AI Transparency & Data Handling

Publica.la integrates AI capabilities into the reader experience through a suite of tools that help readers comprehend, interact with, and get more value from digital content. This page documents what AI features exist, how data flows through them, what providers are involved, and what controls publishers have.

For the public-facing policy summary, see AI Transparency & Policy.

AI Feature Overview

All AI features are reader-facing, activated on-demand by the end user during a reading session. No AI processing occurs in the background or without explicit user action.

We make every effort to keep this list current, but it may not reflect the very latest additions at all times. Regardless, the commitments on this page regarding how content is processed, stored, and handled will not change without prior notice to publishers.

Feature	Description	Provider	Data Sent
Explain	Explains selected text in simpler terms	OpenAI / Anthropic	Selected text, book title, language
Expand	Provides additional context on selected text	OpenAI / Anthropic	Selected text, book title, language
Summarize	Generates summaries of selected text or page ranges	OpenAI / Anthropic	Selected text or page content, book title, language
Translate	Translates selected text to reader's language	OpenAI / Anthropic	Selected text, target language
Dictionary	Defines words and terms with part of speech	OpenAI / Anthropic	Selected word(s)
Assessment Questions	Generates comprehension questions from content	OpenAI / Anthropic	Page content (min. 150 words)
Q&A	Answers reader questions based on the publication text	OpenAI / Anthropic	Selected text passage, reader question
Text-to-Speech	Converts text to spoken audio	AWS Polly	Text content for the current page/section

Data Flow

What gets sent to AI providers

When a reader triggers an AI feature, only the following data is transmitted:

• Text excerpt: The selected passage or page content relevant to the request
• Metadata: Book title and language (used for context in prompts)
• Reader question: For Q&A features only, the question the reader typed

The following is never sent:

• Reader identity or personally identifiable information (PII)
• Authentication tokens or session data
• Full book content (only the relevant excerpt)
• Purchase history, reading habits, or behavioral data

Encryption

• All AI requests are encrypted using AES-256-CBC before transmission
• Data travels over TLS 1.2+ connections
• API keys are stored server-side; they are never exposed to the client

Response handling

• AI responses are delivered directly to the reader's device
• Each request is independent; no conversation history or context is maintained between requests

Data Retention

Our AI providers contractually commit to not retaining API inputs for model training:

• OpenAI: Zero-retention API option enabled. Inputs are not used for training or improving models.
• Anthropic: Zero-retention API. Inputs are not used for training or improving models.
• AWS Polly: Audio is synthesized in real time. No input retention.

AI Providers

OpenAI

• Used for: Text comprehension tools (explain, expand, summarize, translate, dictionary, assessment, Q&A)
• Compliance: SOC 2 Type II, GDPR compliant
• Data processing: API inputs are not used for training or improving models.
• DPA: Data Processing Agreement in place.

Anthropic

• Used for: Text comprehension tools (explain, expand, summarize, translate, dictionary, assessment, Q&A)
• Compliance: SOC 2 Type II, GDPR compliant
• Data processing: API inputs are not used for training or improving models.
• DPA: Data Processing Agreement in place.

AWS Polly

• Service: Text-to-Speech synthesis
• Compliance: SOC 2, ISO 27001, HIPAA eligible, GDPR compliant
• Data processing: Text is synthesized into audio in real time.

Tenant Controls

Publishers can opt out of AI features at any time. To disable AI for your catalog, contact your account manager or email [email protected].

Security

AI feature requests are protected by the same security infrastructure as the rest of the platform:

• JWT authentication: All API requests require a valid JSON Web Token
• Reader-token validation: AI features are only available to authenticated readers with valid access to the content
• Per-request authorization: Each AI request is authorized individually; there is no batch processing or background analysis
• Audit logging: AI feature usage metadata is tracked for billing and analytics (feature used, timestamp, tenant)

What Is Not AI

For clarity, the following platform features do not currently use AI, machine learning, or any form of computational intelligence:

Feature	Technology
Search	Traditional full-text database search on indexed metadata fields
Recommendations	Rule-based matching on BISAC (Book Industry Standards and Communications) category taxonomy
Content intake	ONIX XML parsing and PDF processing (Ghostscript); purely mechanical

Frequently Asked Questions

Does Publica.la use publisher content to train AI models?

No. Publisher content is not used for training, fine-tuning, or improving any AI model. Our AI providers (OpenAI, Anthropic, and AWS) contractually commit to not using API inputs for model training.

Can publishers opt out of AI features?

Yes. AI features can be fully disabled at the tenant level. Contact [email protected] to adjust settings.

What happens if an AI provider changes their data policies?

We continuously monitor our providers' data handling policies. If a provider changes their terms in a way that conflicts with our commitments, we will either negotiate acceptable terms or migrate to an alternative provider. Publishers will be notified of any material changes.

Does AI processing comply with GDPR?

Yes. No personal data is sent to AI providers. Our providers maintain GDPR-compliant data processing agreements. For details on our broader privacy practices, see our Privacy Policy.

Are AI-generated outputs considered derivative works?

AI outputs (explanations, translations, summaries) are delivered to the individual reader who requested them and are not redistributed. They function as reading aids, similar to a dictionary or thesaurus lookup.

Can we get a written statement about AI usage for our agreements?

Yes. Contact [email protected] and we will provide documentation suitable for inclusion in your distribution or licensing agreements.