I built the wrong half of my product first
A few months back I promised some friends I'd have something for them by morning. We'd been going in circles about how everyone should start their own thing, and the sticking point was always the same: no way to tell whether an idea had any room in it. I'm a frontend dev with a day job, so instead of thinking it through I started writing code. That's my usual mistake.
The plan was a validation tool. Point it at a company's site, get back the boring research a person would otherwise spend two hours on: what the business actually does, who else does it in a given market, whether anyone writes about it, whether anyone searches for it, whether the company is even still alive. By morning it worked, roughly, and I had a few thousand sites sitting in a database.
The output was useless. Not broken, just useless. "Task manager for small teams" doesn't answer anything you actually need to know before committing six months of your life to something. So I spent another week going deeper: real competitor lookups, press mentions, legal and revenue data, SEO metrics. I was pretty pleased with it. I showed it to a handful of people, got "interesting" from all of them and a second visit from none. Meanwhile every deep report cost real money in data and model calls, so running the whole database was never going to happen. It wasn't the code that killed it, it was the bill.
Somewhere in the middle of all that I added a thing on the side, in one evening, mostly out of curiosity. It started asking AI assistants the questions a buyer would ask about a category, and writing down what came back: did the brand get mentioned, how was it described, who got recommended instead, what sources did the model lean on. I'd noticed I'd stopped googling for tools myself and just asked an assistant which one to use, and it bothered me that there was no way to see any of that from the other side.
That evening's work is the only part anyone has wanted to talk about since.
What surprised me most is that assistants don't answer the same question twice the same way. Ask twice, get different brands. So "am I mentioned" isn't a yes or no, it's a rate across a lot of runs, which is both annoying to build and annoying to explain. Prompts turned out to matter more than anything I wrote in code: feed a bare keyword and the model answers like a search engine, all generic, no brands in it at all. Mentions only show up when the question has someone specific behind it, with a constraint. And cost turned out to be an architecture problem rather than a billing one. At ten domains you don't notice it. At a few thousand it decides what you cache, what runs daily, and what you simply never compute.
The part I still find uncomfortable is that the version collecting far more data is the one nobody wanted, and the version collecting far less but answering one specific question is the one people ask me about.
So that's Ideata now. It tracks what AI assistants say about your brand, who gets recommended instead of you, and where those answers come from.
The free plan is genuinely usable, not a teaser: no card, one brand, prompts tracked across the assistants, and a research run on your own site. Today I also opened the blog writer on it, which was previously behind a paid plan. Everyone gets 100 credits, an article costs roughly fifteen, so that's a handful of real posts you can use to build out a content plan and see whether assistants start citing you. When the credits run out they run out, no card involved.
Right now I need people telling me where this is wrong more than I need the money. If you've got a site, run it and tell me if the numbers look like nonsense. That would be the most useful thing anyone could do for me this month.
