Metadata is data about data. It can be generated automatically or added manually depending on the context. Metadata can be used to e.g. describe the context of a specific data element, describe how a collection is organized, or used to identify licensing conditions for a specific element etc.
Metadata can be placed either separately from the data in external files/tables, or embedded in the actual file format. An example of external metadata is a ReadMe file that explains the structure of the files.
An example of embedded metadata is EXIF data in an JPG file containing data about about camera, date and time, location etc.
You should always consider using generic or subject specific standards for your metadata, and you may end up having to select several metadata formats depending on your research data type and where the metadata is used. E.g. you may have some metadata associated with the files in your data set, and create some other metadata for the data set itself if you deposit it in an external repository.
Research Data Alliance maintains a list of metadata standards.
Metadata are useful for both humans and machines, but the ability to understand metadata is challenged by using e.g. natural language. A lot of things are ambiguous, e.g. names of people or terms with different meanings across subjects. In order to avoid this, a lot of metadata standards allow for use of vocabularies or dictionaries, e.g. using ORCID identifiers for people or a specific vocabulary within your field of research.