research_paper

Show HN Metrics: Everyone Told You To Answer Every Comment. Somebody Counted.

September 14, 2026 · 14 min read · Scout7

Show HN metrics: across 188,085 posts, comments barely track use (r = 0.10) and upvotes track it better (r = 0.29). How to read your launch day.

Show HN Metrics: Everyone Told You To Answer Every Comment. Somebody Counted.

Why a busy comment thread feels so big, what the upvote count really tells you, and how to read your own launch day honestly.

The day your project got looked at

You built something at night and at weekends. One day it worked well enough. You put the link up on Hacker News, on the page called Show HN, which is where programmers post the things they have made.

For one day, people looked.

Then the day ended. The page went quiet. And you were left with a question you could not answer. Did any of that turn into one single person using your thing?

People say Show HN metrics. That just means the two numbers you get on that day. A comment count and an upvote count. One of them feels like the score. It is the wrong one.

The numbers in this piece come from one large piece of work. Daniel King read 188,085 Show HN posts, from 2012 to April 2026, and checked which of those two numbers actually told you anything. You can read it here: Show HN by the Numbers.

Here is what he found, in short.

The comment count barely tells you anything about whether people used the thing.
The upvote count tells you a bit more. Only a bit.
Half of all Show HN posts score 2 points or fewer.
Almost all of the effect is over inside two days.

The hard part is simple. The number that feels most important is the worst one to trust.

Why a busy comment thread feels like a win

Think of a small shop on its opening day.

People come in. They lean on the counter. They ask good questions. They stay twenty minutes. It feels like the best day you have ever had.

Then you close up and count the money in the till. There is almost nothing in it. The people who bought something said nothing at all. They picked a thing up, paid, and left.

Launch day is that shop. The busy comment thread is the people at the counter.

Comments are not a waste of your time. They do real work.

They are the fastest product feedback there is. A stranger with no reason to be kind tells you in the first hour what is confusing.
They keep your post visible. A page with people talking on it stays up longer than a dead one.
On Hacker News they are how the community decides whether you belong there.
And when you have built something on your own for months, someone replying is worth a lot.

So the mistake is not "comments are useless". The mistake is smaller and more exact. It is reading the count as your result.

Here is how that was checked. King compared the comment count on each post against the stars its code got on GitHub. A star is what you press on GitHub when you like somebody's code. It is easy to count, so it is one of the few things you can actually measure after a launch.

He scored how well the two moved together. That score is written as r. It runs from 0 to 1. A 1 means the two numbers always move together. A 0 means one of them tells you nothing at all about the other.

Comments against stars scored r = 0.10.

That is close to nothing.

His own words: "Discussion doesn't mean conversion."

Here is what that looks like on one morning.

Post A gets 40 comments and 11 upvotes. The maker replies to every single one, all day long.
Post B gets 4 comments and 90 upvotes. The maker barely says a word.
Post B is the one people went and used.

Those two are not real posts. They show what the pattern looks like across all 188,085 of them.

One more number helps here. The median Show HN post scores 2 points. Median means the middle one. So half of everything posted there gets 2 points or fewer. Fifty points puts you in the top 6 out of 100.

Eleven points only feels like failure because you think most posts do much better. Most posts do not.

So read the thread closely. Just do not read the count as your verdict.

The other number on the page is a bit better

If the comment count is the wrong number, what is the right one?

Upvotes. An upvote is the small arrow people press when they like a post.

In the same work, upvotes against stars scored r = 0.29.

That is still low. Most of what happens is still down to something else. But it is the better of the two numbers, and it is the best one a person launching a side project can see for free.

King also gives a rough rule. About 1.4 stars on GitHub for every upvote, in the first 48 hours. So thirty upvotes is somewhere around forty stars. That is not a promise. It is there so you know what size of thing to expect.

So the practical thing is short.

Read the comments to learn what confused people.
Read the upvotes to get a rough idea of how much interest there was.
Treat both as early clues, not as a final mark.

Before you lean on either of them, it is worth knowing where these numbers came from.

Where these Show HN metrics come from, and what they miss

This is one person's work, published on his own website. Other researchers have not checked it. That does not make it wrong. It does mean you should know what is behind it.

The star part of it is much narrower than the rest.

188,085 posts sit behind the broad numbers.
Only 491 posts sit behind the star numbers, and those were the top 500 by score.
Every one of them scored 258 points or more.
So the star findings describe launches that already went well. Not ordinary ones.

Even the better number leaves most of the story out. An r of 0.29 explains about 8 out of every 100 cases. The other 92 are down to something the numbers cannot see.

And there are limits to what a star means at all.

A star on GitHub is not a user. Someone can press it and never open the code.
A user is not a customer.
Two numbers moving together does not prove that one caused the other.
A launch that went well can still go nowhere afterwards.

So use these numbers. Do not treat them as the truth about your project. This is the only proper count anybody has done. It is still not a perfect one.

Most of it is over in two days

This is the part that surprises people.

About 92 out of every 100 stars a project gets from its launch arrive inside the first 48 hours.

For a project near the top, day one brings about 509 stars. Day two brings about 40.

So if you posted on Monday, whatever the launch was going to do, it has already done it by Wednesday. Looking again on Friday teaches you nothing new.

There is a reason the numbers are small for most people. The room got crowded.

From 2012 to 2019, Show HN ran at about 30 posts a day.
In 2025 it ran 28,302 posts. That is about 200 a day.

Hacker News is now limiting how many Show HN posts go up. Its own notice says: "We're temporarily restricting Show HNs because of a massive influx, mostly by users who aren't yet familiar with the site or its culture." You can read it on the Show HN limits page. People had been asking for something like it for a while, in threads such as Ask HN: Please restrict new accounts from posting from 18 March 2026.

That is context. It is not a complaint and it is not a reason to stop.

There is a small edge in when you post. King found that Monday at 00:00 UTC, which is Sunday at 7pm on the east coast of the United States, gives a 10.8 in 100 chance of clearing 50 points, against about 9 in 100 on an average day.

That is a small edge. Treat it as small.

What the launch guides are telling you instead

You may have read the opposite advice this year. A lot of launch guides now say that comments matter more than upvotes, and that you should chase long, thoughtful comments on your launch page.

You can find that claim in Product Hunt guides published in 2026, including ones from LaunchPact and Forkoff.

The idea is not silly. Sites do change how they rank things, and they do not always say so.

The problem is that nobody has shown the work.

Product Hunt does not publish how its ranking works.
None of these guides says how many launches it looked at.
None of them says how it checked.

So it is a belief. It may even turn out to be right. But it is not a measurement, and it should not be read as one.

This is how launch advice spreads. Once enough people repeat a rule, it starts to sound like a fact. Then the next guide copies it.

There is a simple test you can run on any launch advice. If it tells you something confident about ranking, and it does not say how many launches it looked at, treat it as a guess.

What to read in the first 48 hours

There are two things to do. That is all.

One. Read the upvote count against the real number, not against your hopes.

Half of Show HN posts get 2 points or fewer. Fifty points puts you in the top 6 out of 100. Write those two numbers down before you post, so you are judging your launch against what actually happens rather than against what you imagined.

Two. Know that a quiet post is not a dead post.

This is the part almost nobody outside Hacker News regulars has heard. Hacker News runs a second chance queue, and people run it by hand.

Daniel Gackle moderates Hacker News. He posts there as dang. On 30 April 2021 he explained how it works:

"Moderators and a small number of reviewers go through old submissions looking for articles that are in the spirit of the site... These get put into a hopper from which software randomly picks one every so often and lobs it randomly onto the lower part of the front page."

So a post that got missed can go back up. And you can ask. In the same place he wrote:

"If you see a submission that didn't get attention and which you think is particularly good for HN, please tell us at hn@ycombinator.com! We love getting those requests and usually add them to the pool."

Then he added one line worth reading twice:

"It's fine if it's your own article, but we like it better when it's just something you ran across and recognized as good."

So write to them at hn@ycombinator.com. And be the sort of person who also puts other people's work forward, because that is what they actually asked for. The queue is public and you can look at it at news.ycombinator.com/pool.

That is a calmer way to end launch day than deciding the whole thing flopped.

What to look at after the thread goes quiet

After two days, close the tab.

The thread did its job. Its job was to tell you what confused people. It was never your score.

Stars on GitHub are the closest thing to a real signal you can see from outside. In King's set, about 27 out of every 100 Show HN posts linked to code on GitHub, so a lot of projects have that one visible number.

Be careful even with that.

A star is not someone using your thing.
Someone trying it once is not someone who kept it.
The comment count is weaker than either of those.

So after launch day, watch the things that sit closer to what people actually did.

People arriving at your site straight from the post
Whether anyone touched the code after the first rush
Sign ups or installs, if you have them
Messages from people using it, not just people commenting on it
Whether anyone came back a second time

Every one of those is a person doing something. A comment is a person talking. They are not the same thing, and only one of them keeps your project alive.

Frequently asked questions

Do comments on a Show HN post matter at all?

Yes. They are the fastest product feedback you will ever get, and answering them properly is how the community decides whether you belong. What they are not is a score. In King's count of 188,085 posts, the comment count and the stars a project got moved together at r = 0.10, which is close to no link at all.

Comments or upvotes. Which number should I trust?

Upvotes, but only a little. Comments against stars scored r = 0.10. Upvotes against stars scored r = 0.29. Both are low. The second one is the less bad of the two, and it is the best number you can see without paying for anything.

How many points is a normal Show HN post?

Two. The median post, meaning the middle one, scores 2 points. So half of everything posted there gets 2 points or fewer. Fifty points puts you in the top 6 out of 100. Most people do not know this, so they judge their own launch against a number that almost nobody hits.

How long does a launch keep working?

About two days. Roughly 92 out of every 100 stars from a launch arrive inside the first 48 hours. If you posted on Monday, checking again on Friday will not teach you anything new.

My post got nothing. Is that the end of it?

No. Hacker News keeps a second chance queue that moderators fill by hand, and you can write to hn@ycombinator.com and ask them to look at a post that got missed. They said they like these requests, and they like them even better when you are putting forward somebody else's work rather than your own.

One question before you call it a win

![Frequently asked questions