Which AI platform cites what? Five LLMs, the same questions, almost no overlap.

The headline finding

We ran the same 500 standardized commercial-intent queries through five AI search engines — ChatGPT, Perplexity, Claude, Gemini, and Grok — and recorded every cited source. The corpus covers 12,500 individual responses.

The result almost no SEO veteran expects: only 6.8% of cited domains appeared on three or more platforms. The cross-platform overlap most teams plan around does not exist at the level practitioners assume. Optimizing for ChatGPT does not, in any meaningful sense, optimize you for Claude.

This finding aligns with SparkToro and Gumshoe's 2025 research, which found AI models produce different brand recommendation lists more than 99% of the time when asked the same question repeatedly. The probability of receiving identical lists in the same order drops below 0.1%.

“Any tool that gives you a "ranking position in AI" is full of baloney. — Rand Fishkin, SparkToro”

Why the platforms diverge

Each AI platform pulls from a different search backend, weights different source classes, and applies different freshness rules.

ChatGPT leans heavily on Wikipedia and Bing-indexed editorial. Perplexity disproportionately surfaces Reddit and forum content. Gemini favors YouTube and Google-indexed pages. Claude pulls a more diversified mix via Brave. Grok pulls disproportionately from X conversations.

If your strategy treats “AI visibility” as a single channel, you will systematically miss four out of five surfaces.

Ranking position is the wrong metric

SparkToro's research surfaced a finding that fundamentally reframes how visibility should be measured. Asked the same brand-discovery question 71 times, ChatGPT named City of Hope hospital in 97% of responses for West Coast cancer care — but as the #1 mention in only 25 of those answers.

The implication: AI engines are probability machines. They are designed to generate unique answers every time. Treating them as deterministic ranking systems is “provably nonsensical.”

What is a valid metric: visibility percentage across many runs. Certain brands consistently appear in 80–97% of responses for their core categories. Others appear in 2%. That delta is the real signal — and it requires running each prompt 60–100 times minimum to surface.

What the 6.8% have in common

The minority of domains that did appear across three or more platforms shared a consistent profile:

• Strong entity definition in knowledge graphs (Wikidata, Google Knowledge Graph, schema.org Organization markup with consistent identifiers)
• Editorial mentions in third-party publications — not just owned media. Trust-tier outlets every model crawls.
• Active community presence on Reddit, LinkedIn, and niche forums where Perplexity, Grok, and Claude pull conversational signal.
• Schema-marked structured content with extractable 50–150 word passages.
• Sub-second page performance (FCP under 0.4 seconds), which ChatGPT in particular weights heavily.

None of these are tied to a single platform's ranking algorithm. They build the entity itself, which every model independently learns. The cross-platform brands aren't optimizing for five engines — they're building one credible entity that all five recognize.

The strategic implication

Two operational shifts follow from this data:

1. Stop reporting "AI visibility" as a single number. Break it out per platform. Your Perplexity visibility is a Reddit/forum problem. Your ChatGPT visibility is a Bing/Wikipedia problem. Your Gemini visibility is a YouTube/Google-indexed problem. They have different solutions.
2. Build the entity, not the placements. Every dollar spent strengthening cross-platform signals (knowledge graph, editorial media, community presence) compounds across all five engines. Every dollar spent gaming a single platform stays inside that platform.

The brands present in 4+ platforms are 2.8× more likely to appear in ChatGPT specifically (Digital Bloom). Multi-platform presence is itself a signal.

Methodology

Queries were drawn from a stratified sample of high-intent commercial searches across SaaS, consumer, healthcare, and B2B services. Each query was issued to each platform under identical conditions, with identical session-state controls. Citations were scraped from response pages, deduplicated by root domain, and cross-referenced.

The full per-platform overlap matrix and category-level breakdowns are available on request: research@visigent.com.

Every statistic in this piece follows Visigent's research standards: named sources, dated figures, corrections on request to help@visigent.com.