{ "@context": "https://schema.org", "@type": "Article", "schema_version": "1.1", "content_item_id": "perplexity-fast-search-retrieval-audit.en", "translation_group_id": "perplexity-fast-search-retrieval-audit", "locale": "en", "type": "guide", "section": "knowledge", "slug": "perplexity-fast-search-retrieval-audit", "title": "How to audit Fast Search: returned pages, fact use, and visible links are different events", "description": "A practical Perplexity Search API protocol for separating returned results and their order from factual use in an answer and a source link shown to the reader.", "direct_answer": "A Fast Search result shows what the application programming interface returned and in what order. It does not prove that the system used every result, carried a fact into the answer, or showed a source link to a reader. Audit four stages separately: request and parameters -> results and order -> answer claims and textual support -> shown link and later action. Keep a numerator and a base count for each stage; when a step cannot be observed, record missing evidence rather than zero.", "sections": [ { "heading": "1. Direct answer: a search result is not yet an answer", "paragraphs": [ "Fast Search answers a narrow technical request: which web pages the application programming interface returned under specified conditions. It is not a log of everything that happened after the search. A page may be passed to another component, dropped, used in part, or omitted from the links shown to the reader. Therefore, “the page was in the result set” is a separate observation from “the system linked to the page.”", "The practical boundary is simple: record the returned documents and their order first, then check the answer text, the support for each claim, and the link that the reader could see. This sequence does not reveal Perplexity's hidden formula, but it prevents later events from being attributed to the search step." ], "bullets": [ "Return: the document appears in the search call's result set.", "Fact use: answer text is supported by a checkable passage.", "Shown link: the reader can see the source address beside the answer.", "A brand mention, a visit, and a desired action are separate events." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "2. What Fast Search actually promises", "paragraphs": [ "The official documentation describes Fast Search as a lower-latency, lower-cost mode of the Search API. A call sets `search_type: \"fast\"`; when the parameter is omitted, standard web search is used. On the review date, the documentation states a price of $1 per 1,000 Fast Search requests instead of $5 per 1,000 standard web search requests.", "Perplexity recommends the fast mode for day-to-day tool calls inside automated processes, high-volume work, and time-sensitive applications. It recommends the default `search_type: \"web\"` mode for rare, difficult, or ambiguous requests. Results use the same response format, and `max_results` accepts values from 1 to 20. This is a testable interface contract, not a promise of a particular order or a source link in a later answer." ], "bullets": [ "$1 per 1,000 Fast Search requests.", "$5 per 1,000 standard web search requests.", "Between 1 and 20 results in one Fast Search call." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "3. The four audit stages", "paragraphs": [ "Stage one records the exact prompt, date, language, country, chosen mode, and all call parameters. Stage two records the complete result set: page address, title, position, response state, and errors. Stage three splits the generated answer into checkable claims and opens the returned pages to determine whether each claim is supported. Stage four records the links actually shown to the reader and any available evidence of visits.", "Each stage needs its own unit. For a page-return rate, the numerator is results with the chosen property and the base is every result in the predefined series. For support, count supported claims over all checked claims. For shown links, count claims with a visible source link over all answer claims. If a step cannot be observed, record missing evidence rather than a measured zero." ], "bullets": [ "One log row represents one interface call.", "Page return, fact use, shown link, mention, and visit are separate fields.", "An error, an empty result, and an answer without a link are separate outcomes." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "4. How to compare the modes fairly", "paragraphs": [ "Compare `fast` and `web` only with the same prompt set, language, country, time window, and counting rules. Change one variable at a time. Choose the number of repeated runs, the domain list, and the rule for supporting a claim before collection begins; otherwise convenient examples can reshape the conclusion after the evidence is gathered.", "Order can be described only as the order observed in the returned list. One run cannot reconstruct an internal formula or prove that a higher position caused a shown link. If the model, interface, or parameters change, start a new series and state the date through which evidence was collected. Keep the original responses apart from the editorial explanation; the short Telegram post is a navigation link, not evidence." ], "bullets": [ "One prompt set and one time window.", "One changed setting per series.", "Repeated runs and counting rules chosen in advance." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "5. Worked calculation without mixing metrics", "paragraphs": [ "Suppose 20 Fast Search calls return 10 results each: the base for the first calculation is 200 results. If 30 results contain a page from the selected domain, the observed return rate is 30/200. That does not mean 30 source links were shown. If a later answer layer creates 40 claims and opened pages support 12 of them, support is 12/40. If only 7 claims carry a shown link, the shown-link rate is 7/40.", "Every value in this example is invented to explain the calculation, not to measure Perplexity. Brand mentions, visits, and desired actions require separate logs and their own bases. Even 7/40 does not prove that the search result caused a business outcome. Label every fraction in words so a reader can see exactly what is counted above and below the line." ], "bullets": [ "30/200 — pages from the domain among returned results.", "12/40 — claims supported by opened text.", "7/40 — claims with a shown source link.", "This is a teaching example, not an industry measure." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "6. Limits and alternative explanations", "paragraphs": [ "Fast Search may be only the first layer in a longer system. A result may be transformed by another model, filtered by a safety rule, supplied by a cache, or combined with other sources. A missing link can reflect interface formatting, while a matching fact can come from several sources or from knowledge already held by the model.", "Public documentation confirms product parameters and format but does not expose every internal decision. Limit findings to the recorded series, date, and log. Do not carry them automatically to other languages, models, applications, or periods. A relationship between page order and a shown link may be described as an observed association, not as proven cause." ], "bullets": [ "Mark an inaccessible internal step as missing evidence.", "Do not turn an order-link association into causality.", "Repeat the audit after documentation, model, or response-format changes." ], "source_ids": [ "perplexity-fast-search-docs" ] }, { "heading": "7. Practical checklist and deliverable", "paragraphs": [ "Before publishing a conclusion, check seven points: the full prompt and parameters were saved; every result and error was recorded; the answer was split into checkable claims; sources were opened; support for each claim was tested; shown links were recorded separately; and the Russian and English versions match in facts, numbers, dates, and conclusion strength.", "The deliverable should contain one table at call level and another at claim level. The first answers which pages the search returned and in what order. The second identifies the open passage that supports a claim and whether a source link was shown. This output can be checked again, paired with a visible-link audit, and updated after a product change without blending search, answer text, and reader action." ], "bullets": [ "Responsible desk: GeoAeoAle Editorial.", "Machine assistance is disclosed; manual review is not claimed.", "The publication date matches the Telegram short version: 7 October 2026." ], "source_ids": [ "perplexity-fast-search-docs" ] } ], "published_at": "2026-10-07", "modified_at": "2026-10-08", "data_through": "2026-10-08", "next_review_at": "2026-11-08", "author": "GeoAeoAle Editorial", "origin": "editorial", "topics": [ "perplexity", "search-api", "retrieval", "citation", "measurement" ], "publisher": "GeoAeoAle Editorial", "license": "https://creativecommons.org/licenses/by/4.0/", "canonical_url": "https://geoaeoale.com/en/knowledge/perplexity-fast-search-retrieval-audit/", "claims": [ { "claim_id": "search-results-are-not-answer-use", "text": "A Search API returns search results. That establishes the result set and order only; use of the content in a later answer must be checked separately.", "status": "inference", "confidence": "high", "source_ids": [ "perplexity-fast-search-docs" ], "publication_status": "public", "confidentiality": "public" }, { "claim_id": "fast-search-contract", "text": "Perplexity documents `search_type: \"fast\"`, a price of $1 per 1,000 requests instead of $5 for standard web search, and recommends the mode for day-to-day and time-sensitive work.", "status": "observed", "confidence": "high", "source_ids": [ "perplexity-fast-search-docs" ], "publication_status": "public", "confidentiality": "public" }, { "claim_id": "same-response-format", "text": "Perplexity states that Fast Search uses the same response format as standard web search and accepts `max_results` values from 1 to 20.", "status": "observed", "confidence": "high", "source_ids": [ "perplexity-fast-search-docs" ], "publication_status": "public", "confidentiality": "public" }, { "claim_id": "four-stage-method", "text": "A reproducible audit benefits from separating the call and parameters, results and order, claim support, and a shown source link or user action.", "status": "inference", "confidence": "high", "source_ids": [ "perplexity-fast-search-docs" ], "publication_status": "public", "confidentiality": "public" }, { "claim_id": "no-causal-ranking-proof", "text": "A page returned by Fast Search does not prove that publishing it caused its order, a brand mention, a shown link, a visit, or a desired action.", "status": "inference", "confidence": "high", "source_ids": [ "perplexity-fast-search-docs" ], "publication_status": "public", "confidentiality": "public" } ], "sources": [ { "source_id": "perplexity-fast-search-docs", "canonical_url": "https://docs.perplexity.ai/docs/search/fast-search", "title": "Fast Search", "publisher": "Perplexity API documentation", "source_type": "official", "locale": "en", "published_at": null, "checked_at": "2026-10-08", "sha256": "462cc92931f49d0b8cd040e52d269fe69508390dc842850fc50b400e7d9fb5ab", "license": "Proprietary public documentation; factual reference only", "visibility": "public" } ], "related_slugs": [ "visible-links-search-record", "source-exposure-brand-mention-diagnostic", "ai-citation", "measuring-ai-visibility" ], "distribution_links": [ { "kind": "telegram", "url": "https://t.me/geoaeoale/31", "label": "Short version on Telegram" } ], "limitations": [ "Public documentation does not expose the complete path from a returned page to final text.", "Official documentation can change; recheck after an interface or model change.", "Visits and desired actions are not measured by this protocol.", "The Telegram link opens a short version and is not evidence for a factual claim." ], "corrections": [] }