A regularly updated resource of data and statistics on how brands show up in AI search — for marketers benchmarking their AI visibility and journalists looking for citable research.
Compiled by Evertune, an AI marketing platform that helps enterprise brands monitor, measure and own the AI customer journey across ChatGPT, Gemini, Google AI Overview, Google AI Mode, Claude, Perplexity, DeepSeek, Meta AI and Copilot. All statistics are drawn from Evertune's proprietary platform, which uses data science at scale to provide statistically significant insights.
A summary of what Evertune's research has found about AI model behavior, shopping, and source citations as of March 2026.
Across ~220,000 prompts tested in 9 product categories, the newer model universally returned fewer brand recommendations. The top three-to-five brands were mostly stable across both model versions, but brands outside of that cohort saw markedly different visibility with GPT-5.4 mini.
When tested on Spring Break travel recommendations in March 2026, Gemini included specific warnings about Mexico's February 2026 cartel violence. Copilot recommended Guadalajara and Playa del Carmen with no safety caveats. Perplexity added only generic boilerplate disclaimers — regardless of destination-specific risk.
The remainder of the time, ChatGPT and Gemini execute live searches. When AI searches live, those sources become your roadmap — showing where AI looks for answers about your industry. But more than half the time, AI draws from patterns learned by reading the entire internet during its training cycle.
In October 2025, ChatGPT's shopping experience widget appeared in 20% or fewer of responses to all tested product/brand questions. By February 2026, that figure had surged to 80%+ for most categories — a fourfold increase in under five months.
Evertune asked six models the same 41 questions every day for 20 days. The volume of unique URLs cited per model differed by nearly 8×, underscoring how differently models search and surface content.
Most of 180+ newswire-syndicated links were essentially invisible when LLMs used search to respond to prompts via Retrieval-Augmented Generation (RAG). Newswire distribution does not translate to AI search visibility — a critical distinction for PR and communications teams.
Reddit was Perplexity's most-cited domain — by a wide margin — from October 2025 through February 2026, across 200 million prompts. (Notably, Reddit is currently suing Perplexity for alleged unauthorized scraping.)
Across 200 million prompts over 5 months, ChatGPT cited Wikipedia most often among Reddit, YouTube, LinkedIn, and Wikipedia. Even the single most-cited domain rarely exceeds 5% of total citations — a sharp contrast to how SEO dominance works.
From October 2025 through February 2026, across 200 million prompts, YouTube's share of AI citations trended upward across multiple models — a meaningful signal for brands investing in video content as part of their GEO strategy.
The other 95% of citations are spread across thousands of other domains. No single domain is dominant, and each domain's share changes over time. This is a fundamental difference between GEO and SEO: AI models draw from a vastly wider variety of sources than traditional search engines.
All statistics are sourced from Evertune's proprietary GEO and AI marketing platform, which tracks hundreds of brands across 250+ categories.
Research is conducted by running large volumes of prompts across major LLMs — including ChatGPT, Gemini, Perplexity, Google AI Mode, Copilot, and others — and analyzing the results at scale.
This page is updated as new research is published. For the full research library, visit evertune.ai/resources/research.
Our AI marketing platform can be seen in
