# Overview: Document Data Extraction Into Structured Tables

> ImageToTable.ai turns images, screenshots, PDFs, and other documents into structured data: it extracts the values you name, computes ones the document does not print, locates visual elements such as a signature or logo, and merges multi-page results into one table — so you get the table without retyping anything by hand.

ImageToTable.ai turns documents into structured data, and structured data here means more than text values. You name the [columns](/docs/reference/glossary#column) you want and the AI fills them — reading a value off the page, computing one from figures the document shows, judging one the document does not state, or locating a visual element such as a signature or a logo and returning its position on the image. Results from a document that arrived as several pages are merged back into one record, and the finished table exports as rows and columns you can sort, sum, filter, and feed into another system — with no manual retyping at any point.

## What ImageToTable.ai is

Documents in, structured data out, with no manual retyping in between. Point ImageToTable.ai at images, screenshots, PDFs, Word documents, or plain text, and it returns the information they hold as rows and columns you can sort, sum, filter, join, and feed into another system. Along the way it does more than read text: it can compute values from figures the document shows, judge values the document does not state, and locate a visual element such as a signature, a logo, or a stamp on the page.

The work it replaces is manual data entry: a person reading values off receipts, invoices, bank statements, or forms and typing them, row by row, into a spreadsheet. With ImageToTable.ai that typing step is gone; the table comes back already filled.

The path from a file to a finished table is the same each time: the values are extracted, the columns you set up to compute or judge produce their results, results from a document that arrived in several pieces are merged back into one record, and the completed table is exported as structured data.

It works on a set of files rather than one at a time. The files you hand in together form a [batch](/docs/reference/glossary#batch), which is the unit you process, review, and export; how many rows a file contributes depends on what it contains.

## What it is not

Two things this product is often grouped with, and where the difference lies:

- **Not a text-only OCR tool.** OCR reads the characters on a page and hands back the text. It can tell you which words appear on a receipt, but not which of them is the vendor, the total, or the tax. ImageToTable.ai reads what a document *means* as well as what it says, so it can find the values you asked for even when no fixed label tells you where they sit — and it can locate a visual element such as a portrait or a signature without reading any text at all.
- **Not an enterprise document-workflow platform.** It is not a system for designing approval chains, routing documents between teams, or integrating with a records system before anything useful can happen. It starts from a file and the list of values you want back, and its output is a table.

The category it belongs to is structured data extraction: documents in, structured data out. It is also described as semantic or intent-based extraction, because what decides the result is what you ask for, not where the data happens to be placed on the page.

## You name the columns; the AI finds the values

That is the whole model. You list the [columns](/docs/reference/glossary#column) you want — the same names you would type as the headers of a spreadsheet — and the AI reads each document and returns the value that belongs under each name. There is no parsing template to design, no zones to draw around fields, and no model to train on your documents.

Because the column names carry the meaning, the same list works across documents whose layouts differ: a value does not have to sit in the same place on every page to be found. A list can be used for a single run, or saved as a reusable [template](/docs/reference/glossary#template); if you would rather not name columns at all, the AI can read the document and propose a table structure itself.

A column's value does not have to be text that is printed on the page. What a column returns is set by what its instruction asks for:

- **Read a printed value** — the default, taking the value as it appears.
- **Compute a value** — calculate one from figures the document shows, such as a line total or a difference between two amounts.
- **Judge or infer a value** — decide one the document does not state, such as the category a receipt belongs to.
- **Locate a visual element** — instead of reading text, find a portrait, a signature, a logo, or a stamp on the page and return where it is; the app shows a cropped preview of that region, and the located element is what lands in the table.

Computing, judging, and locating are what set the product apart from a plain reading of the page, and all four are set up the same way, by what the column's instruction asks for. [Custom Columns](/docs/concepts/custom-columns) covers how to write each kind.

Where an account sits on the [Model Tier](/docs/reference/glossary#model-tier) scale sets which AI model does the work, and therefore how many [credits](/docs/reference/glossary#credits) each file costs. When one real-world document arrives as several photos or pages, [Multi-Page Merge](/docs/reference/glossary#multi-page-merge) can fold those results back into a single record.

## Where to go next

This page is the orientation, not the manual. The shortest path to a first result is on [Quickstart](/docs/quickstart); the vocabulary the rest of the site uses is defined in [Glossary](/docs/reference/glossary).

- [**Quickstart**](/docs/quickstart) — the shortest path from signing up to your first structured table.
- [**Templates & Columns**](/docs/concepts/templates-and-columns) — what a template is, what a column is made of, and where saved templates can be reused.
- [**Custom Columns**](/docs/concepts/custom-columns) — how a column can read a value directly, compute one, infer one, or locate a visual element.
- [**Model Tier**](/docs/concepts/model-tier) — the processing-quality level behind an account and what it means for credits.
- [**Multi-Page Merge**](/docs/concepts/multi-page-merge) — folding a document that arrived as several files back into one record.
- [**Batches**](/docs/concepts/batches) — how files are grouped, named, and tracked through processing.
- [**Limits & Specs**](/docs/reference/limits-and-specs) — supported file types, page and size caps, and plan limits.

For building on the API rather than the web app, the developer documentation is a separate site: see the [API documentation](/developers/).

---

Source: https://imagetotable.ai/docs/