Hacker News for AI Products: What Survives the Front Page
What HN upvotes and what it flags, how a Show HN differs from a first-person technical post, and how to handle the comments without damaging the product.
Published 2026-08-22 · 4 min read
Hacker News is the highest-variance community available to a technical founder: a front-page post can produce more qualified attention in a day than a month of everything else, and a mishandled one produces a permanently indexed thread of criticism with your product name in the title.
The variance is not random. What survives follows consistent patterns, and most of them are about the post rather than the product.
What HN upvotes
- Something technically novel, explained by someone who built it. The mechanism is the content; the product is incidental.
- Original data or measurement. A benchmark, a cost analysis, a systematic comparison. Numbers nobody else has.
- A first-person account of a hard problem. What broke, what you tried, what actually worked. Specificity is the credibility signal.
- A contrarian position defended with evidence. Not contrarian for its own sake — the evidence is what carries it.
- Something genuinely useful and free. A tool, a dataset, a reference implementation, with no gate.
What gets flagged or ignored
| Pattern | Why |
|---|---|
| A launch announcement | Read as marketing. "Show HN" exists for this and has its own norms |
| Marketing language anywhere in the post | "Game-changing", "revolutionary", "the future of" — instantly discounts everything else |
| Voting solicitation | Asking colleagues to upvote is detectable and can get the account and the domain penalised |
| A thin wrapper presented as novel | The audience knows the model landscape better than most founders and will say so |
| Vague claims without numbers | "Much faster" invites "than what, measured how" |
| A post that only links | No context, no author present. Usually dies at one point |
Show HN versus a technical post
Two different things with different expectations, and choosing wrong is a common mistake:
| Show HN | First-person technical post | |
|---|---|---|
| What it is | A product submission, explicitly labelled | A write-up about something you learned |
| What the audience wants | To try it and tell you what is wrong | To learn the thing |
| Title format | "Show HN: [product] – [one line on what it does]" | The finding or the problem, not the product |
| Product mention | Expected and appropriate | Secondary, often only in the last line |
| Comment tone | Critical and specific, about your product | Critical and specific, about your claim |
| Best when | The product is usable and you can answer questions all day | You have something genuinely interesting to explain |
Writing the post
- 1
Title: plain and specific
Say what it is or what you found. No adjectives, no capitalisation games, no clickbait. The title is the entire hook and understatement works better than emphasis here.
- 2
First paragraph: what and why
What you built or found, and what problem it addresses. Assume the reader is technical and impatient.
- 3
Middle: the actual substance
How it works, what the numbers are, what was hard. This is the part that earns the upvotes and it is the part most founders shorten in favour of describing features.
- 4
Include the limitations
What it does not do, where it breaks, what you would need to add. Stating limits pre-empts the top comment and signals that you are describing rather than selling.
- 5
End with availability, not a call to action
A link and whether it is free. Do not ask for feedback, upvotes or signups — asking is what marks a post as marketing.
The comments, which are the real event
A front-page post generates hundreds of comments, many of them critical, all of them permanently indexed against your product name. How you handle them matters more than the post did.
- Answer the technical objections first, with specifics. A detailed reply to a hard criticism is the single most reputation-building thing available in the thread.
- Concede what is true, immediately. "You are right, it does not handle that case yet" costs nothing and buys credibility for everything else you say.
- Do not argue about taste. If someone says the approach is wrong, explain your reasoning once and move on. Extended defence reads as insecurity.
- Never ask anyone to change a comment or downvote. It is visible, it is resented, and it escalates.
- Stay for the whole day. An abandoned thread with unanswered criticism is the worst outcome, because the criticism stands as the last word.
- Take the useful parts offline. If three people identify the same problem, that is your roadmap, and saying so in the thread is both honest and well received.
Timing and expectations
- Post when the audience is awake. Weekday mornings in US time zones see the most activity; a post that gets no early traction rarely recovers.
- One attempt, or one attempt per genuinely different thing. Resubmitting the same product after it died is noticed and resented. A substantially different release is a legitimate new post.
- Expect the traffic to be poor quality for conversion. This audience is evaluating, not buying. The value is feedback, credibility and links — not signups.
- Do not do it before the product is stable. A front-page post that breaks your infrastructure is a bad day twice over.
Hacker News does not reward a good product. It rewards an honest technical account from someone who stays to answer for it — which is a narrower and more achievable thing.
More in Community playbooks
- How to Find and Join AI Newsletter CommunitiesNewsletter authors and their reader communities are a reachable audience for a new AI product. How to find them, get on their radar, and get mentioned.
- Product Hunt Community: How to Engage Before You LaunchA launch-day platform rewards the people who were already there. How to build standing in a launch community in the months before you need it.
- How to Answer Questions in Public Without SellingThe helpful-answer channel: how answering strangers' questions builds an audience for an AI product, what a good answer contains, and where the line is.