When they talk about efficacy they're always citing a study, which means you can only apply it so far out of a study.
The endpoints of the vaccine studies which were done late last year were symptomatic covid, and severe covid. I put the Pfizer and Moderna studies up there. So when people say 95%, they are referencing those studies or studies like those.
If you ask about being infected but not sick, I guess you'd define that as being PCR positive but no symptoms. I don't think any of the vaccines tested for that... because it doesn't really matter, in a practical sense, and it's much more difficult to do. You'd have to periodically test everyone, regardless of symptoms.
But they did test for people who had
any one of those symptoms, plus being PCR positive. In those tests (which were pre-delta) you were roughly 19 times more likely to get sick - any of the symptoms plus PCR positive - without the vaccine than with.
To re-define efficacy, you'd have to do another study, like say Pfizer efficacy vs delta, and then that efficacy number would be again compared to something.
This study did what's called a test-negative design* to do just that, and found that against the alpha variant it was 93% effective and against delta 88% compared to not taking the vaccine. Efficacy is always relative to something, so should always be talking about some study, trial, whatever.
The media shorthand for 95% is just sort of lumping all these into a basket and hand-waving at them. It sounds science-ier than saying "really effective" or "big numbers".
*Test negative is a different style of studies than the phase 3 trials, because instead of starting with a test (vaccinated) and a control (placebo), you count everyone who presents with symptoms that match your case criteria (in this case having covid) then find out afterwards which patients are your test group (vaccinated) and which are your control (unvaccinated). Common for monitoring for vaccine efficacy.