Perplexity cites the pages it retrieves at answer time — so getting cited comes down to being retrievable for the question, being quotable once retrieved, and being corroborated by the other sources it pulls alongside you. Unlike training-data recall, citations respond to changes you make this month, which makes Perplexity (and ChatGPT search, and Gemini with grounding) the fastest-moving surface in GEO.

Why citations matter beyond the click

A citation is more than a traffic source. When an assistant cites your page, your content is shaping the answer itself — often verbatim. And the citation lists across many answers reveal the source graph of your category: which directories, review sites, and publications the engines actually trust. That graph is your outreach roadmap.

In geopsy's model this is the source rank signal — distinct from being named in the answer text (text rank), and worth tracking separately, because you can be cited without being named and named without being cited.

What retrieval-based engines select for

Grounded engines like Perplexity work in two stages: retrieve candidate pages with web-search queries, then generate an answer conditioned on them (RAG). Each stage has its own bar:

To survive retrieval, a page needs to:

Be crawlable without JavaScript — AI crawlers (PerplexityBot included) don't render client-side apps.
Match the question's intent in its title, headings, and opening copy — retrieval queries look a lot like the user's question.
Be indexed and fast; a page that can't be fetched reliably doesn't get a second chance in a real-time answer pipeline.

To be quoted once retrieved, a page needs to:

Answer the question in the first sentence or two of a clearly headed section — engines lift passages, not pages. This is answer-shaped content.
Be specific: numbers, comparisons, concrete criteria. Vague marketing prose gives the model nothing to quote.
Be internally consistent with what other retrieved sources say about you — an outlier claim tends to get dropped in favor of the consensus.

The playbook

Fix the machine-readability floor first. Server-rendered copy, clean headings, structured data, no accidental crawler blocks. (Steps 1–5 of our AI visibility audit checklist.)
Build one page per buyer question. Question-shaped H1/H2, answer up front, evidence after. Ten focused pages beat one sprawling "ultimate guide" for retrieval.
Mine the citation graph. Run your money prompts, collect every domain the engines cite, and get present on the third-party ones — those pages are already winning retrieval for your questions. Being mentioned there is faster than beating them.
Keep freshness signals honest. Visible dates, updated content, accurate structured data. Real-time engines have a working notion of stale.
Measure per platform. Perplexity cites almost always; Gemini and ChatGPT search cite selectively; Claude rarely browses. Track citations and mentions separately per engine so you know which lever moved. That per-platform split is exactly what geopsy tracks run over run.

What doesn't work

Keyword-stuffed "AI SEO" pages. Retrieval is semantic; the generation stage actively drops low-substance sources.
Optimizing only your own site. If the engines' trusted sources for your category don't corroborate you, your page gets retrieved but your brand doesn't get recommended.
One-off checks. A screenshot from today says little about next week — grounded answers move with the web. Treat citations as a metric, not a milestone.

For the wider strategy this fits into, read how to rank in ChatGPT and AI search; for the conceptual foundation, what is GEO.