Reddit accounts for about 46.7% of retrieved sources in Perplexity queries, according to one third-party audit. However, most of those Reddit pages never make it into an inline citation. They are fetched, scanned, and quietly dropped. YouTube, conversely, is heavily cited for how-to and product queries. Same retrieval pipeline, opposite citation outcomes.
This observation should reframe how B2B SaaS teams think about AI search visibility. The question isn't whether Perplexity can find your content; it's whether it will cite it.
What the stream actually shows
Suganthan Mohanadasan, co-founder at Snippet Digital, published a detailed teardown of Perplexity's raw answer stream in late June 2026, captured from a Pro account and re-verified a month later. His method involved intercepting the server-sent event stream before it disappears (Perplexity's streaming data isn't replayable once the answer finishes rendering). The structural findings are significant, even with the caveat that they come from a single user's captures across eight queries.
The core architecture he found: a 16-head classifier that routes every query based on intent, with fixed thresholds for content types like maps, video, image, and finance. Each query receives a probability scorecard. Depending on the scores, Perplexity decides what results to surface and how deeply to search.
Two things stand out. First, Perplexity always runs a web search. Unlike ChatGPT, which sometimes answers from parametric memory, Perplexity hits the web on every query. Second, the default fan-out is shallow, typically running a single search using the exact query phrasing. Deeper multi-round searches only occur for local queries or when the system escalates into what Mohanadasan calls "Study mode."
Trust is scoped, not scored
This is where most SEO-adjacent takes get it wrong. Perplexity doesn't assign a generic authority score to domains. Instead, domains carry what the stream data labels as "trust notes" with two tiers: "credible" and "trusted." Each trust entry includes a scope description specifying what that domain is credible for. A SaaS company's domain might be trusted for its product information but not for general industry benchmarks.
This distinction matters operationally. Your product pages, comparison pages, and help docs are the content types most likely to earn scoped trust. Generic thought leadership with no first-party data? Probably not.
Content published in the last 30 days receives roughly 3.2× more citations than older content, per the same third-party analysis. Freshness isn't just a tiebreaker; it appears to be a primary signal. Multi-constraint queries average about 17.7 citations each, indicating that Perplexity pulls from a wide range of sources when the question is complex enough.
Retrieved versus cited: where the real gap lives
The practical distinction that should change your content operations: a page can be retrieved by Perplexity without ever receiving an inline citation in the final answer. Retrieved means Perplexity fetched and read the page; cited means it linked to it in the visible response. The gap between those two outcomes is where most B2B SaaS content currently sits, and it's where the optimization opportunity is sharpest.
What gets cited? Based on observed patterns, Perplexity favors answer-ready formats: procedural steps, numerical data, concise definitions, structured tables, and FAQs. Content that's easy to extract a discrete fact from is more likely to earn a citation. If your page requires three paragraphs of context before delivering the answer, it's less likely to earn the citation even if retrieved.
Deep Research mode behaves differently. It reads full pages, takes longer to process, and favors the most comprehensive source available. So the optimization splits: for standard queries, be concise and extractable; for Deep Research queries, be the most thorough page on the topic.
What this means for your content operation
Three moves worth implementing this quarter. First, audit your existing pages for extractability. Add structured elements (tables, definition blocks, numbered steps) to high-value pages. Second, build a freshness workflow. If content published in the last 30 days gets 3.2× the citation rate, a monthly update cadence on your top pages is essential. Third, start tracking citation presence as a distinct metric from traffic and rankings. A page that gets retrieved but never cited presents a different problem than one that doesn't get retrieved at all.
The strongest quantitative claims here come from third-party reverse-engineering, not official Perplexity documentation. Treat them as observed patterns, not universal constants. Perplexity's recent product updates describe expanded research depth and agentic workflow improvements but don't publicly specify ranking weights.
Still, the directional signal is clear enough to act on. Perplexity behaves more like a traditional search engine than most marketers assume, with one critical difference: it doesn't just rank your page; it decides whether to name you in the answer. The gap between being found and being cited is the new gap between visibility and credit. Most B2B content operations haven't even started measuring it.
