@shmerdloff That was EXACTLY the point of the Stanford study. Many years ago I worked on and managed two help desks, where the answers were quite clear cut, but could depend on how a specific question was asked, and the details provided by the caller.
We regularly had "shoppers" who would call repeatedly, tweaking their questions each time, until they got the exact answer they wanted. Then they would ask the Help Desk person for their name and employee ID, so they could get the action they wanted from the organization.
Calls were not recorded back then, so we finally required each helper to end their call with a short script saying "The answers provided are based on the information as presented by the caller, and may not apply if details are different or have been omitted."
Automating help systems, recording calls, and caller ID, have made that more difficult. But now we have AI, ChatBots, etc...UGH!