What is Perplexity and how does the AI answer engine work?

An answer engine, not a chatbot or search engine

Perplexity AI was founded in 2022 (co-founders include Aravind Srinivas) and positions itself as an "answer engine". The idea: ask a question in natural language and get a direct, sourced answer rather than ten blue links. It combines two building blocks that used to be separate: a web search and a large language model (LLM) that reads and summarises the retrieved content.

The difference from a plain chatbot is the grounding in current web sources with visible footnotes; the difference from classic search is the composed synthesis. Perplexity thus fills a gap between ChatGPT (open-ended conversation) and Google (list of links).

How Perplexity works: search, RAG and sources

Technically, Perplexity uses a pattern called Retrieval-Augmented Generation (RAG): it first turns the question into search queries and searches the web, then the language model reads the retrieved pages and composes an answer, linking each key claim to a numbered source. Because the model is grounded in freshly retrieved content, it can give more current and more verifiable answers than an LLM relying only on its training knowledge.

• Models: Perplexity runs its own Sonar models and lets paying users choose among leading third-party models (e.g. from OpenAI, Anthropic, Google) - the exact line-up changes over time.
• Sources: answers include numbered footnotes and a source bar, so each statement can be traced back.
• Follow-ups: as a conversational service, Perplexity keeps context so you can refine with follow-up questions.

Key features at a glance

• Pro Search: multi-step research that breaks a question into sub-steps and consults more sources than the standard search.
• Deep Research: longer automated investigations compiled into a structured, multi-source report.
• Focus: restrict a query to specific domains such as academic papers, the web or social platforms.
• Spaces: shared workspaces to bundle your own files and context.
• Comet: an AI browser from Perplexity that integrates research directly into browsing.
• Other outputs: image upload, file analysis and a developer API - availability depends on plan and region.

Perplexity vs ChatGPT vs classic search

• Perplexity: prose answer with visible sources and live web search - strong for cited, current research and source discovery.
• ChatGPT & co.: stronger at open text generation, brainstorming and tasks; web search and citations exist depending on mode but are less central.
• Classic search (Google, Bing): returns a list of links, increasingly with AI summaries like Google AI Overviews - full control over original sources but no ready-made synthesis.
• Rule of thumb: Perplexity for "answer and cite", chatbots for "create and phrase", classic search for "show me all the originals".

Strengths and limits

The main strength is verifiability: because sources are visible, an answer is faster to check than with a source-less chatbot. Even so, care is still required - citations do not fully solve the hallucination problem.

• Source quality varies: Perplexity may cite solid studies as readily as weak or outdated pages - selection is up to the model.
• Misattribution possible: a statement may diverge from the linked source; a click to verify remains important.
• No substitute for primary sources: for legal, medical or financial decisions, the linked original is authoritative, not the summary.
• Freshness and coverage: content that is poorly indexed or behind paywalls may be missing or skewed.

Swiss and DACH perspective: data protection and responsible use

Perplexity is a US provider. For professional use in Switzerland, mind the revised Data Protection Act (revDSG/nFADP) and avoid entering personal data or confidential business information into prompts without checking the provider's terms and data processing. The supervisory authority is the EDOEB (FDPIC); for offerings with an EU nexus the EU AI Act also applies extraterritorially and requires, among other things, transparency about AI systems.

For particularly sensitive or sovereignty-relevant use cases, it is worth looking at European and Swiss alternatives as well as open models such as Apertus (ETH Zurich/EPFL). Principle: use Perplexity as a research assistant, verify claims against the linked sources, and for pricing always consult the provider's current pricing page.