Perplexity AI is a search tool that behaves less like a list of links and more like a research assistant. Ask it a question and it returns a written answer in plain language, with numbered citations pointing to the web pages it drew from. For students, writers, analysts and anyone who spends time gathering and checking information, that combination of speed and traceability is the product’s main appeal.
The company describes Perplexity as an “answer engine.” It does not aim to replace a conventional search engine for every task, such as finding a specific website, but to handle the questions where a user wants a synthesized explanation and the ability to verify it.
How it works
When a query comes in, Perplexity runs web searches, reads the top results and uses a large language model to compose a response. Each claim in the answer carries a footnote that links back to its source, so the reader can check the original page. The system also suggests related follow-up questions.
Conversation is the second key feature. Perplexity keeps the context of the exchange, so a user can ask “how do neural networks work?” and then follow up with “and how are they trained?” without restating the subject. Users can also narrow a search to particular source types, such as academic papers or discussion forums, and Pro users can upload documents for analysis.
Free and Pro
The basic service is free, with limits on the number of advanced searches per day. Perplexity Pro costs $20 a month, or less on an annual plan, and provides a larger allowance of advanced searches, file uploads, image generation and access to a choice of frontier models from OpenAI, Anthropic and others alongside Perplexity’s own Sonar models. In early 2025 the company also added a Deep Research mode that runs dozens of searches and produces a longer report on a topic.
Model lineups and limits change frequently, so the current details are best checked on the company’s pricing page before subscribing.
Strengths and cautions
Perplexity’s citations make it easier to audit than a chatbot that answers from memory, and it works reasonably well in languages other than English for straightforward questions. But a citation is not a guarantee: the model can misread a source or combine sources in a misleading way, and the quality of the answer depends on the quality of the pages it finds. Users should treat the answer as a starting point and read the cited pages for anything that matters.
The company has also faced criticism and legal complaints from publishers over how it summarizes their articles, a dispute that remains unresolved and that is shaping how AI search tools handle attribution and licensing.
For research-heavy work, Perplexity’s model of answer-plus-sources has become influential: Google, OpenAI and Microsoft have all added similar cited-answer features to their own products over the past year.
