What can be seen is what people search for and what they ask on forums and what they tell surveys. Ask the models those questions in the words people use and ask them more than once and count what comes back outside the AI and you have a reading. A proxy prompt is a real question people type. A test prompt is one written to test a claim the real questions do not reach. Both are kept with the reading and both carry a version so that the next reading can be compared with this one. That is a method and not a guess and it turns the anxiety about AI search into a number with a range beside it.
My details are on the contact page if you have any questions or comments. Please get in touch.