What is metadata?
Metadata is information about data rather than the data itself. For a contract stored as a PDF, metadata includes the file name, author, last modified date, the matter or customer it belongs to, and whether it is marked confidential. For a database table, it includes the schema, the owner, the refresh frequency, the pipeline that produces it, and the reports that read it. Metadata is commonly divided into three kinds: technical (formats, schemas, locations), business (definitions, owners, classifications), and operational (run times, row counts, lineage), and a data catalog collects all three.
Why metadata matters
Metadata is what makes data findable, trustworthy, and governable. A city clerk’s office with decades of scanned council records can answer a public records request quickly only if each document carries the meeting date, agenda item, and retention class. The same holds for AI. A retrieval-augmented generation system that indexes documents without metadata returns passages it cannot attribute, date, or filter by permission. One that stores source, date, department, and classification with each chunk can cite its answer, prefer current documents, and enforce permission-aware retrieval. Data lineage, data classification, and data quality scores are all metadata, and together they are what governance acts on.
What to watch for
- Capture metadata automatically from systems wherever possible, because manually maintained metadata decays.
- Decide which business metadata is mandatory at creation, such as owner and classification, and enforce it.
- Treat metadata as sensitive in its own right, since file names and access logs can reveal what the files contain.