This seems nice, I will try it tomorrow. The N is a bit low though, to draw conclusions.

A comparison with docs+gemini but of quality of articles where an LLM uses the tool could be interesting as well, I may test it that way. Just a random thought tho!