Perplexity AI: Privacy, Training & Output Ownership

Sensitive Data

User inputs and files are retained for 30 days and utilized for model training by default.

Training

Data is used for 'AI Data Enhancement' unless the user manually opts out in account settings.

In the free tier, Perplexity and its model partners (e.g., Anthropic, OpenAI) may utilize your prompts and uploaded documents to refine future iterations of their LLMs. This is the default 'AI Data Enhancement' setting.

Output Ownership

The user owns the outputs generated by the service, though Perplexity retains a license to use inputs for service operation.

Perplexity explicitly states that they do not claim ownership of user content or the resulting output. However, users are cautioned that output may not be unique across different accounts due to the nature of web-search synthesis.

Data Retention

Uploaded documents and images are typically retained for 30 days to facilitate continuous conversation and context retrieval.

Security Measures

Standard encryption for data in transit and at rest is applied, but the tier lacks advanced enterprise-grade administrative controls.

Your Rights & Control

Users have the right to access, delete, and export their search history, but data already incorporated into model training during the default opt-in period may be difficult to fully purge.

Special Considerations

The free tier includes advertising cards which may involve limited tracking of search topics for ad personalization.

Sensitive Data

Offers a cleaner privacy environment without ads, but remains under standard consumer terms with no BAA or DPA.

Training

Opt-out from model training is the default configuration for paid Pro subscribers.

While the default for Pro accounts is to exclude data from training, users should verify their 'AI Data Enhancement' toggle remains off. Perplexity respects this setting for its internal models and upstream API providers.

Output Ownership

Users maintain ownership of inputs and outputs with similar licensing grants as the free tier.

Full ownership of output is assigned to the user. This tier provides access to advanced models (Claude, GPT-4o), ensuring professional-grade content generation under user ownership.

Data Retention

Chat histories are generally retained for up to 90 days to support 'Spaces' and research persistence, though users can manually delete threads at any time.

Security Measures

Enhanced security relative to the free tier, including the removal of all third-party tracking pixels and search-related advertisements.

Your Rights & Control

Users benefit from standard GDPR/CCPA rights, including data portability and the right to be forgotten.

Special Considerations

Ideal for individual power users and researchers who require privacy-by-default without the administrative overhead of a team plan. However, the lack of BAA or contractual confidentiality guarantees makes this tier unsuitable for attorney-client privileged data or PHI.

Sensitive Data

Fully compliant with SOC 2 Type II, with HIPAA Gap Assessment completed in 2025 and BAAs available for enterprise customers. Configurable data retention as low as 24 hours.

Training

Strict zero-training policy; customer data is never used to improve global models.

Enterprise agreements legally prohibit Perplexity and its third-party model providers from using customer content (Inputs or Outputs) for training, fine-tuning, or retraining any AI models.

Output Ownership

The organization retains all ownership rights in inputs and owns all outputs; Perplexity assigns all legal rights to the customer.

Ownership is clearly defined at the organizational level. Perplexity asserts no rights over outputs and provides legal indemnification for certain intellectual property claims in the Max tier.

Data Retention

Administrators have granular control over data retention policies, allowing for automatic deletion of files and search logs after 1, 7, or 30 days to meet internal compliance standards.

Security Measures

Includes Single Sign-On (SSO), SCIM for user provisioning, audit logs, and SOC 2 Type II certified infrastructure. The Sonar API (developer-facing) operates with zero data retention by default.

Your Rights & Control

The organization controls all user rights, including the ability to perform discovery across the team's 'Spaces' and manage permission levels globally.

Special Considerations

The 'Perplexity for Public Safety' initiative provides these enterprise-grade protections specifically for high-stakes government and law enforcement use cases. BAAs are available for enterprise customers through Perplexity's sales team.