Analyzer Guide

Learn how to detect PII entities in your text

The Analyzer is the first step in the anonymization process. It scans your text and identifies personally identifiable information (PII) like names, emails, phone numbers, and more.

How the Analyzer Works

The Analyzer uses multiple detection methods to identify PII:

Pattern Matching

Regular expressions detect structured data like email addresses, phone numbers, credit cards, and IBANs with high accuracy.

Machine Learning (NER)

Named Entity Recognition models identify context-dependent entities like person names, organizations, and locations using spaCy, Stanza, and Transformers.

Checksum Validation

Credit cards, IBANs, and other financial identifiers are validated using checksum algorithms (Luhn, MOD-97) for reduced false positives.

Using the Analyzer

Step 1: Enter Your Text

Navigate to the Anonymize page
Paste or type your text in the input area
The interface shows a character count and token estimate

Step 2: Select Entity Types

Choose which types of PII to detect:

Category	Entity Types	Example
Personal	PERSON, EMAIL_ADDRESS, PHONE_NUMBER	John Doe, john@email.com
Financial	CREDIT_CARD, IBAN_CODE, SWIFT_CODE	4111-1111-1111-1111
Location	LOCATION, ADDRESS, COORDINATES	123 Main St, New York
Government	SSN, PASSPORT, DRIVER_LICENSE	123-45-6789
Technical	IP_ADDRESS, MAC_ADDRESS	192.168.1.1

Tip: Use Presets

Instead of selecting entities manually, use Presets to quickly apply common entity configurations like "GDPR Compliance" or "Financial Data".

Step 3: Select Language

Choose the language of your text for optimal detection accuracy:

Auto-detect - Let the system determine the language
Specific language - Select from 27 supported languages

Language Selection Matters

Selecting the correct language significantly improves detection accuracy, especially for person names and locations.

Step 4: Run Analysis

Click the Analyze button
Wait for the analysis to complete (typically 1-3 seconds)
Review the detected entities in the results panel

Understanding Results

After analysis, each detected entity shows:

PERSONJohn Doe95% confidence

Position: characters 0-8

Result Fields

Entity Type - The category of PII detected (PERSON, EMAIL, etc.)
Text - The actual text that was identified as PII
Confidence Score - How certain the system is (0-100%)
Position - Start and end character positions

Confidence Threshold

Adjust the confidence threshold to control sensitivity:

Threshold	Effect	Best For
0.3 (Low)	More entities detected, more false positives	Maximum coverage, manual review
0.5 (Default)	Balanced detection and accuracy	General use
0.7 (High)	Fewer entities, higher confidence	Automated processing
0.9 (Very High)	Only very confident matches	Minimal intervention

Selecting Results

After analysis, you can refine which entities to anonymize:

Select/Deselect All

Use the checkbox in the header to select or deselect all results
Only selected entities will be anonymized

Individual Selection

Click individual checkboxes to include/exclude specific entities
Useful when the analyzer detects false positives
Useful when you want to keep certain information visible

Filter by Type

Click on an entity type badge to filter results by that type
Quickly select/deselect all entities of a specific type

Pro Tip

Review results before anonymizing. The analyzer may occasionally detect false positives, especially for names that are also common words.

Token Costs

Analysis operations consume tokens based on:

Cost = 2 + 1.0 × text_k + 0.2 × entities_enabled + 0.1 × entities_found

Final = ceil(Cost × 0.5)

Where:

text_k = text length in thousands of characters
entities_enabled = number of entity types selected
entities_found = number of entities detected

Cost Examples

Text Length	Entities	Typical Cost
100 characters	3 types, 2 found	2 tokens
1,000 characters	5 types, 5 found	3 tokens
5,000 characters	10 types, 15 found	6 tokens
10,000 characters	15 types, 30 found	10 tokens

See the Token System documentation for complete pricing details.

Best Practices

✅Select only the entity types you need - reduces costs and false positives

✅Use language-specific presets for better accuracy in non-English text

✅Review results before anonymizing, especially for names and locations

✅Use higher confidence thresholds for automated processing

✅Process text in reasonable chunks (under 10,000 characters) for best performance

Troubleshooting

Entity not detected?

Ensure the entity type is enabled in your selection
Try lowering the confidence threshold
Check that the correct language is selected
Verify the text format matches expected patterns

Too many false positives?

Increase the confidence threshold
Deselect broad entity types like LOCATION
Use entity-specific presets instead of selecting all

Analysis taking too long?

Break large texts into smaller chunks
Reduce the number of entity types selected
Use presets to avoid loading unused detection models

Next Steps

Anonymizer Guide

Learn how to anonymize detected entities

Presets

Save and reuse entity configurations

Last Updated: February 2026