The typical Show HN gets 2 points and no comments. I pulled every Show HN submitted to Hacker News between November 21, 2025 and November 14, 2026 from the public HN Algolia API: 48,331 posts. 62% finished on 1 or 2 points. 1.6% reached 100.
The thing most closely tied to the posts that did break out wasn't the hour, the day or a clever title. It was the maker writing a real explanation in the submission text. URL posts with 400+ words of text reached 10 points 21% of the time; posts with no text managed 6%.
Disclosure: I build Needle, which finds people asking for tools like yours on Hacker News and 13 other networks. Nothing in this study uses Needle data; it's all public HN data you can pull yourself.
This is the data piece. For the rules and the title format, read the Show HN guide. For picking an hour on the day, use the best time to post on Hacker News post and the free HN best-time tool.
In brief
- Most Show HNs go nowhere, and that's the base rate. Median 2 points, median 0 comments. 7.9% reached 10 points, 2.9% reached 50, 1.6% reached 100.
- Write the maker text. Posts with a URL and text did roughly twice as well at the 50-point mark as URL-only posts (4.0% vs 1.9%), and the effect held for GitHub links and own-domain links alike.
- AI is everywhere and underperforms. A third of titles mention AI, LLMs, agents, MCP or a model name. Those posts reached 10 points 6.5% of the time vs 8.6% for the rest.
- Concrete beats generic. Titles with a measurable number ("3kb", "70x smaller", "$1,600") reached 10 points 14.5% of the time. Titles with "free", "platform", "startup" or "for teams" did worse than average.
- The link matters. github.io demos did best (10.7% to 10+). App Store, Chrome Web Store and video links did worst (2.3% and 1.4%).
How I did it (method box)
| Source | Public HN Algolia search API, search_by_date with tags=show_hn |
| Window | Posts created Nov 21, 2025 00:00 UTC to Nov 14, 2026 23:59 UTC (359 days) |
| Fetched | Nov 21, 2026, so the newest post was 7 days old and scores had settled |
| Collection | One request per UTC day (no day came close to the API's 1,000-hit cap; the busiest had 317), about one request per second, raw JSON cached |
| Posts | 48,331 stories with a Show HN title, from 31,685 accounts |
| Spot-check | 16 posts (12 random low scorers, 4 top posts) re-checked against the official HN Firebase API: points, comment counts and timestamps matched exactly |
| Thresholds | 10+ points (someone beyond the maker engaged), 50+ (a real audience), 100+ (a launch day most founders would be happy with) |
What this can't tell you. Everything below is correlation. People who write 400 words about their project probably also built something more substantial, so the text may be a marker of effort rather than a cause. Posts that were killed, flagged or deleted may be missing from Algolia, and posts blocked by HN's new-account restriction never appear at all. Scores are a snapshot from Nov 21, 2026. Algolia gives final scores, not vote velocity, so I couldn't measure how fast top posts picked up their first votes. Small groups (a few hundred posts) have wide error bars, so I've only drawn conclusions where the gap is large or the group is big.
The whole method is reproducible: the Algolia endpoint is public and needs no key.
How many points does a Show HN get?
Every submission starts at 1 point (the submitter's own). Here's where the 48,331 ended up:
| Final points | Posts | Share | Share at or above this level |
|---|---|---|---|
| 1 | 16,615 | 34.4% | 100% |
| 2 | 13,400 | 27.7% | 65.6% |
| 3-4 | 10,027 | 20.7% | 37.9% |
| 5-9 | 4,473 | 9.3% | 17.2% |
| 10-24 | 1,717 | 3.6% | 7.9% |
| 25-49 | 695 | 1.4% | 4.3% |
| 50-99 | 611 | 1.3% | 2.9% |
| 100-249 | 548 | 1.1% | 1.6% |
| 250-499 | 182 | 0.4% | 0.5% |
| 500+ | 63 | 0.1% | 0.1% |
Comments look the same: 62.5% of posts got none, and the median is 0. Even among 1-point posts, 23% had at least one comment, which is often the maker's own.
What to do with this: set your expectations by the base rate, not by the launches you remember. A 10-point Show HN is a top-8% result. If yours stalls at 2, you're in the majority, and it says more about a crowded queue than about your product.
Show HN volume exploded, then HN pushed back
The monthly numbers explain a lot of the "Show HN is dead" talk this year.
| Month | Show HN posts | Median points | Reached 10+ | Reached 50+ | Reached 100+ |
|---|---|---|---|---|---|
| Dec 2025 | 3,520 | 2 | 7.8% | 2.7% | 1.5% |
| Jan 2026 | 4,756 | 2 | 8.3% | 3.3% | 1.6% |
| Feb 2026 | 6,217 | 2 | 5.9% | 2.2% | 1.3% |
| Mar 2026 | 5,435 | 2 | 6.6% | 2.4% | 1.3% |
| Apr 2026 | 3,430 | 2 | 9.6% | 3.7% | 2.4% |
| May 2026 | 2,641 | 2 | 9.9% | 3.8% | 2.2% |
| Jul 2026 | 4,192 | 2 | 9.5% | 3.1% | 1.7% |
| Oct 2026 | 4,261 | 2 | 7.0% | 2.7% | 1.5% |
(I've dropped a few months for space; the partial months at each end and June, August and September follow the same pattern.)
Volume peaked in February 2026 at over 200 posts a day, and the share reaching 10 points fell to 5.9%, the lowest of the year. Arthur Cnops' essay "Is Show HN dead? No, but it's drowning" (Feb 17, 2026) described the same flood; his roughly 4.8k posts for January 2026 lines up with my 4,756.
On March 8, 2026, in an Ask HN thread asking HN to restrict new accounts, moderator dang wrote "We're going to at least restrict Show HNs for a while," and added it applied to new accounts. HN's /showlim page (checked Nov 21, 2026) still says Show HN posts are temporarily restricted for new users. The Show HN guidelines (checked Nov 21, 2026) also now warn against "quickly-generated one-offs; anybody can do that now."
After that, volume fell to 2,641 in May and the 10+ rate rose to 9.9%. Comparing the two periods: Nov 21 to Mar 8 averaged 162 posts a day with 6.9% reaching 10+; Mar 9 to Nov 14 averaged 123 a day with 8.5%. By October, volume had climbed back to about 4,200 a month and the hit rate drifted back to 7%.
What to do with this: if your account is new, expect to need some real participation before a Show HN goes through, and HN won't tell you the exact bar. dang has said that publishing technical criteria would just put them on an LLM checklist. Comment on other people's threads for a few weeks first. It's also the best way to learn what lands.
The biggest signal: explain it yourself
HN lets you submit a URL and text together. About half of URL posts (51%) did. Here's how they compare by the length of that text:
| Submission | Posts | Reached 10+ | Reached 50+ |
|---|---|---|---|
| URL only, no text | 22,335 | 6.1% | 1.9% |
| URL + 1-49 words | 4,746 | 7.0% | 2.6% |
| URL + 50-99 words | 5,479 | 8.3% | 3.4% |
| URL + 100-199 words | 7,769 | 8.6% | 3.3% |
| URL + 200-399 words | 4,511 | 13.6% | 5.7% |
| URL + 400+ words | 1,101 | 21.1% | 10.3% |
| Text only, no URL | 2,390 | 6.1% | 1.3% |
I checked this within link types to rule out the obvious confounder. For GitHub links, posts with text reached 10+ 10.7% of the time vs 6.9% without. For own-domain links, 9.2% vs 5.6%. It held in both halves of the year too.
Two of the five highest-scoring posts of the year were text-only: "I replaced a $120k bowling center system with $1,600 in ESP32s" (2,935 points) and "Subvocal – Talk to your computer without making noise" (849). Text-only posts did poorly on average, though, so the lesson isn't "skip the link." It's that the write-up does a lot of the work.
What to do with this: write 200+ words in the text box: who you are, why you built it, how it works, what's rough, what feedback you want. The guide's maker-comment template is a good skeleton. Write it yourself: HN's guidelines say "Please don't put generated text in HN posts."
Does the link type matter?
| Link points to | Posts | Median points | Reached 10+ | Reached 50+ | Reached 100+ |
|---|---|---|---|---|---|
| github.io (GitHub Pages demo) | 1,172 | 2 | 10.7% | 4.9% | 3.0% |
| github.com repo | 17,331 | 2 | 8.7% | 3.4% | 1.8% |
| Own or other domain | 24,642 | 2 | 7.8% | 2.8% | 1.7% |
| No URL (text post) | 2,390 | 2 | 6.1% | 1.3% | 1.0% |
| Free hosting subdomain (vercel.app, netlify.app, pages.dev, itch.io and similar) | 1,160 | 2 | 4.9% | 1.4% | 0.8% |
| App store or browser extension store | 979 | 2 | 2.3% | 0.4% | 0.2% |
| Video (YouTube, Vimeo, Loom) | 212 | 2 | 1.4% | 0% | 0% |
Store links are the clearest loser: 2.3% to 10 points, and 2 out of 979 reached 100. A store page asks readers to install something before they've seen anything. Video links did even worse.
The github.io result is interesting: a demo page that opens in the browser, hosted next to the code. Two of the year's top ten were github.io pages: "Gemini Pro 3 imagines the HN front page 10 years from now" (3,346 points, the top post of the year) and "Performative-UI – A react component library of design tropes" (1,181).
What to do with this: link to something that works in the browser in one click. If you have a mobile app, link to a web page with screenshots and a short demo, and put the store links there.
Which title words go with traction?
The "with" columns are posts whose title (after "Show HN:") contains the feature; "without" is everyone else.
| Title contains | Posts | 10+ with | 10+ without | 50+ with | 50+ without |
|---|---|---|---|---|---|
| "HN" / "Hacker News" | 442 | 17.2% | 7.8% | 8.1% | 2.9% |
| A measurable number ("3kb", "70x", "200ms") | 626 | 14.5% | 7.8% | 5.3% | 2.9% |
| Any number | 7,983 | 10.3% | 7.4% | 4.1% | 2.7% |
| Rust | 1,069 | 11.1% | 7.8% | 4.7% | 2.9% |
| game | 1,308 | 10.8% | 7.8% | 4.2% | 2.9% |
| open source / OSS | 2,661 | 10.4% | 7.8% | 3.9% | 2.8% |
| browser / WebAssembly | 1,881 | 9.9% | 7.8% | 4.3% | 2.9% |
| "I built" / "I made" | 2,425 | 9.2% | 7.8% | 3.5% | 2.9% |
| Name – description (dash format) | 22,731 | 7.6% | 8.2% | 2.7% | 3.1% |
| AI / LLM / agent / MCP / Claude / GPT / Gemini | 15,757 | 6.5% | 8.6% | 2.0% | 3.3% |
| free | 1,673 | 6.2% | 8.0% | 1.9% | 2.9% |
| SaaS / startup / platform | 1,037 | 5.6% | 7.9% | 2.0% | 2.9% |
| vibe coded | 269 | 5.6% | 7.9% | 1.5% | 2.9% |
| MCP (alone) | 1,273 | 5.4% | 8.0% | 1.0% | 3.0% |
| productivity / teams / business / marketing / sales | 529 | 3.6% | 7.9% | 1.9% | 2.9% |
A few things stood out to me.
AI underperformed, and it's a third of the queue. 32.6% of titles matched my AI pattern. Their 10+ rate trailed in every month of the year, and the gap held inside GitHub links (7.0% vs 9.9%) and own-domain links (6.2% vs 8.1%). Only 18 of the top 100 posts had an AI term in the title. That doesn't mean AI projects can't win; "I built a tiny LLM to demystify how language models work" got 915 points. It means "AI tool for X" is now the default, and defaults don't stand out.
"Free" didn't help. Posts with "free" in the title did worse than average (6.2% vs 8.0%), and so did business-flavoured words. HN readers seem to respond to what a thing is and how it works, not to the offer.
Specific numbers did well. "TinyPDF – 3kb pdf library (70x smaller than jsPDF)" (253 points) and "My $300 embroidery machine needed a subscription, so I built my own app" (743) both put a number on the claim.
The dash format is neutral. "Name – what it does" is the format our guide recommends, and nearly half of posts use it. It didn't move the numbers either way. It's still the clearest format, but it won't do anything for a vague description.
Shorter titles did slightly better. Under 30 characters (after "Show HN:"): 9.7% to 10+. 70+ characters: 7.2%. The median title was 59 characters.
Small groups, big caveats. Question titles (501 posts, 11.8% to 10+) and meta-HN projects (442 posts) did well, but the error bars are wide, and HN-themed tools have a built-in audience you probably don't share.
What to do with this: describe the concrete thing, with a number if there's an honest one. Don't lean on "AI-powered" or "free" to carry the title. If it's open source, saying so went with slightly better results.
When should you post a Show HN?
Share of posts reaching each threshold, by submission time (UTC):
| Submitted (UTC) | Posts | Reached 10+ | Reached 50+ | Reached 100+ |
|---|---|---|---|---|
| 00:00-02:59 | 3,268 | 8.5% | 2.7% | 1.6% |
| 03:00-05:59 | 2,943 | 6.8% | 2.6% | 1.5% |
| 06:00-08:59 | 3,835 | 5.6% | 1.9% | 0.9% |
| 09:00-11:59 | 5,303 | 5.9% | 2.4% | 1.4% |
| 12:00-14:59 | 10,623 | 7.6% | 2.8% | 1.5% |
| 15:00-17:59 | 10,356 | 9.3% | 3.7% | 2.1% |
| 18:00-20:59 | 7,211 | 8.6% | 3.0% | 1.7% |
| 21:00-23:59 | 4,792 | 8.9% | 2.9% | 1.7% |
The 15:00-17:59 UTC block (morning on the US East Coast, early on the West Coast) had the best hit rate, and 06:00-11:59 UTC the worst. The gap is real (about 1.6x at 10+), but it's smaller than the text-length or link-type effects.
Day of week barely mattered. Sunday had the highest 10+ rate (8.8%) and Wednesday the lowest (7.4%). Weekends overall: 8.4%, weekdays 7.8%. Weekends had about a third fewer posts a day (99 vs 149).
One caveat: HN's moderators sometimes re-up promising posts, and I can't tell from Algolia whether a timestamp reflects the original submission.
What to do with this: if you can, post in the US morning, and pick a day when you can stay in the thread. For the live picture on the day, the HN best-time tool shows how busy each hour is right now, and the timing post explains why early votes matter.
How many comments should you expect?
Roughly one comment for every three points, and the ratio held steady across score bands: posts with 10-49 points had a median of 6 comments, 50-99 points had 30, 100-249 had 52, and 250+ had 139. A 100-point launch means around 50 comments to read and answer, which matches the guide's advice to block several hours. Posts well above that ratio were usually debates: "I quit coding years ago. AI brought me back" drew 433 comments on 314 points.
Does reposting a Show HN work?
HN's FAQ (checked Nov 21, 2026) says "a small number of reposts is ok" if a story hasn't had significant attention, and asks people not to delete and repost.
In this dataset, 2,674 URLs were submitted as a Show HN more than once (exact URL match, median gap about 7 days). Of the 2,567 whose first attempt stayed under 10 points, 238 (9.3%) reached 10+ on a later attempt and 49 (1.9%) reached 100+. That's a little better than the 7.9% base rate, so one honest retry isn't hopeless. Changing what you submit (better text, a working demo link) is the bigger lever, going by everything above.
Posting more often didn't change the odds, either way: accounts with one Show HN, 2-4, or 5+ all reached 10 points about 8% of the time.
A pre-post checklist, from the data
- It's not a mobile store link or a video. Link to a page that works in the browser (a github.io demo did best).
- You've written 200+ words in the text box, in your own words: problem, how it works, what's rough, what you want feedback on.
- The title names the concrete thing. A number if there's an honest one. No "AI-powered", no "free" as the hook.
- It's open source? Say so. Small but consistent lift.
- Your account has some history. New accounts are restricted; comment on other threads first.
- You're posting in the US morning, or whenever you can stay in the thread for 3-4 hours.
- You've budgeted for about 1 comment per 3 points, and you'll reply to all of them.
- You've accepted the base rate. 2 points is the median. If it stalls, fix the submission before you try once more, not the timestamp.
For the rules, the maker-comment template and the failure modes, the Show HN guide covers it. If your audience is very technical, Lobsters vs Hacker News is worth a read too, and if you're planning a wider launch, see startup launch directories, ranked and how to launch on Product Hunt.
Ten of the year's top Show HNs
Illustrations, not a recipe. Points as of Nov 21, 2026.
| Post | Points | Link |
|---|---|---|
| Gemini Pro 3 imagines the HN front page 10 years from now | 3,346 | github.io |
| I replaced a $120k bowling center system with $1,600 in ESP32s | 2,935 | text post |
| An e-ink frame that hears birds and draws them as 1800s illustrations | 2,391 | github.com |
| HackerBooks – HackerTube, but for books mentioned on HN | 1,870 | own domain |
| Elevators | 1,680 | own domain |
| isometric.nyc – giant isometric pixel art map of NYC | 1,325 | own domain |
| Performative-UI – A react component library of design tropes | 1,181 | github.io |
| A game where you build a GPU | 964 | own domain |
| SimCity, but with Permits | 941 | own domain |
| I built a tiny LLM to demystify how language models work | 915 | github.com |
What they share is playfulness or a concrete, slightly surprising claim. None of them is a business tool, and only two mention AI at all.
Evidence notes
All checked on Nov 21, 2026.
- Dataset: HN Algolia API,
https://hn.algolia.com/api/v1/search_by_date?tags=show_hn&numericFilters=created_at_i>=START,created_at_i<END&hitsPerPage=1000, one call per UTC day from Nov 21, 2025 to Nov 14, 2026. API docs: hn.algolia.com/api. - Spot-checks: official HN Firebase API (
/v0/item/<id>.json) for 16 items; all matched. - Script: a small resumable Python script fetched each day, cached the raw JSON, and a second script computed the tables. Title features are regex matches on the title after "Show HN:"; "measurable number" means a number followed by x, %, KB, MB, ms, bytes or lines. Link types come from the URL's hostname.
- HN restriction on new accounts: dang's comments of Mar 8, 2026 (47300772, 47300821) in Ask HN: Please restrict new accounts from posting; news.ycombinator.com/showlim.
- Rules: Show HN guidelines, HN FAQ.
- Prior analysis: Arthur Cnops, "Is Show HN dead? No, but it's drowning", Feb 17, 2026.
- Limits: killed or deleted posts may be missing; posts blocked by the new-account restriction are invisible; Algolia doesn't expose vote timing; all findings are correlations.