I agree, when you look at the source its amazing how often it is either not relevant or based on an insufficent data size/subject e.g. mice.
One problem I hit quite often is that the reference is hidden behind an expensive paywall. I could blame the search engine but it ain't necessarily so.
There are a very big US companies buying up uni libraries, research labs etc. research and "leveraging" their agreements so that new research is only published on their paywall site.
The Uni's students can access for free or a nominal fee (only whilst still students), the publisher allows search engines to see the index and a short summary or extract.
One of those companies has a very high share price and income stream, and apparently close links with the big alphabet companies developing AI.
Dan, law is fasinating but AI is already causing redundancies and law is easily coded to automate it.
I have a 1980's paper flowchart version for the very complicated JCT contract (Joint Construction Tribunal) designed to do just that. It works really well and saves a huge amount of work and time.
Relevant facts are mainly yes/no's and reduces the human decision points down to a list that takes very little time to decide on by the deciding power. Even without AI, such programmes can very quickly show a client whether they have any chance of success. Partners simply do not need many, or any, trainees, just a few unqualified cheap clerks and maybe one secretary to impress clients. Conveyancing firms already work this way.