Essentially you can boil "tracking" to two main sources: when there's data collected without a legitimate purpose for doing so, and when data is collected to the point that could identify a user, but no explicit consent is given.
Take for example a Facebook comment section on a third party site. It'd be fine to click the comments and have a quick prompt for Facebook to interact – the comment is public, so it's known to all. But if the user never comments, Facebook has no right to be aware the user was ever there; that's tracking.
You could make the extended argument that overcorrelation of data for advertising is tracking in a sense, as this would cover intra-site tracking (e.g. a shopping site knowing you're pregnant before you know yourself.) This is a little more nebulous to define, as it's hard to define who it benefits If your phone launcher suggests an app, or Uber suggests a location, that's because it wants to save time. But if a shopping site suggests a product, that's advertising, and should be given explicit consent.