Research
Stable, source-linked views of Hacker News topics and years, generated from the same verified data as the Attention Lab.
Data through 2026-07-26
The big arcs
Which topics led HN titles each year
Each line tracks a topic's yearly rank among HN titles, #1 at top; follow a line's rise or fall and read a crossing as one topic overtaking another.
Ranks compare within-year title shares, not front-page position, readership, or importance. The retained corpus is complete through 2025; 2026 is partial.
Hacker News ↗Bitcoin / crypto title attention came in waves
The line is the yearly share of retained HN story titles matching bitcoin, crypto, or blockchain. Peaks describe title attention, not asset value, adoption, or cause.
Observed local highs: 2014 1.371%; 2018 2.089%; 2021 1.515%.
The link economy shifted
Two lines track each source's yearly share of high-scoring (50+) HN links: the solid line is GitHub, the dim line hosted blog platforms.
✦ Insights GitHub’s share of high-score Hacker News links increased significantly from 0.5% to 8.3% between 2008 and 2016.
Hosted personal blogs (Blogspot/WordPress/TypePad) fell 5× as Medium rose and code moved to GitHub. Census-only, share-based — not a coverage artifact.
See GitHub links on Hacker News ↗The big shift
Domain breakout rates differ sharply
Each dot is a domain's breakout rate — the share of its HN submissions that scored 100+ — read against the vertical line marking the 4.0% site-wide average.
Personal blogs & trusted institutions dominate; high-volume aggregators (YouTube, Vimeo, news wires) rarely clear 100 points. Census only (2007–2020); domains with 400+ stories. Every rate is a verified claim.
Hacker News ↗High-scoring source ranks changed
Among domains represented by retained stories scoring 100 or more points, GitHub held #1 while TechCrunch, Ars Technica, and NYTimes ranked lower in 2020 than in 2012. Each line shows one plotted domain's within-year rank.
Descriptive rank among domains of retained stories scoring 100+ in each census year. This is not a historical front-page ranking, a measure of reading behavior, or evidence that source type caused engagement.
From “Startup News” to AI-family titles
Two title-only series from the same tracked-topic model: startup/startups and the AI / LLM family (ai, llm, gpt, chatgpt, neural). This is not front-page placement.
HN launched in 2007 as “Startup News.” Both lines are annual shares within retained live-story records. The retained corpus is complete through 2025, and 2026 is partial.
See “startup” on Hacker News ↗Company names occupy less title space
Each line is one company's yearly share of HN story titles; the y-axis is how often the name appears, so lower means a smaller slice of titles.
Partial 2026: Google 1.216%, Facebook 0.077%, Microsoft 0.584%.
All three terms occupy a smaller share in partial 2026 than in 2007. This is descriptive title vocabulary and does not identify a cause.
See these on Hacker News ↗What tech is talking about
Inside the AI vocabulary shift
Three keyword families — ChatGPT, LLM/LLMs, and agent/agents/agentic — are plotted as yearly shares of retained HN titles. The generic agent family cannot isolate AI-agent usage, so interpret it as vocabulary rather than a clean topic measure.
Partial 2026: ChatGPT 0.467%, LLM / LLMs 2.400%, Agent family 6.527%.
Exact title-token shares; the terms can overlap, remain keyword approximations, and partial 2026 is not a complete-year comparison.
See AI agents on Hacker News ↗The language churn
Each line is a programming language's yearly share of retained HN titles; read the slopes across 2007–partial 2026 as title vocabulary, not language adoption.
Partial 2026: Python 0.630%, Rust 1.028%, JavaScript 0.236%.
Exact title-token shares; language names are keyword approximations, and 2026 is partial.
See “rust” on Hacker News ↗Open-source title share rose
The line is the share of each year's retained HN story titles containing the phrase “open source,” from 2007 through partial 2026.
Partial 2026: Open source 1.903%.
Exact phrase-match shares. The current 2026 value is partial, and nearby events do not establish why the share changed.
See “open source” on Hacker News ↗Framework title attention changed hands
Each line is one front-end framework term's yearly share of retained HN titles; crossings describe title vocabulary, not technology adoption.
Partial 2026: React 0.205%, Angular 0.009%, jQuery 0.006%.
Exact keyword shares with complete years through 2025 and partial 2026. “React” can also occur as an ordinary verb, so all three series are approximate topic measures.
See React on Hacker News ↗Container terms rose at different times
Each line is a tool term's yearly share of retained HN story titles. The sequence describes observed title vocabulary and does not imply substitution or cause.
Partial 2026: Docker 0.148%, Kubernetes 0.152%, Serverless 0.044%.
Exact title-token shares with complete years through 2025 and partial 2026. Sequence and co-movement do not establish product substitution or cause.
See Docker on Hacker News ↗Database title attention changed
Each line tracks one database term's yearly share of retained HN titles; slopes describe title vocabulary rather than adoption or market share.
Partial 2026: Postgres 0.381%, MongoDB 0.024%, NoSQL 0.004%.
Exact keyword shares with complete years through 2025 and partial 2026. These are title mentions, not database adoption or market share.
See Postgres on Hacker News ↗Risers & fallers
Two ranked lists of title words — Rising and Fading — with each word's share of titles and how many percentage points it moved from the 2014–16 census to the partial 2026 corpus.
Auto-discovered — the title words whose share of titles moved most from the census (2014–16) to the live, partial 2026 corpus. Share-based (era-safe); HN post-type words and stopwords removed. Each links to its Hacker News search.
Hacker News ↗The bigger picture
Events arrive as spikes from zero
Each stem is one named security event; its height is that event's peak-month share of HN titles, all rising from the same 0% baseline.
✦ Insights Heartbleed reached 2.28% of titles in April 2014, with a peak share that rose from 0.
Named-event keywords are absent for years, then dominate in the exact month of disclosure. The absolute share is the signal — the from-zero z-score is degenerate.
See Heartbleed on Hacker News ↗Raw correlations weaken after detrending
Each row is a topic pair on a fixed −1 to +1 correlation axis. Hollow marks use monthly title-share levels; filled marks use month-to-month changes after first-difference detrending.
The hollow neutral mark uses monthly title-share levels. The filled accent mark uses first differences—the change from one month to the next—to reduce shared long-run trend. Both use Pearson r on the same fixed −1 to +1 scale.
- AI · ML: raw r = +0.460; detrended change r = +0.020. Little linear co-movement remains after detrending.
- blockchain · AI: raw r = -0.059; detrended change r = +0.018. Little linear co-movement remains after detrending.
- Google · Docker: raw r = -0.478; detrended change r = -0.001. Little linear co-movement remains after detrending.
These coefficients describe linear co-movement in the retained monthly title-share series. They do not show that one topic influenced another, and sampling noise or other events can affect either estimate.
HN became a discussion forum
The line is average comments per story each year, plotted across the full census from 2007 on the left to 2017 on the right.
✦ Insights Comments per story increased significantly, rising from 3.2 in 2007 to 7.8 in 2017, demonstrating a 2.4-fold expansion.
Comment volume grew ~32× while stories grew ~13× across the census. Census-complete; the cultural-vs-UI cause is not established.
HN’s very first item ↗The conversation went deeper
Each line tracks one depth bucket's yearly share of all comments — top-level versus replies nested four-plus deep; watch where they cross.
✦ Insights Top-level comments decreased from 48 percent in 2007 to 21 percent by 2018, while deep threads increased around 2015.
Climbing every comment’s parent chain to its story: top-level comments fell as depth-4+ replies grew, crossing ~2015. Census 2007–2018 (every comment); the cultural-vs-UI cause isn’t established.
HN comment guidelines ↗Threads don’t run out of steam
Each bar is a comment-tree depth; its height is the share of comments at that depth that drew at least one reply, in the 2016 census.
✦ Insights Reply probability increased from 46.2 percent at depth 1 to 51.3 percent at depths of 10 beyond.
A reply is about as likely at depth 6 as at the top — even rising to 51% at depth 10+. Census 2016; this is why threads got so deep, and it means the never-replied comments are the short ones, not the deep ones.
How HN threads work ↗The world leaves a mark
Each line is a keyword's yearly share of HN titles — NSA and COVID — sitting near zero except for one sharp spike, then settling back.
- COVID — 3.4% of titles in 2020
- NSA — 1.2% in 2013 (the Snowden leaks)
- Ukraine — 1.1% in 2022 (the invasion)
✦ Insights COVID reached 3.4% of titles in 2020, marking the largest single-event spike observed.
The world surfaces sharply in the conversation, then fades — Snowden (2013), the pandemic (2020), the invasion of Ukraine (2022), all census-exact from the frozen 2007–2022 census. Ukraine's spike was held back until 2022 finished crawling — the thin early sample read 0.6%; the closed census says 1.1%, nearly double. We don't chart a number until we can stand behind it.
See these on Hacker News ↗Most-discussed isn’t most-upvoted
Each bar is a leaderboard size; its height is the share of census stories that land on both the points-ranked and comments-ranked top list at that size.
✦ Insights Only 26 of the top 100 stories by points also appear in the top 100 by comments. At the top 1,000, the overlap is 35 percent.
Ranking the census by points vs by comments gives near-different lists — 16–35% overlap across sizes — and 37 of the 100 most-discussed are Ask HN (mostly “Who is hiring”), though Ask HN is ~3.5% of posts. Discussion and approval are different axes.
The hiring threads on HN ↗Topics
Annual title-share histories for durable technologies and community themes.
- AI agents1.296% of titles in 2025; peak 2026
- AI / LLM12.901% of titles in 2025; peak 2026
- Zig0.112% of titles in 2025; peak 2026
- DuckDB0.067% of titles in 2025; peak 2026
- SQLite0.112% of titles in 2025; peak 2024
- Postgres0.361% of titles in 2025; peak 2024
- Open source1.518% of titles in 2025; peak 2026
- Rust0.883% of titles in 2025; peak 2026
- HTMX0.027% of titles in 2025; peak 2024
- Redis0.048% of titles in 2025; peak 2011
- WebAssembly0.150% of titles in 2025; peak 2022
- Linux0.792% of titles in 2025; peak 2026
- TypeScript0.177% of titles in 2025; peak 2019
- Image gen0.110% of titles in 2025; peak 2022
- Security0.945% of titles in 2025; peak 2018
- Svelte0.025% of titles in 2025; peak 2023
- Kafka0.057% of titles in 2025; peak 2019
- Python0.681% of titles in 2025; peak 2020
- Terraform0.039% of titles in 2025; peak 2020
- Elixir0.063% of titles in 2025; peak 2017
- Privacy0.438% of titles in 2025; peak 2013
- Docker0.150% of titles in 2025; peak 2015
- Startups0.695% of titles in 2025; peak 2007
- MongoDB0.034% of titles in 2025; peak 2013
- JavaScript0.305% of titles in 2025; peak 2013
- React0.257% of titles in 2025; peak 2017
- Kubernetes0.146% of titles in 2025; peak 2019
- Bitcoin / crypto0.518% of titles in 2025; peak 2018
- Serverless0.060% of titles in 2025; peak 2018
- Remote work0.029% of titles in 2025; peak 2020
- Vue0.020% of titles in 2025; peak 2018
- jQuery0.005% of titles in 2025; peak 2009
Years
A year-by-year view of the strongest tracked title themes.
- 2007full calendar year; 32 tracked topics
- 2008full calendar year; 32 tracked topics
- 2009full calendar year; 32 tracked topics
- 2010full calendar year; 32 tracked topics
- 2011full calendar year; 32 tracked topics
- 2012full calendar year; 32 tracked topics
- 2013full calendar year; 32 tracked topics
- 2014full calendar year; 32 tracked topics
- 2015full calendar year; 32 tracked topics
- 2016full calendar year; 32 tracked topics
- 2017full calendar year; 32 tracked topics
- 2018full calendar year; 32 tracked topics
- 2019full calendar year; 32 tracked topics
- 2020full calendar year; 32 tracked topics
- 2021full calendar year; 32 tracked topics
- 2022full calendar year; 32 tracked topics
- 2023full calendar year; 32 tracked topics
- 2024full calendar year; 32 tracked topics
- 2025full calendar year; 32 tracked topics
- 2026partial live year; 32 tracked topics