Perplexity: a plain-English definition

Full explainer: What is Perplexity good for? covers this in more depth, with examples and limits.

What it means at work

In its API docs, Perplexity describes its Agent API as giving web-grounded answers with built-in citations in one call. For work, the appeal is speed on research questions: market checks, competitor summaries, and background on a company or topic, with links you can open.

What the documentation says

Perplexity separates finding sources from writing an answer, which is useful to know when you judge a result:

• Search step. The Perplexity Search API returns ranked web results as JSON with extracted page text. Perplexity says the Search API gives no generated answer and is billed per request.
• Answer step. The Perplexity Search API guide says to use the Agent API for a model-written answer with citations.
• A model does the writing. In the search-only setup, the Perplexity guide says your model reads the text and writes the answer. That writing step is where a summary can drift from its source, so open the links.

An example

You ask for the latest public information on three competitors’ pricing pages and product launches. Perplexity returns a summary with links. You open each one, confirm the details and paste verified points into your brief.

A common misconception

Citations are not a guarantee. An answer engine can summarize a source inaccurately or rely on a weak page; Liu et al. audited four generative search engines and found only 74.5% of citations support their associated sentence. The same audit found that on average a mere 51.5% of generated sentences were fully supported by citations. The audit dates from 2023, so current tools may score differently. The habit that matters is opening the links for anything important. Also check what your company allows before pasting internal material into any external tool.

Sources

• Perplexity API Docs, Overview: “Get web-grounded answers with built-in citations in one call”
• Perplexity API Docs, Overview: “Get ranked web results as JSON with extracted page text”
• Perplexity API Docs, Overview: “No generated answer … billed per request”
• Perplexity API Docs, Search API: “For an LLM-generated answer with citations … use the Agent API”
• Perplexity API Docs, Search API: “Your model reads the text and writes the answer”
• Liu et al., “Evaluating Verifiability in Generative Search Engines” (arXiv): “only 74.5% of citations support their associated sentence”
• Liu et al., “Evaluating Verifiability in Generative Search Engines” (arXiv): “on average … a mere 51.5% of generated sentences are fully supported by citations”