Perplexity AI is an AI search engine that answers your question with a summary and references to the sources it used, rather than a list of links. The guide How to use Perplexity explains step by step how to apply each feature in your work.
The answer engine works in two steps: it searches the web, then a language model summarises the results. Alongside its own model, Sonar 2, Perplexity uses language models from providers such as OpenAI, Anthropic, Google and xAI.
The service is owned by Perplexity AI, Inc., a US company founded in 2022 that is not part of any large tech company. Its CEO is co-founder Aravind Srinivas.
The citations let you check every answer, but a source does not prove that the summary is accurate. What Perplexity does with your questions and uploads should also help you decide which data you enter into the answer engine.
What is Perplexity AI?
Perplexity is an answer engine from the US company Perplexity AI, Inc. When you ask a question, it searches the web and gives you a summary with references and links to the original sources.
Google responds to a search with a list of pages, and you choose which ones to open. Perplexity puts together an answer from several pages and lists the sources alongside it. A full question in plain language therefore works better than a few separate keywords.
You can use Perplexity at perplexity.ai and in the apps for iPhone, iPad and Android. According to the App Store, the iOS app supports 67 languages, including Dutch.
How does Perplexity AI work?
Perplexity works in two steps. First, the system searches the web for pages that match your question. A language model then uses those pages to generate an answer with source references.
How does Perplexity search the web?
Perplexity searches the web at the moment you ask your question. As a result, the answer can contain current information that is not yet in a language model’s training data.
Pro Search and Deep Research increase the search depth. According to the Perplexity Help Center, Pro Search runs several searches and combines dozens of sources into one answer. Deep Research, Perplexity says, runs dozens of searches, processes hundreds of sources and turns the results into a report.
How does Perplexity put together an answer?
Perplexity builds an answer by having a language model summarise the pages it found, with in-text references that link to the matching sources. You can ask follow-up questions in the same thread, and these build on the previous answer.
The default setting is called Best. According to the Perplexity Help Center, Best is not a separate mode but the automatic model selection within Search, which picks a suitable model for each question. On a paid plan, you can choose a model yourself instead.
Which models does Perplexity use?
Perplexity uses its own model, Sonar 2, as well as models from seven other providers, including OpenAI, Anthropic, Google and xAI. Your plan determines which of those models you can choose yourself.
The table below shows the Search models per plan, based on the overview in the Perplexity Help Center as of 4 September 2026. The changelog of 5 October 2026 also lists GPT-6.1 Sol and Claude Opus 5.5, without a breakdown by plan. Treat the model picker in your own account as the current list.
| Model | Provider | Pro | Max |
|---|---|---|---|
| Sonar 2 | Perplexity | Yes | Yes |
| GPT-5.6 Terra | OpenAI | Yes | Yes |
| GPT-5.6 Sol | OpenAI | No | Yes |
| Gemini 3.7 Flash | Yes | Yes | |
| Claude Sonnet 5 | Anthropic | Yes | Yes |
| Claude Opus 5 | Anthropic | No | Yes |
| Kimi K3 | Moonshot AI | Yes | Yes |
| GLM 5.3 | Z.ai | Yes | Yes |
| Grok 4.6 | xAI | Yes | Yes |
| Nemotron 3 Ultra | NVIDIA | Yes | Yes |
The free version gives you Best and a limited number of Pro Searches, but you cannot choose a model yourself. Choosing from the models in the table comes with Perplexity Pro and Max, which each have their own monthly price.
Model Council sends the same question to several models and summarises where the answers agree and where they differ. According to Perplexity, Model Council is available to Max users and to Pro users within Perplexity Computer.
Who owns Perplexity AI?
Perplexity is owned by Perplexity AI, Inc., a US company founded in 2022 and led by co-founder and CEO Aravind Srinivas. It is not part of OpenAI, Google or any other large tech company, and it raises money from investors.
Who founded Perplexity?
Perplexity was founded by four people with backgrounds in AI research and software:
- Aravind Srinivas, CEO, who previously worked at OpenAI and holds a PhD from UC Berkeley.
- Denis Yarats, CTO, previously a researcher at Facebook AI Research.
- Johnny Ho, previously an engineer at Quora.
- Andy Konwinski, who is also a co-founder of Databricks.
According to Srinivas, the first version of the search engine went live on 7 December 2022, a week after ChatGPT.
Who invests in Perplexity?
Perplexity is funded by venture capital firms and private investors. In the funding round Perplexity announced in January 2024, the investors included NVIDIA and Jeff Bezos (through Bezos Expeditions), and Databricks invested again.
What can you do with Perplexity?
You can use Perplexity to look up information with cited sources. It also offers five features for research and tasks:
- Deep Research, which runs dozens of searches and summarises the results in a report with sources.
- Spaces, where you group threads, files and standing instructions by project or topic.
- File upload, which lets you have a document summarised or ask questions about it.
- Comet, Perplexity’s web browser with a built-in AI assistant.
- Computer, an AI agent that breaks a requested outcome into tasks and carries them out with different models.
I tested Perplexity to understand how it fits into a professional tool stack. Which feature you pick depends on the task: Pro Search for a question that needs more sources, Deep Research for a report and Spaces for recurring research on one topic.
How reliable is Perplexity AI?
Perplexity’s answers are verifiable because each one cites its sources, but a source reference does not guarantee that the summary accurately reflects what the source page says.
A language model that generates a summary sometimes adds a claim or distorts one, even though the source says something different. This kind of error is called an AI hallucination, and it can happen even when Perplexity has found the right source.
What does Perplexity do with your data?
Perplexity processes your questions, prompts and uploads, and by default uses them to train its AI models. You can stop this by switching off the AI data retention setting in your account preferences.
According to the Perplexity Help Center, this setting is on by default for Free, Pro and Max. Switching it off only applies to data collected from that point onwards.
Enterprise accounts follow a different rule: Perplexity never uses Enterprise data to train AI models. For that data, Perplexity has agreements with model providers such as OpenAI and Anthropic not to retain it or train on it.
According to Perplexity’s privacy policy, last updated on 8 July 2026, VeraSafe is its EU representative under the GDPR, and Perplexity also processes data in the United States. In Incognito mode, Perplexity does not store your search activity between sessions.
How is Perplexity different from ChatGPT?
Perplexity searches the web and shows sources by default when you ask a question. ChatGPT, by contrast, is a general assistant for tasks such as writing, coding and conversation. According to OpenAI, ChatGPT searches the web when a question needs current information or when you select the search function. The main difference between Perplexity and ChatGPT therefore lies in what each tool does by default.
What does perplexity mean in language models?
In language models, perplexity is a metric for how well a model predicts a text. The lower the perplexity, the higher the probability the model assigns to the words that actually follow. Researchers use this value to evaluate language models during training.
Can you use Perplexity with client data?
Not with a personal account. Data you enter there is used for model training by default, and Perplexity decides for itself what it does with it. You should not process clients’ personal data in a personal account without a lawful basis and arrangements with the processor.
With an Enterprise account, Perplexity acts as a processor according to its privacy policy and does not train on your data. Your organisation still needs to carry out its own assessment.
Is Perplexity suitable for accountancy firms?
For accountancy firms, Perplexity is suitable for public questions, such as an explanation of laws and regulations with sources, but not for client data. In the Netherlands, for example, the duty of confidentiality under the rules of conduct of the NBA (the Dutch professional body for accountants) and the Verordening gedrags- en beroepsregels accountants (VGBA) also covers what an employee types into a search query. Client names and figures do not belong there.
Is Perplexity suitable for notaries?
For notaries, Perplexity is suitable for general preliminary legal research without any case file data. In the Netherlands, for example, a notary’s duty of confidentiality under Article 22 of the Dutch Notaries Act (Wet op het notarisambt) means that names, deed contents and other file information do not belong in a personal Perplexity account. The AI-weegschaal (AI weighing framework) published by the KNB, the Dutch association of civil-law notaries, in November 2025 is the starting point for that assessment.
Is Perplexity suitable for marketing agencies?
For marketing agencies, Perplexity is suitable for market and competitor research based on public sources. Customer lists, campaign data containing personal data and confidential briefs do not belong in a personal account. You should not process personal data there without a lawful basis and arrangements with the processor.
When should you check the sources of a Perplexity answer yourself?
Check the sources of a Perplexity answer yourself as soon as you carry a claim from it over into advice, a document or a decision. Only the source page shows whether the answer engine’s summary is correct. For figures, dates, statutory provisions and quotes, always open the source and read the passage the answer refers to.
A source from a government, a regulator or the original publication carries more weight than a blog summarising the same information. Since February 2025, Article 4 of the EU AI Act has required organisations that use AI to ensure sufficient AI literacy among their staff. In the AI Literacy Course, you learn to assess AI output with a set of standard checking questions and work out which data you should never enter into an AI tool.
