Perplexity tells its own users to check every citation

The writer's problem

A writer using an AI search assistant for background research needs a way to tell which sentence in an answer came from which web page, rather than accepting an unsourced paragraph on faith. Perplexity markets itself around exactly that distinction, and its own help hub and getting-started guide, as retrieved on 16 September 2026, describe how the citations work and what the user is expected to do with them.

What the documents show

The help hub describes Perplexity as the world's first answer engine, one that searches the internet in real time to deliver fast, clear answers to any question, with sources and citations included, a description of the product's own positioning rather than an independent comparison against other search tools. The getting-started guide is more specific about what a user should do with those citations, instructing under the heading Check every citation that every Perplexity answer links to numbered citations: hover to preview it, or click through to read the full source, and that trust labels like Government, Academic, and Trusted let you judge credibility at a glance. That instruction appears in the guide's section on Perplexity's Computer agent feature, though the numbered-citation mechanism it describes is the same one Perplexity uses across its core search product. Neither document states how a citation is selected from among the pages a search returns, nor what happens when a page a citation points to is later removed or changed.

The editorial choice

This is an editorial habit worth adopting rather than a step the software performs for the writer: the guide's own instruction to check every citation concedes that a numbered link is a pointer to be verified, not a verified fact by itself, and a trust label describing a source as Academic or Government describes the type of publisher, not whether that particular page states what the answer claims it states.

What stays with the author

Perplexity's documents describe a mechanism for pointing at sources; they do not describe a mechanism for confirming that the surrounding generated sentence accurately represents what a cited page says. Opening the numbered link and reading the passage before repeating a claim in a manuscript remains the writer's task, exactly as the getting-started guide itself instructs.

• Has every numbered citation behind a research claim actually been opened and read, not just noted as present?
• Does a Trusted or Academic label describe the specific page cited, or only the domain it came from?
• Would the cited page still say the same thing if reopened today, or has it since been edited or removed?

An answer engine that numbers its sources is easier to check than one that does not, which is a design advantage; it is not the same as the checking having already been done.