# Create a table and define its schema

Use the App's Tables section to create a named table. Define the columns the App asks for, including each label and type. A useful table name describes the subject, while clear column definitions distinguish similar values such as invoice date, payment date, and due date.

## Choose types carefully

A text column preserves words and identifiers. A number column holds numeric values. A date column represents dates. The available types depend on the table creation experience. Do not use a numeric column for identifiers that require leading zeros, or a date column when the source states only an ambiguous partial date.

For document extraction, a field definition also determines the source meaning. “Amount” is less useful than “Total payable stated on the invoice,” especially when a document includes tax, subtotal, and multiple currencies.

## Proposals

An App can show an AI-proposed table with a name, columns, and rationale. Review it before accepting. Acceptance can create or refresh the target table, after which the App can run its configured population step. A schema proposal and completed population are separate outcomes; inspect population status and errors.

Dismissing a proposal does not mean every underlying source was deleted. It means you did not accept that proposal as the table definition.

## Document drafts

A captured-document table draft includes both schema and source-backed values. Review its columns and evidence before choosing **Create table**. Changing a draft's definition or type requires a new preview; simply renaming a meaning-bearing field is not sufficient to validate old values.

After creation, open the saved table and verify its actual columns and row count. An empty newly created table can be valid while awaiting rows. See [editing](/docs/tables/edit-rows-and-cells), [draft review](/docs/tables/review-and-accept), and [limits](/docs/tables/limits).