Perplexity
Perplexity is an AI answer engine that answers any question in writing with inline citations. Built for researchers, analysts, teams and developers, it orchestrates several leading models and adds file search, the Comet browser and a developer API.
What is Perplexity?
Perplexity is an AI answer engine: you type a question and get a written answer carrying inline citations that point to the source behind each claim. Rather than running a single model, it orchestrates several leading ones and selects the best per query; the pricing page names ChatGPT, Google Gemini, Anthropic Claude and NVIDIA Nemotron, while Pro and Max let you pick a preferred model yourself. Model Council and Brain are reserved for Max.
Four products sit on that foundation. Search handles sourced research across the web, your files and your applications, with more than 400 app connectors from Pro upwards and premium data sources (PitchBook, Statista, S&P Capital IQ) on Pro and Max. Computer runs longer tasks, producing documents, assets and applications against a monthly credit allowance. Comet is an assistant browser for Mac, Windows, iOS and Android that combines Comet Assistant and Comet Agent, keeps part of its data (history, site permissions, downloads) locally and offers an Incognito mode. An API platform exposes Gateway (open-weight models hosted by Perplexity behind an OpenAI-compatible endpoint), Agent (third-party models with built-in web search and search presets), Search (ranked web results) and Embeddings (pplx-embed, for retrieval-augmented generation and semantic search). Agent tools billed per invocation include web_search, fetch_url, people_search, finance_search and a code execution sandbox.
An Email Assistant reads sent and received messages to suggest replies, without that content feeding model training. The Enterprise tier adds SSO and SCIM, user management, configurable file retention with automatic deletion down to one day, and private search; Perplexity says more than 50,000 organizations use it. Published sector use cases span finance, legal, consulting, healthcare, tech, government, education and advertising, with customer references including Copy.ai, Doximity, Zoom, SAP Joule, Apollo, Rox and NationGraph.
The publisher is Perplexity AI, Inc., with offices announced in San Francisco, Palo Alto, New York and Seattle, and it founded the Secure Intelligence Institute in 2026 for research into the security of AI systems. The stated limits come from its own documents: the service is provided as is, and generated answers may contain mistakes.
What it does
- Ask a question and receive a written answer with inline citations
- Run a Deep Research investigation on a complex subject
- Query your own files and your organization's connected applications
- Choose a preferred model per query on Pro and Max
- Delegate browser tasks to Comet: email, shopping, scheduling, monitoring
- Produce documents and applications with Perplexity Computer
- Build products on the Gateway, Agent, Search and Embeddings APIs
When to use Perplexity / When not to
A quick filter to help you decide if Perplexity is the right fit.
When to use Perplexity
- Researchers and analysts who need traceable answers: every response carries inline citations showing exactly where each claim comes from.
- Consultants, finance and legal professionals running deep investigations, with Deep Research from Pro, expert-level research on Max, and premium sources such as PitchBook, Statista and S&P Capital IQ on the paid plans.
- Teams working on their own material, uploading files and querying the documents and connected applications of their organization; announced use cases cover marketing, sales, product and IT.
- Developers building search or retrieval products on the four APIs (Gateway, Agent, Search, Embeddings), with an OpenAI-compatible endpoint, SDK, CLI and MCP servers.
- Busy professionals who want routine browser work delegated: the Comet browser handles email, shopping, scheduling and coding tasks.
When not to use Perplexity
- Anyone who needs guaranteed accuracy: clause 8.1.1 of the terms supplies the service AS IS and AS AVAILABLE, with no warranty of accuracy, availability or reliability, and the API documentation states that responses are generated using AI and may contain mistakes.
- Children under 13, who are excluded by clause 1.2 of the terms; minors below the age of majority also need the agreement of a parent or guardian.
- Privacy-conscious individuals who expect training to be off by default: on Free, Pro and Max, data feeds AI training unless the AI Data Retention setting is turned off by hand.
- Organizations bound to European data residency: hosting is announced in North America, and no European storage option appears on the pages collected.
- Developers hoping for a permanently free API tier: API access is billed on usage, by token, by request and by tool invocation, with no free allowance.
How to use Perplexity
A typical end-to-end flow, from setup to results.
- Open the Perplexity website and ask a first question: basic free use requires no account.
- Create an account to keep your history, your preferences and your workspaces.
- Type into the single Ask anything field on the home page, or choose Work in a project for longer assignments.
- Subscribe to Pro or Max from the pricing page; Enterprise goes through a sales form.
- Install the iOS or Android application from the App Store or Google Play.
- Download the Comet browser separately for Mac, Windows, iOS or Android.
- Open Settings, then Preferences, to turn off AI Data Retention if you do not want your data used for training.
- For development work, create a key in the API console and point an existing OpenAI-compatible client at the endpoint.
- Install the SDK or the CLI, or connect the MCP servers provided; the API documentation is indexed for agents through its llms.txt file.
- Buy API credits directly or through AWS Marketplace.
Pros & Cons
Pros
- Every answer carries inline citations, so each statement can be traced back to its source.
- A single subscription reaches several leading model families instead of stacking separate accounts.
- The free plan is permanent, with no time limit.
- The API documentation is public and detailed, with a complete price grid and an OpenAI-compatible endpoint that accepts an existing client.
- Zero Data Retention is documented for the Chat Completions API, and every consumer plan can switch off the use of data for training.
- The DPA is published and reachable without a prior request, Article 27 representatives are named for the EU and the UK, and a DPO is designated.
- SOC 2 Type II, a 2025 HIPAA gap assessment and CAIQlite are held by Perplexity itself, alongside detailed public security documentation and availability on web, iOS, Android and the Comet browser for Mac and Windows.
Cons
- No about page and no contact page: /hub/about and /hub/contact return a not-found page, and no dedicated legal or GDPR email address exists, everything going through general support.
- Use of data for training is active by default on Free, Pro and Max, so the user has to go and switch it off, and switching it off only applies going forward: data already used is not withdrawn.
- The subprocessor list and the audit reports sit behind the Trust Center and cannot be consulted freely.
- Subscriptions are non-refundable (clause 2.4), and clause 9 imposes individual arbitration with a class-action waiver unless the user opts out in writing.
- Hosting is announced in North America, with no European data residency option on the pages collected.
- No list of interface languages is published.
- The address in the terms of service (2261 Market Street) differs from the developer address declared on Google Play (115 Sansome St), and the site returns 403 errors to non-browser clients, which hinders automated verification.
Pricing & Plans
Perplexity operates a permanent free plan: Free costs 0 USD per month with limited daily usage. The lowest paid entry point is Pro at 20.00 USD per month, followed by Max at 200 USD per month; Enterprise pricing is not published and is quoted through a sales form. The API is billed on usage rather than by subscription: 5.00 USD per 1,000 requests for the Search API; 1 USD per million input and output tokens for Sonar, and 3 and 15 USD respectively for Sonar Pro; embeddings from 0.004 to 0.05 USD per million tokens; and Agent API tools charged per invocation (web_search 0.0025 USD, fetch_url 0.00025 USD, people_search and finance_search 0.005 USD, sandbox 0.03 USD per session). Credits may be purchased directly or through AWS Marketplace. Subscriptions are non-refundable.
- search with sourced answers and citations
- limited daily usage
- access to base models.
- extended Perplexity Computer access
- 4
- 000 bonus credits
- Deep Research
- access to the best models
- model selection
- document and application creation
- higher limits and expanded memory.
- maximum Computer usage
- 35
- 000 bonus credits and 10
- 000 monthly credits
- expert-level research
- frontier models
- the highest limits and priority access to new features.
- SSO and SCIM
- user management
- configurable file retention
- and no use of data for AI training.
- billing per token
- per request and per tool invocation
- with credits bought directly or through AWS Marketplace.
Data, GDPR & hosting
A consolidated view of how Perplexity handles your data.
GDPR overview
The Enterprise page states that Perplexity is GDPR and HIPAA compliant. The privacy notice was last updated on July 8, 2026, and the terms of service on January 23, 2026. Legal bases are declared by context: legitimate interests, performance of the contract, legal obligation and consent. VeraSafe Ireland Ltd. (Cork) and VeraSafe United Kingdom Ltd. (London) act as Article 27 representatives for the EU and the UK, and VeraSafe, LLC is the designated DPO, reachable at experts@verasafe.com. Transfers rely on certified adherence to the EU-U.S. Data Privacy Framework and its UK Extension, with standard contractual clauses available on request; the published DPA refers to the clauses of Decision 2021/914. Rights include access, portability, the list of third-party recipients, rectification, erasure, objection, restriction, withdrawal of consent and appeal, exercised through a dedicated form or support@perplexity.ai. Global Privacy Control is honored as a valid objection; Do Not Track is not.
Who owns the data?
Clause 6.4.1 of the terms states that Perplexity claims no ownership of Your Content, which includes your inputs. The company and its licensors do own the Services and the technology used to generate any Output. Under clause 6.4.2 you grant Perplexity a worldwide, royalty-free, transferable and sublicensable license, irrevocable for as long as the content is stored, to host, reproduce, transmit, display, publish, distribute and modify that content in order to operate, improve, promote and provide the Services. Feedback you send becomes the company's exclusive property (clause 6.3). For enterprise customers, the DPA makes Perplexity the processor and the customer the controller.
Reuse rights
The terms do not hand the user explicit ownership of generated Output: Perplexity claims no ownership of Your Content, which covers your Input, while it and its licensors keep the Services and the technology that produces Output. On the processing side, the privacy notice lists among the purposes to improve or create services and products, including our AI models. The AI Data Retention setting is on by default for Free, Pro and Max and is switched off in Settings, then Preferences; switching it off applies only going forward and does not erase data already used. Enterprise customer data is never used for AI training, and the Chat Completions API runs with Zero Data Retention, keeping only billing metrics (token counts, model, timestamp, key identifier). Email content processed by the Email Assistant is not used to create, train, improve or fine-tune models. Perplexity states that it does not sell personal data and passes neither queries, prompts nor conversation content to advertisers, but it does share data with group companies, service providers, business partners, professional advisers and advertising partners for ad measurement.
Data retention & training
Hosting summary
The API FAQ is explicit: compute is hosted via Amazon Web Services in North America. The security page describes separate AWS accounts for production, staging and test environments, with access managed through AWS IAM behind single sign-on, multi-factor authentication and short-lived session credentials. Cloudflare provides DDoS protection, a web application firewall, rate limiting and TLS encryption; DNS resolution for the domain returns 104.18.27.48, on Cloudflare's anycast network. Cloud posture is monitored with Wiz and logs are aggregated in a Panther SIEM. Storing customer data on employee workstations is prohibited, and data stays confined to production environments. No European data residency option appears on the pages collected, so European customers should expect North American hosting, covered contractually through the EU-U.S. Data Privacy Framework and standard contractual clauses rather than by local storage.
Things to keep in mind
Risks and trade-offs to weigh before adopting Perplexity.
- Inline citations make an answer feel already verified, yet clause 8.1.1 provides the service as is with no warranty of accuracy and the API documentation warns that responses may contain mistakes: opening the cited source remains the user's job.
- AI Data Retention is active by default on Free, Pro and Max, so conversations feed model improvement unless the user goes into Settings, then Preferences, and turns the setting off.
- Turning that setting off only applies going forward: data already used for training is not withdrawn.
- Clause 6.4.2 grants a license over stored content that is irrevocable for as long as the content is stored, including the right to modify and sublicense it, and clause 6.3 makes any feedback sent the exclusive property of the publisher.
- Clause 9 imposes individual arbitration and a class-action waiver unless the user objects in writing to support@perplexity.ai, and clause 2.4 makes subscriptions non-refundable.
- The privacy notice describes sharing with advertising partners for ad measurement; objection goes through the cookie banner and the Global Privacy Control signal, while Do Not Track is not recognized.
- Sensitive categories may be involved, including health data, voice biometrics and neural data through Perplexity Health, while the subprocessor list stays behind the Trust Center and hosting is announced in North America only.
Setup & Integrations
Technical difficulty
Very low for everyday use: a single input field on the home page, no installation, and no account needed for basic free use, an account becoming necessary only for history, preferences and paid plans. The mobile applications and the Comet browser are separate installs. For developers the effort stays moderate: create a key in the API console and, because the endpoint is OpenAI-compatible, point an existing client straight at it, with SDK, CLI and MCP servers provided. Enterprise deployment is the demanding end, since SSO and SCIM call for an IT team.
Deployment
Apps stores
Integrations
Behind Perplexity
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Perplexity.
Frequently asked questions
Does Perplexity cite its sources?
Is there a permanently free plan?
Which AI models does Perplexity use?
Is my data used to train the models?
Is there an API?
Where is the data hosted?
What is the minimum age to use Perplexity?
Is a data processing agreement available?
Which security certifications does Perplexity hold?
Are there mobile applications?
Should you pick Perplexity?
Perplexity presents itself as a general-purpose research and document-work tool rather than a chatbot: the answer arrives with its sources attached, and the model behind it is chosen per query from several leading families. Around that core, Computer executes longer tasks, Comet turns the browser into an assistant, and four APIs open the same machinery to developers.
The commercial offer is legible: a permanent free plan, a paid entry point at 20 USD per month with Pro, 200 USD per month with Max, and Enterprise pricing quoted on request. The API is billed on usage, with a public price list detailed down to the tool invocation.
Documentary transparency is a real strength. The DPA is published rather than gated, Article 27 representatives are named for the EU and the UK, adherence to the EU-U.S. Data Privacy Framework is claimed, and the security page describes access management, infrastructure, detection and vendor review. The privacy notice was last updated on July 8, 2026, and the terms of service on January 23, 2026.
The reservations are equally concrete. On Free, Pro and Max, the use of data for training is active by default and must be switched off in Settings, then Preferences, and switching it off only applies going forward. The subprocessor list and audit reports sit behind a closed Trust Center. There is no about page and no contact page, and no dedicated legal or privacy address. Hosting is announced in North America, with no European residency option on the pages collected. Subscriptions are non-refundable, and the terms impose individual arbitration unless the user opts out in writing.
For sourced research, access to several model families and an API you can price precisely, the proposition holds up, provided you go and change the data settings on day one.
- Choosing a selection results in a full page refresh.
- Opens in a new window.