What ChatGPT, Perplexity, Claude, and Gemini Actually Cite

The advice that circulates in most AEO content treats the four major AI platforms as variations on a theme. Fix your directory presence, earn backlinks, update your schema -- and you'll improve citations across ChatGPT, Perplexity, Claude, and Gemini simultaneously.

Our August and September 2026 research shows they're not variations. They're distinct citation ecosystems drawing from sources that barely overlap. What gets you cited in Perplexity can be irrelevant to Claude. What ChatGPT wants from a local plumber is different from what Gemini rewards for the same business.

The Sanbi.ai Source Affinity Study

In our September 2026 knowledge update (`platform-citation-behaviors.md`, session 124, 2026-08-31), we documented findings from Sanbi.ai's cross-engine source affinity study: 119,939 citations tracked across ChatGPT, Gemini, Perplexity, and Claude on the same prompt set. This is one of the first published datasets to map citation sources for all four platforms simultaneously rather than treating AI search as a single channel.

The engine-level breakdown:

**Claude** cites patents and analyst reports as primary citation types. Of the four platforms, Claude is the most differentiated -- it pulls from authoritative, specialized document types that the other platforms don't favor. For a B2B technology company with published research, white papers, or patent filings, Claude is the most achievable citation target. For a local electrician or restaurant, it's close to unreachable by design.

**ChatGPT** cites manufacturer-official domains as the dominant source. This differs from what the Yext study (17.2M citations, 2026) characterized as ChatGPT's directory preference -- the Sanbi.ai data, which is more recent, points toward brand-official and manufacturer-owned domains rather than third-party directories. The Yelp and Foursquare data partnership layer still feeds ChatGPT for recommendation queries, but the baseline citation source is the official channel.

**Perplexity** shows YouTube and Reddit as dominant community source surfaces. We've tracked Perplexity's Reddit citation behavior since session 2 (`perplexity-citation-triggers.md`, 2026-04-22), and the methodology rec from session 123 (2026-08-30) confirmed Perplexity now holds 71% of cross-platform Reddit citations following ChatGPT's August 14 collapse in that channel. The YouTube signal is a newer addition to this picture -- Perplexity is pulling from community substrate, not brand-owned content.

**Gemini** isn't characterized specifically in the Sanbi.ai data we reviewed, but the Yext 2026 study found Gemini pulls 52% of citations from brand-owned websites -- the inverse of ChatGPT's behavior. Gemini rewards strong on-site structure more consistently than the other three platforms.

The Sanbi.ai study included one finding that confirms what we've seen across every cross-platform analysis we've done: sources dominating ChatGPT's citation list are "cited effectively zero times by Claude on the same query set." The platforms are not drawing from a shared pool. Optimizing for one does not transfer to another.

What 129,000 Domains Show About Google Ranking and AI Citations

Session 126 of our platform research (2026-09-02) covered a separate dataset: LeadsNow.ai's analysis of 129,000 domains for AI citation correlation factors. The results challenge the most common piece of advice businesses receive about AI visibility:

- 80% of AI-cited sources were not in Google's top 3 results - Only 29% of AI citations came from Google's first page at all

That means if a business is putting its energy into ranking higher on Google and waiting for AI citations to follow, it's targeting 29% of the AI citation pool. The other 71% is accessed through infrastructure that Google's algorithm doesn't control: directory data feeds, entity recognition layers, publisher licensing agreements, and community presence on platforms like Reddit and YouTube.

The mechanism LeadsNow.ai describes is consistent with what we've documented across multiple studies. AI platforms pull citations from training data, partner data feeds (Yelp, Foursquare, Thumbtack, Angi), real-time retrieval, and licensing agreements -- not from Google's ranking signals. A Foursquare listing that ranks 14th in traditional Google search may appear in a Foursquare data feed that feeds directly into the AI citation layer, completely independent of that ranking.

Reconciling With the Perplexity-Google Overlap

There's an apparent tension here. Our April 2026 research on Perplexity (`perplexity-citation-triggers.md`, session 2) found that 60% of Perplexity citations overlap with Google's organic top 10 for the same query -- suggesting Google ranking matters considerably for at least one platform.

Both findings hold simultaneously. The LeadsNow.ai 80% figure measures across all four platforms combined, not Perplexity alone. When Perplexity (which is the most Google-correlated AI platform, running real-time RAG through web retrieval) is averaged with Claude (patents), ChatGPT (manufacturer domains), and Gemini (brand-owned sites), the cross-platform average shifts heavily away from Google rank.

In our actual audits of local businesses, we found the Perplexity dependency on Google ranking to be real and concrete. A business that doesn't appear in Google's organic results for a category query is not in Perplexity's retrieval set for that query -- regardless of how well-structured its content is. For B-series (category) queries specifically, our audit clients scored 0.0 on Category Authority across all platforms, and the path to improving that score for Perplexity runs through Google ranking and directory presence, not content optimization.

For the other three platforms, Google rank is not the controlling input. Claude cites your analyst report coverage. ChatGPT cites your official domain and data-feed presence. Gemini responds to on-site structural signals. None of those inputs have much to do with whether you appear in Google's top three results.

What This Means for Fix Priorities

The four-platform source map suggests different starting points based on which platforms matter for your business.

For most local service businesses -- contractors, healthcare practices, restaurants, personal services -- the primary AI citation surfaces are Perplexity and Google AI Overviews. The fix work concentrates on directory infrastructure: verified listings with consistent category data, at least one platform with active reviews, and presence in the specific directory your vertical uses (Angi for home services, Healthgrades for healthcare, Avvo for legal).

For B2B businesses and professional services, Claude becomes a realistic target for the first time. The fix is different in kind: published research, analyst coverage, documented methodology, patent filings. The businesses Claude cites have an institutionally recognized body of work -- not just a well-structured website.

For consumer product brands and manufacturers, ChatGPT's official-domain preference means the brand's own infrastructure carries more weight than it does for service businesses. But recommendation queries still flow through the Yelp and Foursquare data layers -- brand ownership alone doesn't get you into those feeds.

A Signal Check gives you a baseline across all four platforms and identifies which citation gaps are structural versus fixable. If you're scoring strongly on one platform and showing near-zero on another, the source affinity data above likely explains why -- and the fix is in a different channel than the one you've been working on.