Study

Perplexity's Citation Funnel: 29 Query Variants, 10 Pages Retrieved, 3 Cited

Perplexity turns one question into 28.94 retrievable query variants, pulls back a median of 10 pages per searched answer, and gives a numbered citation to only 3 of those 10 pages, according to Foglift research covering 61,680 answers. GeoAura updated its Perplexity citation guide on November 5, 2026 with the four-step retrieval funnel measured end to end. Getting retrieved is only the start, and surviving the reranking step is where most pages disappear.

The Four-Step Funnel, End to End

The funnel starts with query rewriting, where one user question is expanded into 28.94 retrievable variants. That expansion is the engine of the whole process: Perplexity does not simply search the words a person typed, but fans out into many related phrasings to cover the question from different angles. For content creators, this means the exact wording of a query matters less than the breadth of the topic coverage. A page that answers only the narrow phrasing of a question may never match any of the variants, while a page that covers the topic thoroughly has far more chances to be picked up somewhere in the spread.

Retrieval then lists a median of 10 pages per searched answer, and reranking cuts that list down sharply. Only 3 of the 10 retrieved pages receive a numbered citation, a read-to-cite conversion of 30.0 percent, and generation prints those three citations above the prose. In other words, seven out of ten pages that Perplexity actually fetches still end up invisible to the user. The funnel is a filter at every stage, and the final step is the most brutal. Being one of the ten retrieved pages is an achievement, but it guarantees nothing about being cited.

Density Versus Coverage Across the Panel

The wider panel behind the research covers 61,680 answers and 372,111 citations across five engines, measured between March 20 and October 23, 2026, and it shows that the engines make very different trade-offs. ChatGPT attached at least one source to 93.9 percent of answers, ahead of Perplexity at 90.2 percent, so ChatGPT is slightly more likely to cite something at all. But Perplexity kept the density crown at 9.68 citations per answer, far ahead of ChatGPT at 3.29, with AI Overviews at 7.51, Gemini at 4.97, and Claude at 2.41. Perplexity cites a lot when it cites, yet the funnel shows how few of the pages it reads actually make the cut.

That contrast matters for anyone measuring AI visibility. A high citation count per answer does not mean a high chance for any single page, because the funnel is narrow at the end and the competition for those final slots is intense. An engine can attach many citations and still leave most retrieved pages unmentioned. For SEOs, the practical lesson is to track both sides of the equation: how often an engine cites at all, and how many of the pages it retrieves actually survive to be shown. Optimising for one without the other gives a misleading picture of how visible a site really is.

What the Volatility Means for Visibility Work

The research also underlines how unstable AI citations are from day to day. Consecutive-day domain overlap runs at just 52.7 percent, measured by LoudFace across 146 B2B questions and 2,044 answers between October 9 and 22, 2026. Nearly half of the domains cited on one day are different the next, which means a citation earned today is no guarantee of a citation tomorrow. For teams that report on AI visibility, that volatility is a warning against reading too much into a single day of results. A snapshot can look strong or weak for reasons that have nothing to do with the quality of the underlying content.

For link builders and content teams, the funnel points to a clear set of priorities. First, cover topics broadly enough to match the many query variants Perplexity generates, rather than optimising for a single phrasing. Second, make pages genuinely useful at the reranking stage, because retrieval alone is not enough when only three of ten pages get cited. And third, measure over time rather than chasing daily swings, since the overlap figures show that citations churn constantly. The funnel is measurable, which is good news, but it is also narrow and unstable, which means visibility in AI answers has to be earned again and again.

Sources

Sources: GeoAura, November 5, 2026; Foglift Research, October 24, 2026.

More from Linkbaits

Turn this playbook into links

Linkbaits helps you create content that earns citations and backlinks. Start with the free tools and build your first linkable asset this week.

Start free trial