Information Extraction — AI Glossary

Reference details

Related terms

• Named Entity Recognition (NER)
• Intelligent Document Processing
• Text Classification

Example: The field that is not in the document

Extract vendor, invoice number, total and purchase-order number into a JSON schema. Three are printed on the page; this particular invoice has no PO number. A model given a required string field will very often produce a plausible one rather than fail. Make the field nullable, instruct explicit abstention, and — the step teams skip — include documents with genuinely missing fields in the evaluation set, scoring “correctly returned null” as its own outcome. Otherwise accuracy is measured only on fields that were present, which is the easy half of the task.

Commonly confused with

Information extraction is the task; structured output is the mechanism. Schema enforcement guarantees the shape of the JSON, never the truth of its contents — a syntactically perfect object can be entirely invented. Grammar-constrained decoding removes parse errors from your problem list and does not touch accuracy.

When to use it

Reach for it when:

• Heterogeneous documents, where a rules-based parser breaks on every new layout
• Few-shot settings with no labelled training data for a dedicated model
• Where downstream systems need typed fields rather than prose

Reach for something else when:

• One stable, high-volume format — a deterministic parser is cheaper and auditable
• Without an abstention path, where a missing field becomes a fabricated one
• Where extracted values feed irreversible actions with no verification step

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Information Extraction — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path