Skip to main content

AI glossary Data & Retrieval

What is information extraction?

← All glossary terms
In short

Turning unstructured text, such as emails, contracts or invoices, into structured data: entities, fields, relations and events.

Explained

Information extraction is the task of pulling structured data out of unstructured content. From a contract it can take the parties, dates and amounts; from an email, the customer, the product and the request; from a report, the events and how they relate. Traditionally this needed rules or models trained for each field and document type. Large language models can now extract many fields from varied documents with instructions and a few examples, often returning the result directly in a fixed format such as JSON, which makes extraction one of the most practical uses of AI in document-heavy processes.

What gets extracted

  • Entities: people, organisations, places, products, account or policy numbers.
  • Fields: dates, amounts, addresses, clauses, line items.
  • Relations: who owes what to whom, which product a complaint refers to.
  • Events: an incident, a change of address, a cancellation, with when and where it happened.
  • Classifications attached to the record, such as document type or request type.

How it works with language models

The document is converted to text (with OCR for scans), split if it is long and sent to a model together with instructions and the target schema: the fields to fill and their types. The model returns the values, ideally with a pointer to where each one appears in the source.

The output is then validated, for example checking that dates are valid, totals add up and identifiers exist in the system of record, before it is written to the destination.

Keeping it accurate

  • Define the schema precisely, including what to return when a field is missing.
  • Ask for the source passage of each value so people can check it quickly.
  • Validate with rules and against existing systems, not only with the model.
  • Send low-confidence or failed validations to a person, and use their corrections to improve the prompts.
  • Measure accuracy per field on a labelled sample before and after going live.

Frequently asked questions

What is an example of information extraction?

Reading supplier invoices in different layouts and returning, for each one, the supplier, invoice number, date, tax, total and line items in a fixed format that the finance system can load.

What is the difference between information extraction and OCR?

OCR turns an image of text into text. Information extraction works on that text to find the specific pieces of data you need and put them into fields. For scanned documents, OCR usually comes first and extraction second.

Can LLMs do information extraction?

Yes, and it is one of their most practical uses. They handle varied layouts and wording with instructions instead of per-template rules. They still make mistakes, so validation and human review of uncertain cases remain necessary.

What is named entity recognition?

Named entity recognition (NER) is a part of information extraction that finds and labels names in text, such as people, organisations, places or dates. Full extraction goes further, filling specific fields and capturing relations between them.

Decision guides that use this term

How to put it to work

Where it shows up at Thinkia

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert