Where AI citations come from: consensus (ChatGPT) vs owned site (Gemini) vs reviews (Perplexity)

AI assistants don't cite the same kind of source: ChatGPT leans on web consensus — the aggregators and reference sites the internet broadly agrees on — Gemini cites your own domain more than any other engine, around 52% of its citations in 2026 studies, and Perplexity leans on reviews and fresh, niche directories. Each engine trusts a different place, which is why appearing in all three takes three different plays, not one.

We spend our days getting WooCommerce and WordPress content into these answers, and the split above is the thing most "AI SEO" advice flattens into a single checklist. Contexta works from a site's own Search Console data precisely because the useful next move differs by engine — and by which engine actually sends you anything. What follows is the citation-share number behind each engine's preference and the concrete asset you build to win it, with one detail worth watching: only one of the three plays happens on your own site.

Where do AI citations actually come from?

They come from three distinct source pools, one per engine, and the pools barely overlap — only about 11% of cited domains appear across more than one platform. ChatGPT draws on web consensus, Gemini on your owned domain, and Perplexity on reviews and recent directories, a split documented across 2026 cross-engine citation studies, one of which analyzed 17.2 million citations.

Two things follow. First, being cited by one engine tells you almost nothing about the others — a divergence we trace to each engine's retrieval design in why ChatGPT and Perplexity cite different sources. Second, and this is the practical part, the work that earns a citation lives in a different place for each engine, so the question stops being "how do I rank" and becomes "which source pool am I even trying to enter."

Why does ChatGPT cite web consensus instead of your own page?

Because ChatGPT rewards what the internet broadly agrees on, not what any single page — including yours — says about itself. On subjective queries, third-party directories like G2, Yelp and TripAdvisor make up roughly 48.73% of its citations, and Wikipedia is its single most-cited domain; your own product page rarely makes the cut because it's one voice, not a consensus.

The actionable consequence is uncomfortable but clear: you can't write your way into ChatGPT from your own domain. Winning it is off-site work — getting consistent facts about your product onto the aggregators and reference sources it trusts, earning reviews on the directories that dominate your category, and keeping your name, category and key specs identical everywhere so a consensus can actually form. That is the experience-and-authority signal E-E-A-T describes applied to citation rather than ranking: corroboration beats self-description.

Why is Gemini the one engine that mostly cites your own site?

Because Gemini inherits Google Search's logic, and Google has always been willing to surface a brand's own domain for its own topic — so your site is directly in play. In 2026 analyses, roughly 52% of Gemini's citations pointed to the brand's own domain, far more than ChatGPT or Perplexity, which makes Gemini the one engine where the content you control is the content that gets cited.

Honest caveat, because the studies genuinely disagree: other analyses find third-party editorial still dominates Gemini overall, so treat "owned domain" as a strong tendency, not a guarantee — it holds most when the query is about you or your category. Either way the play is the highest-control one you have: structure each page answer-first around the real questions people ask, so the passage matches what Gemini's query fan-out is looking for. This is the on-site lever Contexta's AI editor is built for — it rewrites pages answer-first in the site's own voice and targets each page's actual Search Console queries, which are the same queries Gemini's Google-derived retrieval is scoring against. The queries worth targeting first are the ones your own Search Console data already shows you're close on.

Why does Perplexity lean on reviews and fresh directories?

Because Perplexity retrieves the live web on every query and weights recency and specialization heavily — it cites content under 30 days old at roughly an 82% rate, averages the most sources of any engine (about 21.87 per answer), and favors niche, industry-specific directories more than any other assistant. A polished but static owned page loses here to a recent review roundup or an actively maintained directory listing.

So the Perplexity play is freshness plus presence in the right directories: keep your key pages genuinely current — a cadence question more than an editing one, since content from the last 13 weeks takes about half of all AI citations — earn reviews where your category's buyers actually look, and get listed in the niche directories it over-indexes on — the same off-site sources that decide your visibility more than your blog does. One caution, though — none of that pays off until the page clears the basic eligibility floor, because a fresh, review-backed page that PerplexityBot can't fetch, or that hides its text behind JavaScript, still gets cited nowhere. That floor sits beneath all three engines, and it's the same set of five reasons a page never gets cited at all.

How do you cover three engines without running three teams?

You sort the work by where it lives, do the shared floor once, and take the highest-control lever first. Only one of the three engine plays — Gemini's — happens on your own site; ChatGPT and Perplexity are won off it, through corroboration, reviews and directory presence. That single fact reorganizes the whole effort: your content team's on-site writing moves Gemini, while ChatGPT and Perplexity need a reputation-and-listings push that publishing more blog posts won't deliver.

Which of the three to lead with isn't a guess — it's whichever engine already cites you and sends visitors, since citation volume for the same brand swings enormously between platforms. Build the floor, take the on-site Gemini win because you control it outright, then spend your off-site effort on the engine your own traffic says is worth it. The three-way split isn't a burden; it's a map telling you exactly which kind of work — owned, earned, or reviewed — each citation actually requires.