Metadata is data that describes other data: information about the content, structure, origin, format, quality and rules of use of a data asset. Among data fundamentals, metadata is what makes raw data findable, understandable and manageable: a column name, a file creation date, a photo’s camera settings and a dataset’s owner are all metadata.
The distinction between data and metadata is a matter of role, not of content. The timestamp of a sensor reading is part of the data for a time-series analysis, but the date a dataset was last refreshed is metadata about that dataset. What counts as metadata depends on what it is describing.
The main types of metadata
Metadata is commonly divided by the question it answers. Library and archive practice uses descriptive, structural and administrative metadata, while data management adds the business, technical and operational view used in data catalogs.
| Type | Question it answers | Examples |
|---|---|---|
| Descriptive metadata | What is this and how can it be found? | Title, subject, keywords, abstract |
| Structural metadata | How is it organized? | Schema, column types, chapter order, file relationships |
| Administrative metadata | How is it managed? | Owner, creation date, access rights, retention period |
| Technical metadata | How is it stored and processed? | File format, encoding, size, compression |
| Business metadata | What does it mean for the organization? | Definitions, glossary terms, quality rules |
| Operational metadata | What happened to it? | Load times, row counts, job status, lineage |
These categories overlap. A field such as last modified date is administrative for an archivist and operational for a data engineer. The categories are useful for planning which metadata to capture, not as strict boundaries.
Where metadata lives
Metadata can be embedded in the asset itself or stored separately in a catalog, registry or database. Embedded metadata travels with the file, which is convenient but risky when files are shared, while external metadata can be searched and governed centrally.
- File headers: EXIF in photos records camera model, exposure, time and sometimes GPS position; ID3 tags store song titles in audio files.
- Document properties: office files keep author, revision count and editing time.
- Database system catalogs: the database stores table definitions, column types, constraints and statistics as metadata.
- Web pages: title tags, meta descriptions and structured data markup describe content for search engines.
- Data catalogs: central inventories of datasets with owners, definitions, quality scores and lineage.
Structured data is defined largely by its metadata: a table’s schema tells every reader what each column contains. Semi-structured data such as JSON carries its metadata partly inside the records as field names, and unstructured data such as images relies on extracted or attached metadata to be searchable at all.
Metadata standards
Shared standards make metadata interoperable between systems and organizations. Without an agreed vocabulary, one system’s “creator” is another system’s “author”, and automated exchange breaks down.
| Standard | Domain | What it defines |
|---|---|---|
| Dublin Core | General resources and libraries | 15 core elements such as title, creator and date |
| EXIF | Digital photography | Camera settings, timestamps, location |
| ISO 19115 | Geographic information | Extent, reference system, lineage of geodata |
| DCAT | Open data catalogs | Datasets, distributions and catalogs on the web |
| Schema.org | Web content | Structured descriptions for search engines |
| ISO/IEC 11179 | Metadata registries | How data elements are named and defined |
Open data portals depend especially on such standards. A dataset published without a license, update frequency and field definitions may be technically available but practically unusable, because users cannot judge whether it fits their purpose.
Why metadata matters for data management
Metadata is the foundation of data governance, data quality and data lineage. Analysts use it to find the right dataset, engineers use it to trace how a number was calculated, and compliance teams use it to locate personal data and apply retention rules.
| Task | Metadata needed | Benefit |
|---|---|---|
| Data discovery | Descriptions, tags, owners | Analysts find trusted datasets quickly |
| Impact analysis | Lineage between sources and reports | Safe changes to upstream systems |
| Quality monitoring | Expected ranges, freshness, row counts | Early detection of broken pipelines |
| Privacy compliance | Sensitivity classification, retention | Correct handling of personal data |
| Long-term preservation | Format, checksums, provenance | Files remain readable and trustworthy |
Modern platforms distinguish passive metadata, documented once and rarely updated, from active metadata, collected continuously from query logs and pipelines and used to trigger alerts or recommendations. Active metadata stays current, whereas manually maintained documentation tends to drift out of date.
Risks and common mistakes
Metadata can reveal more than the data itself. Photo location tags can expose a home address, document properties can reveal internal names and revision history, and communication metadata, meaning who contacted whom and when, can map relationships even without any message content.
- Leaking embedded metadata: strip EXIF and document properties before publishing files externally.
- Treating metadata as optional: datasets without owners and definitions quickly become data silos nobody trusts.
- Inconsistent vocabularies: free-text tags with spelling variants defeat search; use controlled terms.
- Stale catalogs: metadata that is not updated with the data misleads users worse than no metadata.
The most durable remedy is to capture metadata automatically at the point where data is created or loaded, and to assign a named owner who reviews business definitions. Metadata collected as a side effect of normal work stays accurate far longer than documentation written in a separate cleanup project.
A plain-language summary
Metadata is the label on the box, not what is inside the box. A library card that lists a book’s title, author, year and shelf number is metadata. It does not contain the story, but without it nobody could find the book among millions of others.
Computers use the same idea everywhere. A photo knows when it was taken, a spreadsheet knows who saved it last, and a database knows what each column means. Good metadata turns piles of data into something people can actually use.
An example from practice
A city administration publishes about 400 datasets on its open data portal, from bus stops to air quality readings. Citizens and developers complain that they cannot tell which files are current or what columns such as pm_val mean.
The data team introduces a mandatory metadata profile based on DCAT: every dataset must have a description, owner department, license, update frequency, spatial coverage and a definition for each column. A nightly job adds operational metadata such as the last successful refresh and the row count.
Within months, the portal search returns relevant results, outdated datasets are visible at a glance and developers build apps on the air quality data. The team also strips GPS metadata from published photos of public works to avoid exposing employees’ locations.