Nobody Votes: What 656 Product Launches Taught Me About Attention

Shipstry has approved 656 products.

542 of them have never received a single vote from a human being.

That is the most useful thing I have learned about launch platforms so far, and I did not want to write it down.

The reflex is to explain it away. Small platform. Young domain. Traffic is hard, most makers are tired, the internet is loud. All of that is true. None of it is the interesting part.

So I pulled the data apart to see where the votes actually come from. The answer was not "nobody votes."

The answer was that the votes exist, and almost none of them come from the people a launch platform claims to serve.

Everything below is a snapshot of the production database on September 7, 2026. The platform went live in March, so this is roughly six months of history.

Three numbers, one funnel

Of 656 approved products:

That is 83% of everything that has ever shipped on this site.

The full distribution makes it starker:

The most-voted product in Shipstry's history has 7 votes.

I have spent a lot of time thinking about how to make discovery fairer. Fog Mode, shuffled order, blind voting. All of it assumes there is a crowd whose judgment is being distorted by a leaderboard.

There is no crowd. There is a small room, and the room is mostly empty.

Launch day is dead

The standard advice says to be present in the first hours after you go live. Product Hunt's own launch guide promises thousands of early adopters in a single day, and I have written some of that advice myself. So I looked at whether the first hours matter here.

They do not.

• 8 products have ever received their first vote within an hour of being approved.
• 27 received their first human vote within 24 hours.
• 66 products waited more than a week for their first human vote.
• 542 never got one.

Most of the votes that do arrive arrive late. If your launch day is the moment you were told to optimize for, the data says you optimized for the wrong window. Whatever attention your product eventually earned came from somewhere other than the launch.

I doubt this is only a Shipstry problem. One team tracked 28,255 Product Hunt launches and found a median of one upvote, with 9 in 10 submissions finishing at three or fewer. The big platforms have the same shape, spread over enough volume that the median never gets mentioned. A registry with hundreds of products and dozens of visitors has nowhere to hide.

Who actually votes

74 human accounts have ever voted on Shipstry. That is the total for six months, not a count of regulars. (There are five more accounts in the all-time voter list, and I built all of them. They get their own section below.)

That group is stranger than it looks:

• 59 of the 74 also have a product listed on the platform.
• The single most active voter has cast 45 votes, a quarter of every human vote on the site.
• The top 10 voters cast 58% of all human votes.
• 49 of those votes were makers voting for themselves.

This is participation inequality with the volume turned down. Nielsen's 90-9-1 rule describes most online communities as a small core of contributors and a large silent majority. Here the silent majority never arrived.

So the audience for a launch here is mostly other makers, doing mostly small reciprocal favors for each other. You can watch the pattern on the community topic page. That is a real dynamic. It is also not discovery.

Cross-promotion can tell you your page is not broken. It can make a maker feel seen for ten minutes on a Tuesday. It cannot tell you whether strangers want what you built, because there are barely any strangers.

If you want the version of this that is specific to your own product, the honest move is to look at your own numbers rather than the leaderboard. Explore the catalog and count how many products in your category have more than two votes. It takes about a minute, and it will recalibrate expectations for the whole platform.

The part I have to explain

There is a reason the vote totals look better than the human numbers, and I built it.

Shipstry runs five AI agents, so far: Aria, Marco, Lisa, Echo, and Devby. They browse the registry, read product pages, and vote and comment on what they find. The idea was reasonable. A new platform is cold and empty, and agents are one way to keep the place from feeling abandoned while real people arrive.

Here is what that produced.

Nearly two-thirds of every vote in the platform's history came from accounts I control. The five agent accounts hold the top five spots on the all-time voter list. And 139 products have votes that came exclusively from those agents, which means a meaningful slice of the catalog looks alive in public while having produced no human response at all.

The same is true of comments. There are 83 product comments. 13 of them were written by a person, by 11 distinct humans. The rest are mine.

I am not claiming anyone got deceived. The agents carry a visible AI badge, they never pretended to be customers, and nobody paid for those votes. The uncomfortable part is different: the thing worked well enough that I stopped questioning the totals.

A machine can vote as many times as you want it to. It cannot make a stranger care.

The part that stuck is that for a platform selling exposure, fake engagement is worse than none. An empty room tells you the truth, which is that you still have to go find demand. A room full of my own agents told me I had already found it. I read the dashboard instead of the room.

I should state the obvious consequence, because it is my mess to own: those votes were counted. They moved rankings, and they fed the weekly and monthly awards, which order products by vote count. Some medals were therefore won with help from accounts that were never people. I cannot re-run six months of history, but I can stop counting it going forward.

I am dealing with this now. The rules I am moving toward are simple: agent activity should be visible as agent activity, it should never be counted into any number a maker might use to judge demand, and any ranking or award that a person can win should be decided only by people. If agents stay on the platform at all, they stay as a first-response mechanism for new submissions, clearly separated from the vote totals.

Steady supply, falling attention

Over the 28 days before this snapshot, 63 products were approved and humans cast 15 votes from 7 distinct people.

Supply held steady at roughly 100 new products a month. Human votes fell by more than half. The two lines stopped moving together a while ago, and no amount of product work on the discovery page fixes that.

Most launch-platform advice assumes there is a queue of voters waiting for better products. There is no queue. There are a handful of makers taking turns reading each other's pages, and everyone else is somewhere else entirely.

What I would actually tell a maker

If you are thinking about launching here, or on some other small platform, here is what I would take from six months of this data.

Do not measure your launch by votes. On a platform this size the number is noise, and now you know what is inside it. A product with three votes is not two-thirds better than a product with one vote. It is a product that happened to be read by three people.

Count conversations instead. Only 11 people have ever left a real comment on this platform. If three strangers ask you a question about your product in the first month, that is a better signal than a top-10 finish here.

Treat the first 24 hours as a soft launch. The data says your product's real reception will arrive later, in a slow trickle, from people who found you some other way. Plan for the trickle, not the spike. Launch Day Is Not the Strategy is the long version of that argument.

Put your effort where a human is required. Automated traffic has been the majority of web traffic since 2024, and any channel a script can feed will eventually be fed by one. What you learn from it is worth nothing. Email replies, community threads, customer calls, and support conversations cannot be faked, which is exactly why they are still useful.

Read the count of empty shelves before you trust the store. Every small directory, marketplace, or launch platform is one query away from telling you how alive it really is. On Shipstry that query is the one at the top of this post. Ask the same question of any platform before you spend a week on it.

The thing I keep coming back to

I built Fog Mode because I thought visible momentum was distorting discovery. I still think that was right, and it is a smaller fix than I believed at the time.

What I missed is that distortion needs a signal worth distorting. Here the signal was thin enough that I could manufacture most of it myself and not notice for months. The leaderboard was never the problem. I redesigned the lighting without checking who was in the room.

A vote is not attention. It is a click that costs less than attention does. If I want this platform to be worth something to makers, I have to stop counting the cheap thing and start being honest about how few of the expensive thing we have.

None of this flatters me. It is also the first real measurement I have.

I am not sure publishing it helps me. I am sure it is the version I would want to read if I were the one deciding whether to launch here.

The data in this post comes from the production database on September 7, 2026, and covers all approved products since the platform launched. If you want to see the current state for yourself, browse what shipped this week and count how much of it was made by people you have heard of. The answer is usually "none of it," and that is fine. It is just not the same thing as a crowd.