What is Perplexity AI?

How does Perplexity AI work?

Perplexity is primarily designed for chatbot-based web search. The tool's architecture enables it to interpret user queries, search the internet, analyze user-uploaded content and generate responses.

Perplexity delivers results using a variety of large language models (LLMs), both proprietary and from third parties. This LLM-based structure creates a conversational feel for users interacting with the tool thanks to these models' powerful natural language processing (NLP) capabilities.

Perplexity's default model, used as the foundation model for its standard search, is an in-house modification of Meta's Llama 3 called Sonar. But Perplexity also offers access to several other LLMs for more advanced searches and multimodal capabilities, including the following:

• OpenAI's GPT-4o and o3-mini.
• Anthropic's Claude 3.5 Sonnet.
• Perplexity's Sonar Large, which offers advanced real-time search capabilities.
• DeepSeek's R1.

Pro and Enterprise Pro users can toggle among options to find the best model for their query.

Other design elements include web crawling, indexing and information retrieval. Web crawling techniques enable Perplexity to continuously update its knowledge base, while information retrieval techniques -- such as semantic similarity and keyword matching -- help the tool extract information from indexed web data.

Search options

Perplexity divides its search capabilities into Auto and enhanced queries.

Auto, which is available for all Perplexity pricing tiers, uses the Sonar model to perform basic functionalities, such as answering questions, providing summaries, suggesting and responding to follow-up questions and citing sources. Perplexity describes this mode as best suited for straightforward, everyday queries.

Enhanced queries offer more complex search and generation capabilities. The four types of enhanced searches are Pro Search, Deep Research, Reasoning with R1, and Reasoning with o3-mini:

• Pro Search. Pro Search includes all the features of Auto plus a wider range of sources, more detailed answers, more model options, multimodality and the Internal Knowledge Search feature.
• Deep Research. Deep Research capabilities include iterative search, reasoning and comprehensive reports on queries. It uses a variant of DeepSeek R1.
• Reasoning with R1. This mode uses Perplexity's R1 1776, an open source, multilingual search model. It is a modification of DeepSeek R1 that removes its censorship features.
• Reasoning with o3-mini. This mode uses OpenAI's o3-mini, a lightweight reasoning model, to perform tasks that require logic and problem-solving.

Perplexity AI's key features

Perplexity offers a range of capabilities, from simple question answering to more complex research. Key features include the following: 

• Summaries. When a user queries Perplexity, the tool generates a response summarizing its key findings. Summaries range in detail and complexity depending on a user's search type.
• Follow-up questions. Perplexity suggests follow-up queries to help users refine their searches.
• Categorical search. Perplexity has four tabs users can use to narrow a search query. By choosing either web, academic, social or organization files, users can categorize their search to help Perplexity find a more targeted answer.
• Source verification. Perplexity search results include detailed links to verify the sources used to generate answers. This helps users fact-check the tool's answers.
• Multimodality. Perplexity offers multimodal abilities, enabling users to work with text, images, code and other media types.
• File uploads. Users can attach files via the search bar to provide additional information and context for queries.
• Internal Knowledge Search. Perplexity can search web information as well as internal, proprietary documents with the Internal Knowledge Search feature. Repositories can hold up to 500 proprietary files for Perplexity analysis in XLSX, PPTX, DOCX, PDF or CSV formats.
• Push notifications. Users can create custom push notifications for Perplexity, rivaling Google Alerts and other similar tools.
• Real-time answers. Perplexity offers real-time information, enabling users to query the tool for live updates. For instance, a user could ask the tool for the score of a live football game. However, users note that this feature sometimes lags and does not always provide the most up-to-date information.
• News discovery. Perplexity's Discover tab provides users with a news page, similar to those found on traditional search engines like Google.
• Query management. Perplexity's Library tab saves all of a user's past queries, called threads. The Spaces feature lets users organize those threads into specific folders and projects. Users can also toggle privacy settings and invite contributors to collaborate on projects.
• Customizable outputs. Perplexity's Spaces feature also lets users personalize how they want the tool to answer queries. For example, a user conducting research could instruct the model only to use sources from a given year. Similarly, a student practicing French could instruct the tool to only give output in the French language.

Perplexity AI limitations

Like many generative AI tools, Perplexity suffers from certain limitations that users must consider. Common challenges include the following:

• AI bias. Also known as machine learning bias, AI bias occurs when an AI model trains on biased or skewed data. AI bias can lead to stereotyped and inaccurate responses or elicit outputs that are otherwise damaging to marginalized and underrepresented communities.
• Hallucinations. AI models hallucinate when they produce false output. This can be especially problematic for AI search tools, where users want factual answers to questions as quickly as possible.
• Limited problem-solving abilities. AI search tools like Perplexity have limited problem-solving and reasoning abilities. Perplexity AI might highlight irrelevant sources, make false assumptions about data or create misleading generalizations.
• Copyright infringement. Like many other generative AI tools, Perplexity faces copyright infringement lawsuits, which claim that the search engine illegally copies and regurgitates copyrighted articles. The possibility of infringing on intellectual property laws is concerning for users or organizations looking to use Perplexity AI for content generation.
• Reliance on third-party LLMs. Perplexity offers services tied to multiple third-party LLMs, meaning that Perplexity is subject to the same limitations as those LLMs.

The potential for bias, hallucination or inaccurate reasoning means that users querying Perplexity must stay vigilant and investigate answers for accuracy and trustworthiness.

7 example use cases for Perplexity AI

Due to its search and generative capabilities, Perplexity is useful in a wide range of scenarios. The following are a few examples.

1. Answer questions

From the most basic inquiries to questions that involve multiple lines of reasoning, users can query Perplexity in a conversational chatbot setting.

For example, a user could ask the tool, "Who is Selena Gomez?" Within seconds, Perplexity will provide an answer in several paragraphs, explaining who Selena Gomez is and what she is known for, including subsections for her early life, music career and recent ventures. The tool will also link top sources for users to click on and suggest follow-up questions, such as "What are some of Selena Gomez's notable roles in movies and TV shows?"

2. Conduct research

Users can conduct research using Perplexity. For example, a student could use Perplexity to study by querying the model and asking follow-up questions to better understand a complex topic.

Perplexity's enhanced features, such as deep research, provide in-depth reports for information discovery. Perplexity also lets users upload their own files and organize past queries into folders, aiding the research process.

3. Summarize text

Perplexity's NLP capabilities make it helpful for summarizing lengthy or complex documents. For example, a paralegal could ask Perplexity to summarize legal documents and case law. This is especially useful when combined with the file upload, which lets users ask the tool to summarize documents they upload that might not be available online.

4. Translate language

Perplexity can translate text from one language to another. For instance, a user could ask the model to translate an English phrase into French. Perplexity will return the translated phrase or sentences, plus a breakdown explaining what different words mean in the translated language.

5. Create content

Perplexity's generative capabilities enable users to create new content. For example, Perplexity can generate creative writing, images, music compositions, code, letters, professional correspondence and more.

The tool's multimodality means that users can access various media types when asking the model to create content. For instance, a user could upload an image of a famous artwork and ask Perplexity to create song lyrics inspired by the art.

6. Research market trends

Users can query Perplexity for help with different kinds of market research. For example, Perplexity can check real estate trends in a given area, plan a cost-effective trip to a particular destination, analyze stock market trends or conduct product price comparisons. Perplexity's push notifications feature is also helpful in monitoring specific market research based on a user's interests and needs.

Perplexity also offers a price comparison tool called Buy with Pro for U.S.-based users, which assists with the buying experience from initial product search through payment.

7. Analyze data

Through file upload or the Internal Knowledge Search feature, users can instruct Perplexity to engage with proprietary and web data together. For example, a small business owner could upload past sales material for Perplexity to analyze alongside current market trends to develop unique sales proposals for the company.

Getting started with Perplexity AI

Perplexity offers Standard, Pro and Enterprise Pro plans. Each tier offers different pricing and features.

This list represents product information at press time; be sure to check pricing documentation for a comprehensive and up-to-date list.

Perplexity's Enterprise Pro tier offers Pro features and additional capabilities, including increased data privacy and security, user management, SOC 2 certification, single sign-on and data retention. The Enterprise Pro plan offers a self-serve cost of $40 per month, per seat, or $400 per year, per seat, for companies with fewer than 250 employees. Larger enterprises can contact the Perplexity team for custom pricing.