I read 73,000 product launches with AI. Most of them never died. They just went invisible.

Thousands of products launch every week. Product Hunt, Uneed, Peerlist, BetaList, plus a dozen smaller sites. The founder posts, friends upvote, there's a nice little buzz for a day. Then a new list starts the next morning and the product is gone from everyone's screen.

I know this one from experience. Before Tablif I launched a different product. It got 29 upvotes on Product Hunt and landed in the top 3%. On Fazier it was #1 product of the month. On paper it was a good launch. Over the next six months it got 250 downloads in total, and today it has 110 active users. Launch day felt great. The day after was silence.

That's what made me curious about what really happens to products once launch day is over, so I started collecting them. That was at the end of August. Today there are 73,384 products in my database, and a few things surprised me.

---

Collecting launches from 12 sites

The first part was boring and took a lot longer than I expected. I wrote a small collector for every launch site: Product Hunt, PeerPush, TinyLaunch, Uneed, Peerlist, Fazier, BetaList, Tiny Startups, Microlaunch, OpenHunts, DevHunt and Smol Launch. Twelve in total.

Each one works differently. Some have an API. Some hide clean JSON inside the page. Some render everything in the browser and never show the product's own website anywhere, which makes them almost useless to me, because the website is how I tell that two launches are the same product.

Product Hunt is still the giant. 66,309 of my launch records come from there. Number two is PeerPush with 6,310. The rest are smaller, but that's usually where you find the most honest indie products. A lot of founders launch the same thing on four or five sites in the same week, so matching them by domain was one of the first real problems I had to solve.

---

Reading the products, not the headlines

A name and a tagline tell you almost nothing. "AI-powered workflow tool for teams" could be a thousand different things. So the second step was to actually read every product's website.

So far I've crawled 530,774 pages from 13,490 product sites. A language model reads them and writes a short, plain page for each product: who it's for, what it does, and what it doesn't do. That last part turned out to be the most useful. Founders almost never write it, but it's exactly what a buyer wants to know.

Behind those pages is a fixed list of 80 questions I ask about every single product. Pricing, who it's for, which platforms it runs on, whether there's a free plan, whether you can reach the founder on X, that kind of thing. Across all products that adds up to a little over 5 million answers. Asking the same 80 questions everywhere is what makes the products comparable. Without it you'd just have 73,000 different marketing pages.

This is also where I learned what "we'll just use an LLM" really costs at scale. Since late September I've made 372,464 model calls. About one in five of them, 75,531 calls, came back with a rate limit error. Most of those weren't even my quota running out. The provider was simply saying "no capacity right now, try again later".

For a while my own retry code made things worse. Every failed call got retried straight away on the next API key, which burned more quota, which caused more failures. A classic storm. What fixed it wasn't clever code. It was slowing down. I split the work by priority, gave the background jobs their own budget and let them wait their turn. The pipeline got slower and a lot more reliable at the same time. If you're building on top of LLM APIs, plan for 429 errors from day one. They aren't an edge case. They're a regular Tuesday.

---

The surprise: products don't die

I was expecting a graveyard. Everyone says most side projects are dead within a few months. So I wrote a small checker that visits every product's site and asks one simple question: is it still up?

Out of 73,384 products, 427 are gone. For 808 I can't tell, because the site blocks bots or returns something weird. Everything else, more than 72,000 products, is still online.

That changed how I see the whole problem. These products aren't failing. Their founders are still paying for the domain, still fixing bugs, still answering emails. They just lost the one moment when people were actually looking. For most indie makers, the launch day spike is the entire marketing plan. After that, the product keeps living without anyone seeing it.

---

So I built a town

This is the part people find a bit strange, and that's fine.

Every product I collect gets an office in a small 3D town called Tablif Town. There are 29,808 offices right now. When a founder claims their office, they get a product page written from their own site, they can edit every line of it, and they play a short weekly office game. Every week they play, Tablif writes new pages about the product: answers to real questions, problem pages, guides. Each page is one more way for someone to find that product long after the launch list has moved on.

I know a town can look like a gimmick on top of what's basically a directory with good pages. But a list is something you scroll past and forget. A place is somewhere you come back to. Some founders have shared screenshots of their office on X. Nobody ever shares a screenshot of a directory listing.

It's still small. So far 170 products have been claimed by their founders. But these are real makers who come back every week, and that's the number I care about.

---

Turning the data into videos

The newest experiment started this week. Founders who play the office game get their own short video on YouTube, in a series called Guess the Startup. Three clues, all taken from the product's own page, and viewers try to guess which product it is before the answer shows up.

The channel is four days old. 21 shorts, 975 views, 2 subscribers. One video about a small smartwatch battery app got 701 views on its own, which tells me the format works when the product is interesting enough. It's a very slow start, but every video comes straight out of data the pipeline already has, so keeping it going costs almost nothing.

---

The real goal: visibility that doesn't stop

The town, the game, the videos. From the outside they might look like separate things, but they all serve one goal: a product's visibility shouldn't end on launch day. It should keep going.

The game is built around that idea. Nothing in it is made up. It runs on what founders actually do with their products and what their sites actually say. As a founder plays, that real information turns into new content and goes out through our own channels: the product pages, answers and guides on our website, the Guess the Startup episodes on YouTube, and the new projects we're working on next.

We want these products to show up in two more places. The first is search engines, so when someone looks for a fix to their problem, the right product's page is there. The second is AI assistants. People now ask ChatGPT and tools like it to "recommend something for this", and those assistants build their answers from pages they've read on the web. The more accurate, up to date and founder-approved pages a product has, the better its chances of being part of those answers.

So the game is really just the excuse. What we're actually doing is turning a few minutes of a founder's week into lasting, accurate content about their product on the web.

---

What I learned

A few things I'd tell anyone building something similar:

Collecting is easy. Matching is hard. The same product shows up under five names on five sites. The domain is the only thing you can trust, and even that breaks with app store links and GitHub repos.

Ask the model what a product doesn't do. It's the most honest part of any product page, and founders hardly ever write it themselves.

Respect the rate limits. A slow pipeline that finishes beats a fast one that spends all day fighting the API.

The problem isn't launches. It's attention. Most products are still alive. They just need a second, a third and a tenth moment where someone notices them.

I'm building this on my own and in public. If you launched something in the last few months, there's a good chance it already has an office waiting in the town. You can find it at tablif.com.