Perplexity

The company does not claim ownership of user Input, which remains 'Your Content'. However, users grant a broad, royalty-free, transferable, sublicensable, worldwide and irrevocable license to use, host, modify and distribute that content, which materially dilutes practical ownership.

Output is treated as part of 'Your Content' which the company does not claim to own, but the company and its licensors expressly retain ownership of the Services and any technology used to generate Output. Users are warned that identical Output may be generated for and used by other users, limiting exclusivity.

The Privacy Notice and Terms indicate user inputs and content may be used to improve or create AI models, and the content license explicitly permits use to 'improve' the Services. There is a narrow carve-out only for Email Assistant content, which is not used to train models; there is no general opt-out from model improvement in the consumer documents.

The supplied documents contain no defined retention schedule, no deletion SLA, and no clear self-service deletion mechanism for user content. The Terms state the company is not obligated to delete Your Content and bears no responsibility for its deletion or failure to delete, leaving retention effectively at the company's discretion.

The Privacy Notice discloses sharing with affiliates, service providers, business partners, professional services, and advertising partners for targeted advertising. The company states it does not sell personal data for money and does not send queries/prompts to advertisers, but data such as identifiers, commercial info, internet activity, and geolocation is shared with advertising partners for targeted advertising, and sensitive data may be transferred at the user's direction.

The documents provide several opt-out mechanisms: opting out of arbitration within 30 days, opting out of push/SMS messages, and a 'Your Choices' section referencing access and third-party recipient rights. However, the privacy 'Your Choices' section is truncated in the supplied text and no clear general opt-out from model improvement is described, so overall opt-out coverage is partial.

The supplied documents reference GDPR-style legal bases (legitimate interest, consent, contract) and California privacy concepts (CCPA/CPRA, sensitive data under Cal. Civ. Code § 1798.121), plus FERPA is referenced within a data-category table. However, no compliance certifications or attestations (SOC 2, ISO 27001/42001, NIST) are claimed, and no third-party audit evidence is presented, which is a notable gap for a tool handling health and biometric data.

The documents provide no commitments to model explainability, transparency reports, or enterprise auditing capabilities. They warn that Output may be incorrect, biased, or incomplete but offer no mechanism to audit or interrogate model behavior.

The supplied documents disclose no specific security controls (no mention of encryption at rest/in transit, penetration testing, bug bounty, or incident response), and reference no trust center or security page. The company expressly disclaims responsibility for the security of the Services, and no breach history is provided.

The consumer documents explicitly state that Enterprise and API offerings are governed by separate terms and that Perplexity acts as a service provider/processor for those, implying a materially different (and likely more protective) data posture there. However, the enterprise terms themselves were not supplied, so the actual delta cannot be verified from the provided text.

The Terms reserve broad rights to remove, screen, edit, or delete Your Content and to disclose it where deemed necessary, and the Email Assistant analyzes email content. The documents do not explicitly state routine human review of prompts, but the discretion to screen content and disclose to third parties implies staff access is possible.

The documents disclose that the company may share data with law enforcement, participate in legal processes, and disclose data without consent to defend rights or investigate illegal activity. A mandatory arbitration clause and class-action waiver channel disputes away from courts. No pending litigation or specific government requests are disclosed.

The Privacy Notice discloses collection of extensive PII (name, contact, device/network identifiers, IP, location, browsing history, purchase and payment data) and a broad range of sensitive personal information including account credentials, biometric/voice data, government IDs, health, genetic and neural data, and potentially racial/ethnic, religious, immigration, and sexual-orientation data submitted via prompts. While disclosure is fairly detailed and some is consent-based, the sheer breadth of SPI collected combined with advertising-partner sharing of non-sensitive categories warrants a high-risk rating.

Both the Terms (Last updated January 23, 2026) and Privacy Notice (Last updated July 8, 2026) are current relative to the August 2026 analysis date and demonstrably address the AI capabilities the product ships, including AI model training/improvement, third-party LLMs, the Comet browser, ProShop, Email Assistant, and advertising measurement referenced in the product surface. The policies cover the visible product features, supporting a GREEN rating.

Two documents were supplied (Terms of Service and Privacy Notice) and no material contradictions were identified between them. The Terms' training/improvement license aligns with the Privacy Notice's statement that data may be used to improve AI models, and both are consistent on third-party sharing and dispute resolution.