{"id":2709,"date":"2026-08-20T12:41:07","date_gmt":"2026-08-20T10:41:07","guid":{"rendered":"https:\/\/mairateam.com\/?post_type=glossary&p=2709"},"modified":"2026-08-20T15:55:46","modified_gmt":"2026-08-20T13:55:46","slug":"perplexity","status":"publish","type":"glossary","link":"https:\/\/mairateam.com\/en\/glossary\/perplexity\/","title":{"rendered":"Perplexity"},"content":{"rendered":"

Perplexity<\/strong> is a search-first AI assistant: it answers a query with a single coherent response assembled from current web sources and cites those sources visibly next to each part of the answer. From a brand's point of view it is the most transparent of the large assistants: you can see directly whom it cites, which makes it the easiest place to test citation visibility.<\/p>\n

In short<\/h2>\n\n\n\n\n\n\n
What it is<\/strong><\/td>\nAn AI assistant built on search, assembling answers from live web sources<\/td>\n<\/tr>\n
What makes it different<\/strong><\/td>\nCitations are visible and numbered next to each claim, not hidden in a footer<\/td>\n<\/tr>\n
How to get in<\/strong><\/td>\nPerplexityBot has to reach the site and the content has to offer quotable answers<\/td>\n<\/tr>\n
What to expect<\/strong><\/td>\nVisibility and mentions, not traffic. Clicks from citations come in single digits<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

How Perplexity picks its sources<\/h2>\n

From query to candidate set<\/h3>\n

Perplexity first processes the query: it adds context from earlier messages in the conversation, splits it into partial searches and runs them in parallel. Candidates come from the search index that PerplexityBot fills; a page missing from the index or stale in it can be fetched on the spot via Perplexity-User. Candidates are not treated as whole pages: the text is split into passages and the passage is scored, not the domain. The unit of competition is therefore a paragraph or table that answers a partial question on its own.<\/p>\n

Ranking and synthesis with numbered citations<\/h3>\n
    \n
  1. Candidate passages are ranked by match to the query, freshness and source signals: how often and in what context the domain appears on the topic. Only a handful of top-ranked sources make it into the answer.<\/li>\n
  2. The selected passages go to the language model as context; it writes the answer from them alone (the RAG mechanism) and attaches to each claim the number of its source.<\/li>\n
  3. A source the model took no claim from appears next to no sentence, even if it reached the source list. A numbered citation therefore proves a passage was used, not just listed.<\/li>\n
  4. Deeper search modes run several rounds: after the first, the model asks follow-up queries and widens the candidate set, so sources the first search missed can still enter the answer.<\/li>\n<\/ol>\n

    What this means for the site<\/h3>\n

    Readability.<\/strong> PerplexityBot as a rule does not run JavaScript. Prices, specifications and reviews added only by the browser are not in the index, so a passage containing them cannot be cited. A passage has to stand on its own: the first sentence carries the answer, numbers have a unit and a date, the product or brand name sits inside the passage, not only in the page heading. Freshness.<\/strong> Ranking works with dates: for queries about prices, availability and news, a page with a visible update date and data valid as of that date outranks an older page, however long that page is. Structure.<\/strong> A heading phrased as a question, short paragraphs, specifications in a table, a definition or summary up front. For an e-shop: a category intro answering which product to choose, and a product page with specifications in the HTML. For a B2B site: a service page with scope and price range, and an article with original numbers.<\/p>\n

    Worked example:<\/strong> a product page loads its specifications from the inventory system only through a browser script. PerplexityBot stores the HTML without them, so for the query \"X with specification Y\" it has no passage to cite even though the product qualifies. The same specifications in a server-rendered HTML table create a citable passage.<\/p>\n

    Why Perplexity is useful for brands specifically<\/h2>\n

    With most assistants you can only guess what an answer is based on. Perplexity shows its sources, so it works as a window into how models choose: which domains recur on a topic, which content formats get cited, and in what wording. A regular test of your own queries on Perplexity is the cheapest way to track your citation share and learn what wins citations.<\/p>\n

    From our own practice<\/h2>\n

    Availability for PerplexityBot is verified in every audit with a direct request: the expected result is HTTP 200, anything else means a block somewhere on the path. At the same time, the sober part of the same picture: traffic from perplexity.ai tends to be zero or a handful of sessions even with good visibility. Perplexity is a channel for visibility and learning, not for traffic, and that is exactly how it should be judged.<\/p>\n

    Common mistakes<\/h2>\n