Different genres have different requirements when it comes to composition, so generalisations are guaranteed to miss the mark occasionally. If there was a genre called "scene capture" it would encompass the most basic aspects of what most people want when it comes to composition. Some of the possible exceptions are still life, product photography and the like.

A popular genre which has common, basic composition requirements is landscape photography, and it's common to see landscape scenes being divided into foreground, mid field and far field (or background). In landscape photography, composition is all about finding the right viewpoint (where to stand) and the right line of sight (what direction to look in). Within the camera's field of view there will be a main subject or subjects of interest or a main area of interest, and the photographer's job is to make it obvious what the main subjects or areas of interest are, regardless of whether they are in the foreground, mid field or far field. Compositionally that can be achieved by means of placement within the frame or alternatively it's possible to use various devices like leading lines or clear channels to lead the viewer's eye towards those specific subjects or areas of interest. Other factors like lighting, contrast, sharpness and colour can also be used to draw the viewer's attention but the subject here is composition.

When I have my composition hat on I look for ways to draw the viewer's attention into the shot, and very often that goes hand in hand with ways to amplify the depth aspect of the scene. Photos are a 2D rendering of a 3D reality and anything you can do to enhance the perception of depth within the photo will be a major step towards creating the idea that the viewer is perceiving a 3D reality as opposed to a 2D rendering of it. You want the viewer to forget that he/she is not actually seeing reality but a facsimile of it. I've noticed that 3D depth cues are a pronounced feature of AI-generated images. It's obviously a major aspect of the algorithms that are used to create images from scratch. Perhaps it's our turn to learn from AI instead of it learning from us.

I never "compose". Therefor no rules or advices are ever involved.

I go for "looks right". Im not gonna go for "looks best". "Best" is too judgemental and subjective.. "Right" is absolute. Its binary. Either it is or it isnt "right". I can produce many variants that look right.

Judging "best" does not occur while still shooting. Thaz later, at printing or sharing. And judgement of "best" is only valid for a particular moment.