TL;DR - The answers by John and Mark tell you what to do. This answer addresses why those are good answers.
I don't use Lightroom, so this may not apply completely. The answers by @john and @Mark are very good. I just want to highlight some fundamental issues that they assume, but don't explicitly state here.
1) To be sane, you need to store each photo in just one primary place (not counting backups and, possibly, symbolic links/shortcuts). This is fundamentally a database issue. If you store data in more than one place, then you run into issues if the data gets deleted or modified in one place, but not in the other. Then you have problems knowing what the "current" or "original" data is. (And it also wastes storage.)
2) Most photos cannot be adequately represented by a single attribute and the attribute(s) of interest (and their hierarchy) when searching for a photo later frequently cannot be known and fixed at the time the photo is "catalogued".
3) Because 1) requires you to store the photo in just one place (i.e. in one place in a hierarchy of folders), that place can only match at most one potential type of query.
So, the general solution is to store the original photos in an arbitrary set of folders - with some consideration given to general file management issues.
(I store mine in folders by camera, then in subfolders named as the date of last photo in each group as YYYYMMDD (so the folders naturally sort by date when displayed sorted by name). I try to keep the number of pictures in one folder to around 700 or less because that's an easy number for me and my system to process. But lots of other systems will work.)
That takes care of storing the photos, but what about finding them?
That's where the metadata comes in. Some of this may be stored directly within the photos and some may be stored in your photo organizing software (e.g. Lightroom).
Some metadata, like GPS, camera data, date and time are automatically added to pictures by most digital cameras. Beyond this, you can define your own tags which can either be flat or hierarchical. E.G.just Person, or Person-Friends-Male-Dave (where any photo assigned the "Dave" tag is also automatically assigned the tags Person, Friends, and Male. (KPhotoAlbum works this way on Linux. I don't know how Lightroom handles it.)
The advantage of this is that a single photo can have an arbitrary number of tags (limited only by storage and software implementation constraints).
And software can be used (sometimes, automatically behind the scenes) to build secondary indexes based on these tags so you can quickly access any category of photos based on your needs at the moment. This makes the photos "appear" as if they were stored in folders which are exactly suited to the the current query - which may be quite different from the very next query you may need to make.