Does Perplexity Train on Your Data? Consumer, Enterprise, and API Rules

Does Perplexity train on your data? Perplexity says some consumer data may be used for AI model training unless the user opts out. Its current help guidance names Free and the paid Pro and Max plans as consumer products with AI data retention enabled by default. Enterprise customer content and API customer content receive a separate contractual exclusion from model training. A defensible company policy has to name the plan and route because the Perplexity logo does not reveal which rule applies.

TL;DR

• Perplexity says AI data retention is on by default for Free and the paid Pro and Max plans. Users can turn it off for future training data.
• The consumer opt-out is prospective; Perplexity says previously collected training data cannot be removed through that setting.
• Enterprise terms prohibit Perplexity from using customer content to train or improve generative AI models. The prohibition also covers its third parties.
• Perplexity API terms provide the same exclusion for API customer input and output, while operational usage data remains a separate category.

Does Perplexity train on your data in consumer plans

Perplexity's data collection help article says some personal data may be used to train AI models and improve search quality unless the user opts out. It identifies Free and the paid Perplexity Pro and Perplexity Max plans as the affected consumer products. For those accounts, AI Data Retention is enabled by default.

The setting appears in Account preferences. When the AI data retention toggle is enabled, Perplexity says data is collected for AI training. Turning it off excludes data collected after the opt-out date. The company also says previously collected training data cannot be deleted or removed through the opt-out.

Perplexity's consumer Privacy Notice supports the broader data-use picture. It applies to the public website and apps, including Comet and other consumer services. It expressly excludes Enterprise and API offerings where Perplexity acts as a service provider or processor. The notice says Perplexity processes prompts and queries. It also processes uploads, outputs, and other user content, and may improve or create products, including AI models.

A Pro or Max payment therefore does not create the Enterprise rule. Those plans remain in the consumer opt-out framework named by Perplexity. A procurement record should not treat "paid Perplexity" as a product class.

The consumer opt-out has a narrow effect

Turning off AI Data Retention addresses model-training use of future consumer data. Perplexity says core features remain available, while data may still be processed for service operation and legal compliance. Product improvement remains another stated processing purpose. An enterprise must review each purpose.

Search history and uploaded files can each have different handling. The same applies to account information and device data. Feedback and generated content require their own review. Comet browsing and local browser data add another product surface. Perplexity's Privacy Notice also gives Email Assistant a specific rule: it says email content is not used to create or train AI models. The statement also excludes improving or fine-tuning those models. A feature-specific statement should not be expanded into a claim covering every consumer product.

The practical control is simple. If company policy allows a consumer account at all, verify the account and capture the setting. Restrict the data class before the request. Do not assume an individual user will preserve the toggle after a device change or account change. The Perplexity enterprise security review explains why a managed organization provides stronger identity and administrator evidence than a personal setting.

Enterprise content is excluded by contract

Perplexity's Enterprise Terms define input and output together as Customer Content. Section 1.3.3 says Perplexity will not use, or authorize third parties to use, Customer Content to train or retrain generative AI models. The same clause bars fine-tuning or otherwise improving those models with Customer Content.

The current Enterprise help guidance repeats that commitment for Enterprise Pro and Enterprise Max. It says uploaded files are retained for seven days, while threads and organization data can remain until deletion. It also says third-party AI providers are subject to zero-training protections. Training exclusion and retention are therefore different facts. A thread can be retained without becoming model-training data.

Enterprise terms also distinguish Customer Content from Usage Data. Perplexity may use anonymized or aggregated operational data that does not include Customer Content or identifiable personal information. It may use that data for lawful purposes such as measuring feature popularity or improving categorization. A review should preserve the definition rather than rewriting the contractual no-training promise as "Perplexity uses no enterprise data."

If an Enterprise subscription ends and the organization downgrades to a consumer plan, Perplexity says future activity follows the consumer policy. The user must then turn off AI data usage to prevent future activity from improving models. That transition belongs in offboarding and renewal procedures.

The API has its own no-training terms

The Perplexity API Terms apply to Sonar and Agentic Research API services, not the public website or Enterprise product. They define customer Input and Output as Customer Content. The terms say Perplexity will not use, or authorize third parties to use, that content to train or retrain generative AI models. They also bar fine-tuning or otherwise improving those models with the content.

The API terms permit service processing needed to provide the API and comply with law. They also separate operational Usage Data from Customer Content. Perplexity can analyze non-identifying metadata and performance measures for lawful purposes. Third-party models can have additional terms, so an enterprise should record the selected model route and the current third-party terms list.

An API key alone is weak identity evidence. It proves that a request used a credential. It does not identify the employee or agent behind the request, or the application that initiated it. The approval should pair the API project with authenticated caller identity and an approved data class. It should also state the destination rule and retain the decision record. AI data loss prevention controls should run before the prompt reaches Perplexity, not after the provider's no-training rule takes effect.

A product boundary makes the answer testable

A Perplexity approval should name the consumer account or Enterprise organization. For API use, it should name the API customer relationship. The record should state the training rule and retention behavior. It should also cover enabled features, the third-party model path, and approved data classes. Add an accountable administrator and review date.

An employee may use Perplexity Pro. Separately, a developer may use the Perplexity API while procurement owns Enterprise seats. Each route follows different data rules. Evidence must follow the actual request.

A no-training commitment also does not approve sensitive data by itself. A healthcare workflow still needs the contract and service checks described in Perplexity HIPAA compliance. A legal or financial workflow still needs a defined purpose and access controls. It also needs minimization and retention controls. Provider training is one column in the decision, not the entire decision.

DeepInspect

DeepInspect can inspect authenticated HTTP requests deliberately routed to an approved Perplexity API endpoint. It can classify the assembled prompt and bind the request to an application or user. It can then enforce route and data rules before transmission. Each decision produces a tamper-evident record.

It cannot operate the consumer opt-out or administer an Enterprise organization. It also cannot change Perplexity retention or prove provider-side model-training behavior. The organization first chooses the correct Perplexity product and contract. DeepInspect applies customer policy only to the API traffic that crosses its boundary.

Book a technical deep dive at deepinspect.ai.