honest one for the founders here (+ updates on the open-source YC data platform) 👇

All the above, plus an eventual patent to be comprehensive… https://patents.google.com or https://ppubs.uspto.gov/pubwebapp/

For me, it's usually a mix of Google, Product Hunt, and some other random places.

But honestly, even if I find something similar, I don't see it as a reason to stop. Sometimes it's more about building a better one.

honestly my move is usually just typing the idea into PH search plus a quick GPT sanity check, both feel incomplete because keyword search misses anything phrased differently and GPT just hallucinates competitors half the time. semantic search over an actual startup dataset sounds like the right fix rather than a workaround. one question - does ExploreYC only index companies that made it into a YC batch, or does it also catch stuff that applied and got rejected/pivoted? that second group feels like where the real "someone already tried this and it didn't work" signal would be

Most people I know do a mix
quick Google for exact matches, then browse a few category leaders on Crunchbase or Product Hunt to see what's recent.

Curious if you're filtering by funding stage or geography, since "done before" often means "done before in my market" rather than globally.

I have been thinking about this recently because I am building an Influencer discovery tool using the logic and python setup we used in-house for our agency. Before this, the solutions for influencer discovery were just Modash, fevikon and some of the top players. None of them were affordable nor were they solving the problem with the logic we were using. They were just database depended. Now when I am working on the tool's launch, I come across many influencer discovery tools people are building and some are even crazier than I think. However, as @hosseinyazdi said here in the comments, it doesn't stop me from building my tool. In fact, I use the competitors to study what market really need.

@Konstantin Borimechkov that's exactly the dataset I was hoping you'd say yes to. HN and a16z data alongside YC gives you a much broader "has this basically been tried" signal than YC alone, since plenty of ideas get built and die outside the YC pipeline entirely. does the vector search weight recency at all, or does a 2015 attempt at an idea rank the same as a 2025 one? for fast-moving categories like AI tooling a 10 year old failed attempt might not mean much anymore, the market and the tech underneath it are completely different now.

My actual move before building: exact-match Google + PH/HN for the obvious names, then I flip the question. "Has someone done this" almost always resolves to yes — there's a graveyard for every idea. The signal I'm actually hunting is why the last few attempts stalled — distribution, timing, or the thing being genuinely hard? That's what tells me whether to build.

+1 to @omri_ben_shoham1 's recency point, and I'd push it further. A 2015 attempt at an AI-tooling idea barely counts as prior art, because what killed it was usually a constraint that's since moved — inference cost, context windows, no agent runtime. The useful lookup isn't "does a similar company exist," it's "someone tried this in 2015, it died because of X — is X still true in 2026?"

Which is why semantic search over a real dataset is the right base layer, @konstantimb — existence search that survives different phrasing. The thing I'd kill for on top is outcome: dead / pivoted / acquired / still-going, plus the stated reason. Existence tells me I'm not first; outcome tells me whether first even mattered.

(And agreed with Hossein and Lakshya above — a match was never a stop sign for me either; it's just cheaper to learn from someone else's gravestone than to dig my own.)

I will say all of the above, but here's the thing we learned building Kednus (www.kednus.com): "has it been done" is the wrong question. It almost certainly has. The better question is "why did the existing versions not win my specific customers/ICPs?