Perplexity Averaged 21.87 Citations Per Question in Qwairy's Q3 2025 Study. Most Brands Appear in None of Them.

In Qwairy's Q3 2025 study of 118,101 AI answers, Perplexity averaged 21.87 citations per question against 7.92 for ChatGPT, about 2.8x as many. If your brand does not appear in those citations, someone else's does.

How Perplexity Retrieves and Ranks Sources

Perplexity's own engineering write-up describes a retrieval pipeline built in stages. It queries its search index both lexically and semantically and merges the results into one hybrid candidate set, favouring completeness first; later stages filter and rank for precision (Perplexity Research, Architecting and Evaluating an AI-First Search API). It also parses documents into self-contained spans, each of which can be retrieved and ranked on its own at query time.

That write-up gives no sub-query or result counts. Vendor blogs that state exact counts describe their own reverse-engineering, not Perplexity's documentation, so this page no longer states them. What the primary supports is narrower and still useful: a page is retrieved and ranked in stages, and a page can be dropped at any of them.

The Citation Data That Rewrites Brand Visibility Strategy

Ahrefs' study of 75,000 brands is the finding that should change how you allocate budget: branded web mentions correlate with brand visibility in Google AI Overviews at 0.664 (Spearman), and backlinks at 0.218. Ahrefs studied AI Overviews, not Perplexity, and stresses that correlation is not causation.

That gap is large. In this data, the link-building playbook that worked for Google organic rankings tracks AI Overview brand visibility only moderately to weakly, while off-site brand signals track it most strongly. It is a reason to weigh brand presence at least as heavily as links, not proof that links stop mattering.

Domain Rating, Ahrefs' authority metric, correlates at 0.326, below brand anchors (0.527) and brand search volume (0.392). Ahrefs also found that brands in the top quartile for web mentions averaged 169 AI Overview mentions versus 14 for the next quartile down, a gap of more than 10x, while the bottom two quartiles averaged 0 to 3. That shows how concentrated AI Overview brand visibility is among the brands with the most web mentions. It does not show that adding mentions moves a brand between quartiles.

Why Most Brands Stay Out of AI Answers

Ahrefs' quartile data above already shows how concentrated AI Overview visibility is: the brands with the fewest web mentions barely register. Perplexity adds a second filter. It retrieves a broad candidate set, ranks it in stages, and cites only the few sources that survive. Being retrieved is not the same as being cited. Getting crawled is not the problem. Getting cited is.

Correction, November 8, 2026. An earlier version cited LumenGEO, a GEO software vendor, for Perplexity's daily query volume, for its own brand-audit and retrieval-to-citation figures, and for a re-ask churn rate from its own stability study. Those figures are removed. The monthly rotation and citation half-life in the FAQ now cite Profound and Scrunch, whose studies they come from.

Each Platform Cites Differently and That Is the Trap

Passionfruit's analysis of 11 million citations across ChatGPT, Perplexity, Claude, and Gemini exposed a problem most brands have not faced before. Perplexity includes sources in 95% of its responses, averaging 209 citations per query. ChatGPT includes sources in 72% at 86 citations. Claude includes sources in 64% at 275. Gemini sits at 77% and 82.

Source note, November 8, 2026. Passionfruit's study page now redirects to a product page, so these links point to the archived copy of July 15, 2026.

Only 35% of queries are cited by all four platforms. When all four answer the same question, the highest-citing platform includes 4.4x more sources than the one that cited the least. For 10% of shared queries, that gap reaches 12.7x.

Here is the part that matters most: 68% of queries that generated citations in one month did not generate them the following month. Only 7% maintained visibility for four months or longer. Less than 0.2% remained cited across all seven months studied. Citation stability is a myth. What you have this month you might lose next month unless the underlying source architecture holds.

What Tracks With Citations in Perplexity and Other Engines

Cyrus Shepard's Zyppy analysis of 54 studies, experiments, explainers and patents, covering ChatGPT, Gemini and Perplexity, scores 23 factors from 9.5 down to 2 on repeatability, strength of evidence and official support. The scores are his own weighting of the evidence, not effect sizes, and the factors are correlated with citations rather than proven causes. The top five: URL accessibility (9.5), search rank (9.4), fan-out rank across related sub-queries (9.3), preview control (9.2), and query-answer match (9.2).

Fan-out rank is the one most brands miss. When an engine searches related sub-questions as well as the typed query, you need to rank for the related questions too. If you own "best CRM for B2B SaaS" but are invisible on "CRM pricing comparison mid-market" and "CRM migration risks enterprise," an engine that fans out can filter you out before synthesis begins.

Content format shows up in the data. In BuzzStream's analysis of 4 million citations, press releases distributed through syndication channels such as Yahoo and MSN made up 0.04% of the dataset. Nearly invisible.

Freshness matters, but this page no longer gives a recency lift or a time-to-citation range: the vendor figures that stood here had no stated method. Scrunch's half-life finding in the FAQ below is the better-sourced evidence that citations decay without maintenance.

What Machine Relations Measures and Why Perplexity Proves the Model

Machine Relations is the discipline built around what AI engines actually retrieve when buyers ask questions. The 2026 data is consistent with its thesis.

Earned media is where AI engines find most of what they cite: Muck Rack's May 2026 What Is AI Reading? edition analyzed more than 25 million links from ChatGPT, Claude and Gemini responses across 17 industries and found that earned media, the sources brands neither own nor pay for, accounted for 84% of all AI citations, against 0.3% for paid and advertorial content. That pattern is not random. It is the result of citation architecture: the structural condition where a brand's claims appear as sources across AI-generated answers.

AuthorityTech measures this directly. When a brand comes to us, we do not count placements. We track whether those placements appear in AI-generated answers when buyers ask the queries the brand needs to own. That is the difference between a clip report and a citation report.

Perplexity makes the distinction unavoidable. A platform that averaged 21.87 citations per question in Qwairy's Q3 2025 data is not a niche research tool. Your brand is either in those citations or it is watching from outside the conversation while competitors get named.

FAQ

How many sources does Perplexity cite per response?

In Qwairy's Q3 2025 study of 118,101 AI answers across 8 providers, Perplexity averaged 21.87 citations per question, the highest of the providers in its table, ahead of Google AI Overview (17.93) and Gemini (17.11). ChatGPT averaged 7.92, which makes Perplexity about 2.8x higher, and Microsoft Copilot 2.47 on a small sample of 60 questions.

Do backlinks help brands get cited in AI search?

In Ahrefs' study of 75,000 brands, backlinks correlated with brand visibility in Google AI Overviews at only 0.218, compared to 0.664 for branded web mentions. Domain Rating correlated at 0.326. Ahrefs studied AI Overviews rather than Perplexity and stresses that correlation is not causation, so read it as a signal to build brand presence alongside links, not as proof that link-building does not matter.

How long do AI citations last?

Citation stability is low across all platforms. Profound compared the domains cited for identical prompts a month apart and found 40 to 60% of them had changed. Scrunch's citation half-life study, which tracked citation events from September 2025 to March 2026, found citation activity halves in roughly 4.5 weeks. 68% of queries that generated citations in one month did not generate them the following month.

What content formats earn the most AI citations?

BuzzStream's analysis of 4 million citations across 3,600 prompts and 10 industries (ChatGPT, Google AI Mode, Google AI Overviews and Gemini, collected over one week starting January 27, 2026) found blog/content pages made up 53.46% of citations, ahead of news (14.09%) and social (8.71%). Press releases distributed through syndication channels such as Yahoo and MSN made up 0.04% of the dataset. The measured unit is content-type share of citations in the cited dataset. It is an observed measurement; it does not establish a provider's source-selection mechanism, citation causation, a guaranteed future citation, recommendation lift, pipeline, revenue, or any other business outcome.

Correction

Correction, November 9, 2026. An earlier version stated figures from two GEO software vendors' blog posts as fact: audience shares for Perplexity users, a fixed number of sub-queries and results per answer, citation rates by content format, and a recency lift with a time-to-citation range. The vendors give no method we could check, and the pipeline description was their own reverse-engineering, not Perplexity's documentation. Those figures are removed and the retrieval section now follows Perplexity's own engineering write-up.