What is AI extraction?
Extraction means pulling specific fields out of unstructured text or a file: a name, a date, an address, a dollar amount. The rest of the paragraph is not the point. Missing fields should stay missing.
In plain English
A lead email is a story. Your form needs city, date, and flights of stairs. Extraction is the step that fills those boxes or marks them unknown. It is not a summary and not a label, though you might classify first and extract second. List the fields and the allowed emptiness. A missing phone number should stay missing. The model will otherwise borrow one from a signature or from a different line in the thread. Check ten real messages, including a forwarded one and a one-line one. If a field is wrong more than rarely, do not pour it into the sheet unreviewed.
How an owner would use it
List the fields and the word UNKNOWN. Map them into the sheet or the CRM. Have a person look at the first fifty before you trust the map. If the source is a photo of a form, you may need OCR before extraction.
Common mistakes
Extraction that fills gaps to look complete will schedule the wrong building. Do not extract a price the customer merely hoped for and drop it into the quote as yours. Do not run extraction on a pile of unrelated PDFs and call it bookkeeping. If you need a strict shape for software to read, pair it with structured output.
A worked example
A moving company
Stairs stay unknown until someone asks
This is an illustration, not a customer result. The model pulls pickup city, date, and flights of stairs from the web form's comment box. The comment never mentions stairs. The field stays UNKNOWN, and the coordinator asks. The model does not assume a ground-floor walk.
Related terms
Prompts and guides
Questions owners ask
What is extraction?
Extraction copies specific facts out of a messy document or message into labeled fields.
How do I keep extraction from guessing?
Name the fields and require UNKNOWN when the source does not say. Reject any row that invented a required field, and fix the prompt before you automate the write.
How is extraction different from classification?
Classification picks a label for the whole item. Extraction pulls out the pieces. A move request can be classified as residential and still need the date extracted.
Product guides are on AI apps for business.