There was a paper recently: "LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings"

- The research showed models being able to reproduce purchasability via llms, which I thought was cool because i'm in the advertising space

- the question came up, can you predict which ads will also perform better?

- the answer is yes, you can also do the same with landing pages, logos, and other product shots.

- I've implemented a loop to use public data sets to test many theories on how models and now the J-space gives insight as to how to convert better.

- Let me know if you'd like the loop:)

LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings

https://arxiv.org/abs/2510.08338

here is the paper if you're curious.