Document AI

What it does

Document AI takes documents such as scanned pages, PDFs, images, Word, Excel, PowerPoint, and HTML files and turns their contents into structured output. Depending on the processor, it can return raw text and layout, extract named fields and key-value pairs, classify a document's type, or split a composite file into its component documents. Processing is billed per page, and the results can be sent on to other Google Cloud services.

People typically use it through the Google Cloud console or the Document AI API to build a processor, which then exposes a prediction endpoint for their documents. Custom extractors can be used out of the box or fine-tuned with as few as 10 documents, and custom splitters and classifiers can be trained on an organization's own document classes. The documented BigQuery integration writes extracted metadata to a BigQuery objects table, so parsed fields can be joined with other tables.

Document AI suits developers and technical teams in organizations that process high volumes of forms, invoices, identity documents, or archives. It requires a Google Cloud account and project, with a free trial credit for new customers. Some pretrained processors are limited-access and need an approval request, and availability varies by region. Google's own page advises against entering sensitive or personal information and requires users to be 18 or older.

What it can do

• Extract structured fields from documents with a generative-AI custom extractor that can be fine-tuned with as few as 10 documents.
• Split composite files into single-class documents with a custom splitter.
• Classify documents by type with a custom classifier, including few-shot learning.
• Digitize scanned documents with Enterprise Document OCR, returning text and layout information.
• Parse tables and key-value pairs from structured forms with Form Parser.
• Monitor precision, recall, and F1 score for each processor in the Google Cloud console.

Works with

• BigQuery: extracts document metadata into a BigQuery objects table for joining with other tables.
• Cloud Storage: named on the overview as a service to store and organize documents, without a described integration mechanism in the reviewed information.

Example workflows

Extract fields from invoices

Input: a batch of supplier invoices as PDF or image files. Steps: create a custom extractor processor in the console, optionally fine-tune with about 10 labeled documents, send the files to the prediction endpoint, and review extraction precision in the console. Output: structured records with named fields suitable for a database or BigQuery table.

Split a mortgage package

Input: a single composite PDF containing an application, income verification, and photo ID. Steps: create a custom splitter processor, send the composite file, and receive the individual logical documents. Output: separate single-class documents ready for classification or extraction.

Before you choose

• Pricing is per processed page, so high-volume workloads should be estimated with the pricing calculator before rollout.
• Some processors, including certain pretrained parsers and the procurement splitter, are limited-access and require an approval request that needs a Google Cloud project ID.
• Google's page instructs users not to enter sensitive, confidential, or personal information and states use is limited to adults 18 or older.

Listed under

• Operations
• Data extraction
• Developers
• Document classification
• Document processing
• Enterprise
• OCR

Get started

New customers receive $300 in free credit; the API setup is a separate step from the console.

1. Create a Google Cloud account and project, then open the Document AI page in the Google Cloud console.
2. Create a processor for your task (for example, a custom extractor) and enable the Document AI API for the project.
3. Send a test document to the processor's prediction endpoint and review results in the console before scaling up.

New customers receive $300 in free credit, which is a time-limited trial allowance and not a permanent free tier. Enterprise Document OCR's first 1,000 pages are free. Other processors are billed per 1,000 pages, with volume-tier rates and discounts that require opt-in to consumption models. Pretrained parsers bill per count, where one count equals up to 10 pages. Custom processor hosting is billed hourly. Failed requests are not billed.

Source: cloud.google.com/document-ai/pricing, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Document AI | Google Cloud ↗cloud.google.com
• Document AI pricing | Google Cloud ↗cloud.google.com
• Document AI overview  |  Google Cloud Documentation ↗cloud.google.com

Company website

cloud.google.com: product, pricing and policy pages

YC directory mirror

Batch and company name

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.