I run a weekly experiment on AI recommendations. This week's category broke my assumption.

Part of building Bersyn is that every week I run the same experiment on a new buyer category: ask ChatGPT, Claude, Gemini and Perplexity the questions a real buyer types, then count who gets named and who gets recommended first. It is the product, but it is also free market research, so I share it.

Last week I did databases. The serverless newcomers (Neon, Upstash, Turso) had already become AI's default. My takeaway was optimistic: a challenger can absolutely become the recommendation.

This week I did transactional email APIs, and it humbled that takeaway.

Three buyer questions x four models = 12 conversations. The incumbents held:

Amazon SES, Mailgun and Postmark were each named in all 12.
An established player was the first answer 8 of 12 times.

One challenger broke through: Resend, named 8 of 12, first twice. But on ChatGPT it was named 0 of 3 times. The model with the most users does not say it in this category. So even the winner only won on three of four surfaces.

And the others I expected to see were nearly gone:
Mailtrap showed up only on Perplexity, never on the other three.

The honest lesson for me as a founder: it is not about which week you happened to pick. Some categories the newcomers have already won the inputs, some the incumbents still hold them, and a single model can have a completely different roster than the other three. AI is not evaluating products, it is repeating its inputs. Resend made it into those inputs on Claude and Gemini; on ChatGPT it did not.

That blind spot is exactly the thing I am building Bersyn to make visible. If you want to see who AI recommends in your category, the board for this one is here: https://www.bersyn.com/recommends/transactional-email-apis, and you can run a free scan on your own category at https://www.bersyn.com.

Curious from other founders here: have you checked what AI says about your product? Were you in the answer, or was it a competitor?