# Two Months, Zero Users. So I Asked an AI — The Culprit Was "One More Tweak" > Source: > Published: 2026-07-16 00:00:00+00:00 In the [first post](/blog/origin-story) on this blog, back in May, I made a promise: “I’ll write it up whether this goes well or badly.” Two months have passed. Bottom line: it went badly. More precisely, nothing happened at all. The service itself wasn’t idle. Twelve AI bots posted predictions every day, and nearly 1,400 of them had been scored. Including the baseline bots backtested to 2016, LDBD had passed 128,000 scored predictions. I rebuilt the scoring system, refreshed the design, even ran a security review. I wrote thirteen blog posts. And through all of it, the visitor graph stayed a flat line on the floor. The user count was still zero. ## So I asked an AI: what’s wrong with my service? I couldn’t figure it out alone, so one day I handed the whole question to Claude. What else is out there like this? Does LDBD have any real edge, any commercial angle? And why is nobody coming? The first thing it did was visit my site the way a stranger would. Then it reported back: “Nearly all of the top 30 on the leaderboard are the operator’s own bots. There is one human — a test account with just six predictions.” I look at that screen every day, but reading it described through someone else’s eyes made it land for the first time. I had spent two months decorating an empty room. ## The diagnosis: the product wasn’t the problem The research came back as a full report. The short version goes like this. Similar services already exist, the report said: real-money prediction markets, and play-money platforms where even the market leader peaked at around 2,000 daily users. Since last year, “AI arenas” that pit LLMs against each other with real trading money have been the hot thing, and one of the few clear money-makers in this space sells accuracy data on 8,000 Wall Street analysts. Within all that, LDBD’s differentiators — humans and AI competing on one board, open to any agent, statistically rigorous scoring — are real, the report said. The problem was that none of it was reaching anyone. Then came the part that stung: “Two to three months of zero visitors is not a product defect. It is the expected outcome. You never told anyone.” I wanted to argue, but the facts checked out. All thirteen posts lived only on my own domain. The condition I had set for myself — “once five to ten posts pile up, share it with communities” — had been met long ago, and instead of telling anyone I kept finding things to polish. After the design refresh, after the audit, after the share cards. The preparation was done; only the preparing kept growing. Looking back, there was nothing actually stopping me from launching. I kept postponing because one more tweak always seemed like it would make things a little better, and I wanted to show the thing at its best. But the day it would be perfect never came. New problems surfaced endlessly — while I polished one, two more caught my eye. The better move would have been to ship once the service basically worked and let real users’ feedback drive the fixing. I had the order backwards. The positioning critique hurt too. LDBD’s front page opens with “You called it? Prove it.” — but proof only means something with an audience watching. Asking early users to build a track record on a stage with zero spectators is a bad deal for them. People who want to show off already have places to show off. ## What I found while preparing to launch So I made a launch plan. Picked the venues, wrote the intro posts, prepared answers to likely questions. When I put the checklist together, there was nothing left on it. Everything had been ready for a month. There was one exception. While checking my account on an overseas community, I noticed the comments I’d left there weren’t earning any points at all. It turned out every comment had been caught by a spam filter — visible to me, invisible to everyone else. My two-month-old account wasn’t underperforming; it was quietly dead. If I had posted the launch announcement without noticing, it would have vanished without a sound. I emailed the moderators and pushed the overseas launch back. ## Then I launched Monday morning, I posted to [GeekNews](https://news.hada.io/topic?id=31386), a Korean dev community: “I had 12 AI bots predict stock direction for two months, all publicly scored.” Over two days, 926 people came — more than the previous two months combined. Twenty-one signed up, and two of them left predictions. For the first time, someone other than me was on a leaderboard that had held only my bots for two months. Two is a small number, but the distance from zero to two felt longer than from two to two hundred. The data held a surprise, too. The third most-visited page wasn’t the leaderboard or the predict screen — it was the [developer API docs](/bots). Predictable for a dev community, maybe, but it was the first sign that someone out there reads LDBD not as a site to browse, but as a platform they can plug their own agent into. The comments taught me things as well. The best one was the most skeptical: “no bot in the world has ever beaten humans at trading.” I got to answer that this skepticism is exactly where the service starts. If the bots can’t predict, the leaderboard shows it, publicly. And in fact, two months of data give me no grounds to claim the LLM bots meaningfully beat the “always up” baseline — not hiding that is the only honest thing this service can do. There was one more surprise waiting. On launch night, Vercel (my hosting provider) emailed to say I had used up the free tier, and that going over would pause the site automatically. I upgraded to the paid plan that day and spent the evening tracking down where the usage was going, then shipped a fix to cut the traffic costs. Having visitors means having these problems — the first good problem in two months. ## What I learned **Building is not telling people.** Thirteen posts and 128,000 scored predictions earned me zero visitors. Content doesn’t get discovered just because it piles up. Telling people is a separate job from making the thing, and I put it off for two months because making was the comfortable one. **You can only see your own product through someone else’s eyes.** I knew the leaderboard was all my own bots. I only understood what that meant when I read it described by a first-time visitor. **Having an edge and having users are different problems.** The diagnosis said the differentiators were real — and they didn’t bring a single visitor. Before the differentiator matters, someone has to see it — and understand what they’re looking at. **I said I’d write it up even if it went badly, so here it is.** Two months of zero wasn’t really the product’s fault. It was the finger that never hit publish. ## Next Once the account is restored, Hacker News is next. And the bots’ report cards will keep showing up in this log — this time, not just on my own domain.