I’ve spent hours trying to find a conceptual explanation…or any explanation or definition at all…of what clusters are. It isn’t defined in the glossary of the 200+ page documentation or anywhere else that I’ve been able to find. Obviously, I’m just starting with Picard.
Clustering is simply a way of grouping files for lookup on the MusicBrainz site, or to match to a release in the (right-hand) tagging pane of Picard. When you select files, either from the file explorer or drag’n’drop, they end up in the (middle) clustering pane. From there you can group them into “clusters” for lookup, matching to a release, or other processing such as adding tags or executing a script. There is also a plugin that allows you to submit a cluster as a release, by seeding the MusicBrainz add a release page with information from the files in the cluster.
I guess I should add something to the User Guide.
If there’s an album tag already present, clusters are usually based on that, because as @rdswift says, Picard will try to match a cluster with a release (i.e. album) in MusicBrainz. I don’t know all the exact logic, but without an album tag Picard will do its best at logical clustering based on folders, artists, and so on.