TLDR
Adobe's analysis of more than a trillion US retail visits found AI-referred shoppers now generate 53% more revenue per visit than other traffic, reversing a gap that favoured non-AI visits by 128% a year earlier. Meta's Muse, which reached 5.6 million shopping requests in under two days, now accounts for 72% of all daily AI-agent traffic tracked by HUMAN Security.
KEY TAKEAWAYS
The revenue reversal
A year ago, a shopper arriving from an AI assistant was worth less than one arriving from any other channel. Adobe's analysis of more than a trillion US retail site visits, published in its Q2 2026 report, found that position has completely reversed. AI-referred shoppers now generate 53% more revenue per visit than other traffic, a swing from twelve months earlier when non-AI visits were worth 128% more.[1]
Adobe found AI-referred visitors spent 48% longer per visit, viewed 13% more pages and recorded bounce rates 32% lower than non-AI traffic.[1] Someone arriving via an AI agent has already filtered, compared and decided, so they land on the product page closer to buying than a shopper who found the site through paid search or a social feed.
Muse's two-day sprint
Meta's Muse agent launched on 8 September 2026 and reached 5.6 million shopping requests in under two days, a volume ChatGPT took eleven months to accumulate, roughly a 165-times compression of the adoption timeline.[2] Within HUMAN Security's traffic data, Muse now accounts for 72% of all daily AI-agent requests, and 84% of those requests land on product or search pages rather than homepages or editorial content.[2]
Muse is also 25% more likely than other AI agents to reach checkout.[2] As this publication reported when Muse arrived in Australia on 24 September 2026, the agent bypasses human-initiated advertising entirely, so a sale Muse drives will not appear in any campaign attribution report a media buyer is currently reading.[3]
A fragmented shelf
Jellyfish's Share of Model platform, released 1 October 2026 and covering Australia, the US, the UK and Singapore, tested what each major assistant recommended in the same product category. Alexa for Shopping pulled from 177 toy brands, almost entirely from Amazon's own store. ChatGPT cited 24 retailers and Google AI Mode cited 37.[4]
John Dawson, Vice President of Strategy at Jellyfish, said: "Brands have spent decades optimising products for search engines, marketplaces and retailer shelves. Agentic commerce changes that equation." Dawson said the same product question can return a thirty-retailer shortlist on one assistant and a closed, single-store answer on another, so "winning AI isn't one race, it's many, and most brands cannot yet see the starting line."[4]
Natasha Wallace, Chief Solutions Officer at Jellyfish, said: "Until now, marketers have had little visibility into why AI shopping systems favour one product over another." Wallace said that visibility gap has been the core problem her team built the platform to address.[4] Tinuiti's brand citation data from July 2026 puts numbers on the concentration gap: the top 100 retailers accounted for 34% of Microsoft Copilot citations, compared with 4% on Google AI Mode and 4% on ChatGPT, meaning dominant players own Copilot's shelf while the long tail of retail brands fares far better on the latter two assistants.[5]
What Australian ecommerce teams should measure now
Australia is one of four markets covered in Jellyfish's Share of Model tracking, and local retailers have been receiving Muse traffic since the agent's public launch on 8 September 2026.[3] A Dentsu Creative survey of 1,950 senior marketers across 14 markets, conducted in April 2026, found 59% were already optimising for AI search and 50% were investing in agentic commerce, though investment in measurement has lagged behind investment in optimisation.
AI-agent requests carry distinct user-agent strings and header patterns in server logs, and Muse's requests are now large enough to appear as a named segment in most web analytics platforms. Teams running Google Analytics 4 (GA4) that have not yet created an agent-traffic segment are making budget and merchandising decisions without accounting for a channel that represents 72% of all AI-agent volume hitting their category.[2]
The assistant-specific citation gap Jellyfish identified means product feed optimisation, structured data and pricing accuracy matter differently depending on which assistant a brand wants to win. A retailer absent from ChatGPT's 24-retailer shortlist faces a different problem from one absent inside Alexa's Amazon-dominated 177-brand set, and the fix for each is distinct. Jellyfish's Share of Model platform has covered the Australian market since 1 October 2026.[4]
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
Why do AI-referred shoppers generate more revenue per visit?
How quickly did Meta's Muse grow compared with ChatGPT?
Why does it matter which AI assistant a shopper uses?
How can Australian ecommerce teams start tracking AI-agent traffic?
Xaviery Malinao writes for Prompt the Market on how brands and agencies are adapting to answer engines, drawing on Bushnote's work with clients across search, AI search and content.
