The serverless database newcomers already won AI's recommendation. The specialists are invisible.

Still building Bersyn in public (it shows founders why AI recommends competitors instead of them), and the weekly teardown is becoming the actual distribution engine, so here is this week's.

I asked ChatGPT, Claude, Gemini and Perplexity which database to use, five ways each, 20 answers per company.

The surprise: the serverless newcomers are already the default. Neon recommended first 14/20, Upstash 11/20, Turso 9/20. So a challenger absolutely can become AI's go-to.

But ask about the specialized jobs and the incumbent wins every time: Pinecone/Milvus over Qdrant for vector, S3/R2/Backblaze over Tigris for object storage (named 0 times on three models), ClickHouse over Tinybird, Prisma over Drizzle.

The thing I keep relearning: the ones that won did not win on features, they won because the training data already named them as the answer. That is the whole product thesis, and it is also my own distribution problem in miniature, which is why I run these in public and go straight to the named companies with their own scan instead of waiting in a thread.

What I am testing now: tagging the invisible companies with their own teardown. Last round that got the named tools replying (and a couple of warm conversations). Trying it again here.

Anyone else found that going direct to the person with the pain beats broad posting? That is the bet I am running.