The Maze: AI search is not one channel. In one consumer-brand dataset, Perplexity sourced 61% of citations from earned media. ChatGPT split almost evenly between earned media at 38% and competitor content at 36%. Google’s AI surfaces leaned harder on aggregators and social sources. The same mix changed again as customers moved toward conversion. Brands looking for one universal optimisation checklist are solving the wrong problem. The real unit of work is engine × buying stage × category.
Each engine builds authority from a different neighbourhood. Perplexity put 61% of citations into earned media and another 24% into aggregators. ChatGPT used only 38% earned media, while competitor content supplied 36%. AI Overview drew 28% from aggregators, 18% from social and 32% from earned media. AI Mode landed at 29%, 12% and 43%, respectively. Those are not small tactical differences. A brand could dominate PR coverage and still lose visibility on a surface that prefers comparison sites, community discussion or rival-owned pages.
The source mix also changes as intent hardens. Earned media supplied 62% of citations at awareness but only 31% at conversion. Aggregators moved the other way, from 14% to 35%. Competitor content rose from 5% to 17%. That shift follows shopping logic: broad authority can introduce the category, but a buyer closer to purchase wants comparisons, alternatives and proof that survives side-by-side evaluation. The channel that wins the first question may not win the final shortlist.
This turns AI visibility into a portfolio problem. Perplexity’s pattern rewards credible third-party coverage. Google’s surfaces make aggregator and social presence harder to ignore. ChatGPT’s 36% competitor-content share means a brand’s visibility can depend partly on how rivals frame the market. The practical response is not to publish more everywhere. It is to map priority prompts by funnel stage, record which source classes appear for each engine, then fund the gaps. Otherwise a brand can spend heavily on owned content while the answer engines keep shopping elsewhere.
Independent research points to the same platform split. Google confirms that AI Overview and AI Mode may use query fan-out, issuing multiple related searches across different sources, and says the two surfaces can show different links. Ahrefs found that 65% of Perplexity’s cited URLs matched Google’s top 10 for short-tail terms, versus 10% for ChatGPT. Its March 2026 update also found only 37.9% of AI Overview citations in the first ten search-result blocks. Classic ranking still matters, but it is no longer the whole supply chain.
The percentages are a diagnostic, not a law of nature. The client, product, prompts, sector, sample size and observation period are not disclosed. Small unlabelled source types also had to be grouped as residual `Other`. The value is the shape of the problem: source preference varies by engine and stage, and published research shows those systems and models keep changing. Brands need repeated measurement of their own commercial prompts, not a permanent platform stereotype.
Why it matters: Ecommerce discovery is moving from a ranked page to a portfolio of generated answers. That raises the cost of being strong in only one source class. PR, comparison sites, social proof, product information and competitive positioning now work as inputs to different machines at different moments. The brands that win will not chase every AI tactic. They will build a small engine-by-funnel scoreboard, link visibility to conversion, and move budget toward the sources that actually shape the next customer decision.
