing, reaching 21,407 links in 2025. arXiv is the quiet riser, tripling from ~1,300 in 2020 to 3,436 in 2025 as AI papers became front-page material. TechCrunch, once the second-most-linked domain, has faded from 5,100 in 2012 to under 1,900.
When to Post for Maximum Impact
There’s a persistent myth that weekday mornings US time (early afternoon UTC) are the best time to post on Hacker News. The data tells a more nuanced story. Sunday mornings around 7–8 AM UTC(which is Saturday night to early Sunday on the US West Coast) have the highest average scores in the entire dataset, at 16.1. Saturday evenings UTC also perform well, with several slots hitting 15+.
The explanation is straightforward: weekends see less than half the submission volume of weekdays. Tuesday at 11 AM UTC gets 52,660 posts across the dataset; Sunday at 7 AM gets only 17,961. Fewer posts competing for the same number of upvotes means each post has a better shot at visibility. The worst time to post? Wednesday at 3 AM UTC (a 9.8 average score), with maximum competition from the European morning crowd, minimum attention from Americans who are asleep.
Average Votes by Day & Hour (UTC)
Lower score
Higher score
0
3
6
9
12
15
18
21
Sun
Mon
Tue
Wed
Thu
Fri
Sat
Hour (UTC)
Ask HN vs. Show HN vs. Links
“Show HN” posts have consistently outperformed both “Ask HN” posts and regular links on average score. In 2010, Show HN averaged 18.1 points while links averaged just 10.1; the community has always rewarded builders. By 2022, Show HN hit 20.8, though the gap narrowed as link scores also climbed to 19.2.
The most interesting shift is in 2024: for the first time, regular links (19.4 avg) outperformed Show HN posts (17.9 avg). Meanwhile, Ask HN dropped to 11.8, its lowest since 2012. It appears that as the AI wave brought a flood of new projects, the bar for what constitutes an interesting “Show HN” rose. Link posts, increasingly pointing to substantial AI research and tooling, captured more attention.
The Domains HN Loves Most
Among domains with at least 500 submissions, blog.ycombinator.com unsurprisingly tops the quality chart at 95.7 average score. Posts from YC’s own blog carry weight. But the runner-up is more telling: Dan Luu’s personal blog at 93.3. A single developer writing detailed, opinionated technical analysis outscores OpenAI (79.2), Stripe (74.0), and every major tech company except YC itself.
The list is a who’s who of developer-beloved domains. Paul Graham’s essays (58.2),Julia Evans’ explainers (53.7), Simon Willison’s AI explorations (52.8), and Rachel by the Bay (49.0) all make the cut. The common thread: these are all individual writers producing deep, technical content. The only pure corporate domains in the top 20 are apple.com, stripe.com, and sec.gov (yes, SEC filings go viral on HN).
Top 20 Domains by Avg Votes
Avg votes per submission — minimum 500 submissions
Brevity Wins
Short titles win on Hacker News, and the data is unambiguous. Titles under 20 characters have an average score of 15.8 and a 4.05% chance of hitting 100+ points. At 40–59 characters (the most common bucket, with nearly 2 million posts), the average drops to 12.2. By the time you hit 100+ characters, it’s 5.9 average with a 0.19% hit rate. That said, fewer than 530 titles in the entire dataset are that long, since HN’s title limit naturally keeps things concise.
The sweet spot appears to be under 40 characters. That’s enough for a clear headline but short enough to feel punchy. Think “Show HN: My Weekend Project” not “Show HN: I built a comprehensive full-stack application for managing personal finances with AI-powered insights.”
The Words That Go Viral
What do the biggest hits on Hacker News have in common? We looked at every title that scored 1,000+ points and counted the most frequent words (excluding stop words). “Google” dominates at 137 appearances, nearly double “Apple” at 87. The tech giants command attention: Amazon (38), Facebook (54), Microsoft (29), and YouTube (29) all make the list.
But the most revealing entries aren’t the company names. “Died” appears 43 times; obituaries and memorial posts for influential technologists reliably go viral. “Open” and “source” together appear 107 times, reflecting HN’s deep open-source ethos. “Down” (33 times) captures outage posts, which are perennial attention magnets. “Free” (28 times) and “first” (26 times) suggest that novelty and accessibility are reliable hooks.
The People Behind the Posts
Hacker News has a fascinating power-user dynamic. The most prolific poster, rbanffy, has submitted 36,842 stories, an average of roughly 5 per day for nearly two decades. But volume doesn’t equal quality: rbanffy’s hit rate (posts scoring 100+) is just 2.1%. The most efficient prolific poster is ingve, with 17,087 submissions and an 8.2% hit rate, meaning about 1 in 12 posts reaches the front page with a strong score.
The quality leaderboard tells a different story. Sam Altman (sama) tops it with a staggering 211.6 average score across 125 posts and a median of 100. Virtually everything he submits does well. gdb (Greg Brockman, OpenAI co-founder) follows at 204.8. The whoishiring bot, which posts the monthly hiring threads, sits at third with 171.7 average, a beloved fixture of the community. Notably, kragniz (141.4 avg) and jart (126.4 avg) have huge averages but median scores of 3–4, suggesting a few massive viral hits pulling up their averages.
Most Prolific Posters
UserPostsHit %Avg Votesrbanffy36,8422.1%8.8Tomte26,2113.1%11.8tosh24,1574.2%15.2bookofjoe20,6862%9.3mooreds20,5041.9%8pseudolus19,9445%17.7PaulHoule19,1211.1%6.2todsacerdoti18,8866.4%23.9ingve17,0878.2%26.2thunderbong16,0573.1%11.7jonbaer14,1782.3%9.8rntn13,4101.4%8doener12,8983%12.3Brajeshwar12,6081.9%8.5LinuxBender11,0581.1%6.2
Highest Quality Posters (min 50 posts)
UserPostsAvg VotesMedian Votessama125211.6100gdb59204.8116whoishiring531171.7101kragniz50141.44meetpateltech511137.17jart55126.43ddevault116124.211raphlinus53123.3116whatok72111.34craigcannon206110.744schoen54106.33input_sh50104.23pc6610319isp85102.93antirez84102.322
The Power Law of Success
The distribution of successful posters follows a steep power law. If we filter to only the ~48,800 users who have ever had a post score 100+ points (out of millions of accounts) and ask how they got there, the largest category is “Persistent”: 17,276 users who submitted more than 5 posts but only cracked 100 once. The next group is “Repeat hitters” (15,211 users with 2–5 hits), followed by 7,687 “Lucky shots” who managed it in just 2–5 total posts.
At the very top, only 939 users qualify as “Power hitters” with 20 or more posts scoring 100+. That’s the entire HN “elite”: fewer than a thousand people out of millions of accounts. And 3,507 users are literal one-hit wonders: they submitted a single post, it scored 100+, and they never posted again. Lightning struck once and they walked away.
Moderation, Dead Posts & the Question Question
Hacker News’s moderation is famously opaque, but the data reveals clear patterns. The dead post rate (stories flagged, killed, or detected as spam) peaked at 31.25% in 2011. That was the Wild West era: the site was growing fast, spam detection was primitive, and the community was still developing its immune system. Since then, the rate has stabilized around 20–22%, meaning roughly 1 in 5 submitted stories doesn’t survive. It dipped to 15% in 2013–2014 and again in 2019 (15.8%), but has consistently returned to the low twenties.
Meanwhile, statement titles consistently outperform question titles. From 2011 onward, statements average 1–3 points higher than questions in nearly every year. The one exception is 2022, where questions actually edged out statements (16.1 vs. 15.2), a year when “Ask HN” discussions about layoffs and the tech downturn resonated strongly. But the general rule holds: if you want upvotes, state your case; don’t ask a question.
The Show HN Explosion
Show HN went from 393 posts in 2010 to 28,372 in 2025, a 72x increase. Everyone is shipping now.
Weekend vs. Weekday Content
Normalized for volume, weekends and weekdays look remarkably similar. Show HN is 4.1% of weekend posts vs. 3.9% on weekdays. The biggest gap: fun/hobby projects are proportionally more common on weekends, while startups and security dominate weekdays.
Topic Share by Day Type
Percentage of total posts in each category
Score Inflation
The top 1% score went from 37 in 2007 to 301 in 2025, an 8x increase. The median has barely moved (stuck at 2). Getting noticed is harder; going viral pays off more.
Score Percentiles Over Time
How the bar has shifted for top posts
The Graveyard
Most stories on HN vanish without a trace. In 2025, 68% of stories received zero comments and 40% got zero or one vote. Only about 3% of submissions ever reach 100+ points.
Stories That Disappeared
Percentage of stories with no engagement
What Does HN Ask About?
The most common word in Ask HN titles is “startup” (6,475 appearances), followed by “good,” “software,” and “app.” The community asks for career advice, learning resources, and tool recommendations.
Most Common Words in Ask HN Titles
Top 20 words (excluding stop words)
When Do Viral Stories Get Posted?
The success rate peaks at 7–8 AM UTC (3.7%), drops during peak hours, then climbs again in late evening. Early birds and night owls face less competition.
% of Stories Reaching 100+ Points by Hour (UTC)
Higher = better odds of going viral at that hour
The All-Time Leaderboard
The 20 highest-scoring stories in Hacker News history. Stephen Hawking’s death at 6,015 points is the all-time #1. The xz backdoor (4,549) and CrowdStrike outage (4,489) are the top non-obituary posts.
Top 20 Stories of All Time
#TitleScoreCommentsYear1Stephen Hawking has diedby Cogito6,01543620182A Message to Our Customersby epaga5,77196720163Backdoor in upstream xz/liblzma leading to SSH server compromiseby rkta4,5491,84920244CrowdStrike Update: Windows Bluescreen and Boot Loopsby BLKNSLVR4,4893,85920245Steve Jobs has passed away.by patricktomas4,33836320116Mechanical Watchby todsacerdoti4,29841320227YouTube-dl has received a DMCA takedown from RIAAby phantop4,2401,44320208Reflecting on one very, very strange year at Uberby grey-area4,1071,01320179Google’s copying of the Java SE API was fair useby pdoconnell4,103932202110Replit used legal threats to kill my open-source projectby raxod5024,0221,274202111How I cut GTA Online loading times by 70%by kuroguro3,883697202112Bye, Amazonby grey-area3,8161,097202013Sora: Creating video from textby davidbarker3,6472,231202414Google Search Is Dyingby dbrereton3,6361,561202215Every Google result now looks like an adby cmod3,592972202016Show HN: This up votes itselfby olalonde3,53182201217A search engine that favors text-heavy sitesby Funes-3,441716202118Slack has raised our charges by $195k per yearby JustSkyfall3,4061,470202519My First Impressions of Web3by natdempk3,3931,129202220F.C.C. Repeals Net Neutrality Rulesby panny3,3841,4312017
What the Data Says About Us
Nineteen years of Hacker News data paint a portrait of a community that has grown dramatically without losing its core identity. The site remains overwhelmingly focused on technical depth. Individual developers with personal blogs still outperform corporate PR machines on average score. The topics shift (from startups to AI), the platforms shift (from Medium to Substack), but the appetite for substantive technical content is remarkably consistent.
If there’s one takeaway for anyone who posts on HN, it’s this: build something interesting, write a clear and short title about it, and post it on a quiet Sunday morning. The data says that’s your best shot. Everything else is noise.
Sign in to highlight and annotate this article

Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
analysis
Governments Lock Down Biometric IDs — Investigators Get Left Outside
sovereign biometric data silos are fundamentally changing the landscape for developers in the computer vision and OSINT space. As nations from Guyana to Sri Lanka finalize their digital ID frameworks, we are seeing a shift from public-facing identity data to encrypted, state-controlled biometric repositories. For developers building investigative tools, this represents a technical "Access Wall" that cannot be bypassed with traditional scraping or simple API integrations. The technical implication is clear: the data is moving from the public web into secured, sovereign databases. For the solo private investigator or the small fraud-detection firm, the gap between "what is possible" and "what is accessible" is widening. While government agencies leverage multi-million dollar biometric stacks
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Products
Desktop Canary v2.1.48-canary.33
🐤 Canary Build — v2.1.48-canary.33 Automated canary build from canary branch. Commit Information Based on changes since v2.1.48-canary.32 Commit count: 1 a7e3d198df 🐛 fix(chat-input): memoize mentionOption/slashOption to prevent freeze on paste ( #13551 ) (Arvin Xu) ⚠️ Important Notes This is an automated canary build and is NOT intended for production use. Canary builds are triggered by build / fix / style commits on the canary branch. May contain unstable or incomplete changes . Use at your own risk. It is strongly recommended to back up your data before using a canary build. 📦 Installation Download the appropriate installer for your platform from the assets below. Platform File macOS (Apple Silicon) .dmg (arm64) macOS (Intel) .dmg (x64) Windows .exe Linux .AppImage / .deb
Desktop Canary v2.1.48-canary.34
🐤 Canary Build — v2.1.48-canary.34 Automated canary build from canary branch. Commit Information Based on changes since v2.1.48-canary.33 Commit count: 1 e364b9a516 ✨ feat: skill store add skills tab ( #13568 ) (Rdmclin2) ⚠️ Important Notes This is an automated canary build and is NOT intended for production use. Canary builds are triggered by build / fix / style commits on the canary branch. May contain unstable or incomplete changes . Use at your own risk. It is strongly recommended to back up your data before using a canary build. 📦 Installation Download the appropriate installer for your platform from the assets below. Platform File macOS (Apple Silicon) .dmg (arm64) macOS (Intel) .dmg (x64) Windows .exe Linux .AppImage / .deb
Desktop Canary v2.1.48-canary.35
🐤 Canary Build — v2.1.48-canary.35 Automated canary build from canary branch. Commit Information Based on changes since v2.1.48-canary.34 Commit count: 2 25cf3bfafd 🐛 fix(userMemories): i18n for purge button ( #13569 ) (Neko) 3cb7206d90 ✨ feat: create new topic every 4 hours ( #13570 ) (Rdmclin2) ⚠️ Important Notes This is an automated canary build and is NOT intended for production use. Canary builds are triggered by build / fix / style commits on the canary branch. May contain unstable or incomplete changes . Use at your own risk. It is strongly recommended to back up your data before using a canary build. 📦 Installation Download the appropriate installer for your platform from the assets below. Platform File macOS (Apple Silicon) .dmg (arm64) macOS (Intel) .dmg (x64) Windows .exe Lin





Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!