Amazon Comprehend

What it does

Amazon Comprehend takes in raw text, and for some APIs, PDF and Word documents, then returns structured results such as entities, key phrases, sentiment, syntax, language, topics, and detected personally identifiable information (PII). It can also return the locations of PII so a document can be redacted.

People typically use it to analyze customer support tickets, product reviews, emails, survey responses, and insurance or mortgage documents, or to power search that indexes key phrases and sentiment rather than only keywords. Custom Classification and Custom Entities let teams train a model on their own labels, and the service supports toxicity detection and prompt safety classification for generative AI inputs.

It suits developers and data teams building text-analysis applications on AWS, and organizations with sensitive text that needs PII detection. Use requires an AWS account and is billed by usage, with a 12-month free tier of 50K units per eligible API. Custom Comprehend has no free tier, and endpoints for synchronous custom inference bill while running, so costs should be estimated before deployment.

What it can do

• Entity recognition, sentiment analysis, syntax analysis, key phrase extraction, and language detection on text
• Detect PII and Contains PII APIs for finding and flagging personally identifiable information, including redaction
• Custom Classification and Custom Entities training for organization-specific categories and terms
• Topic modeling over document collections stored in Amazon S3
• Toxicity detection and prompt safety classification for generative AI applications
• Asynchronous and synchronous inference options for custom models

Works with

• Amazon Textract (used to extract text from scanned PDFs before entity detection)
• Amazon S3 (source location for topic modeling jobs)

Example workflows

Redact PII from archived customer comments

Input: customer comments of about 550 characters each. Steps: send each comment to the Detect PII API, receive entity locations, and mask those spans before archiving. Output: redacted text versions of each comment.

Categorize incoming feedback with a custom classifier

Input: website feedback text plus a labeled training set. Steps: train a Custom Classification model, then run asynchronous inference on new feedback batches. Output: a category label for each piece of feedback.

Before you choose

• Applicable text-analysis and asynchronous custom APIs charge in 100-character units with a 300-character minimum. Topic modeling instead charges by total document size per job: a flat charge for the first 100 MB, then per MB. Training and real-time endpoints have separate meters.
• Custom Comprehend has no free tier, and real-time endpoints bill from start until deletion even when idle.
• Scanned PDFs require Amazon Textract, which is billed separately.

Listed under

• Data
• Analytics and BI
• API
• Call center analytics
• Custom classification
• Document classification
• Document processing
• NLP

Get started

Free tier is available for eligible APIs for 12 months from first request.

1. Create or sign in to an AWS account and open the Amazon Comprehend console.
2. Choose the API you need (for example, Detect PII or Key Phrase Extraction) and confirm the AWS region used for requests.
3. For custom models, prepare labeled training data and train a Custom Classification or Custom Entities model before choosing asynchronous or endpoint-based inference.

Eligible APIs include a 12-month free tier of 50K units (5M characters) per API per month, available to new and existing AWS customers from their first request. Custom Comprehend has no free tier. Volumes above 100M units per month require contacting AWS for pricing. Use the AWS Pricing Calculator for estimates.

Source: aws.amazon.com/comprehend/pricing, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• Natural Language Processing – Amazon Comprehend – AWS ↗aws.amazon.com
• Amazon Comprehend Pricing ↗aws.amazon.com

Company website

aws.amazon.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 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.