The Maze: EMARKETER's GEO forecast turns an awkward marketing acronym into a mainstream behavior problem. Search is already the leading generative AI use case in the US, at 34.0% of internet users in 2026. By 2030, it reaches 45.3%. That is not a niche tool for curious prompt people. It is a new discovery layer growing before measurement, playbooks, and budget discipline have caught up.
Search is the breakout use case, not a side quest. EMARKETER's public report anchors the key stat: more than 45% of US internet users will use generative AI for search by 2030. The official forecast starts search at 34.0% in 2026 and ends at 45.3% in 2030. Other functions also rise, including text creation, work use, shopping, image generation, and health. But search sits above all of them across the period. That matters because search is where intent, comparison, and brand memory begin. If AI search becomes a monthly habit for nearly half the US internet population, discovery moves from a ranked-results contest into an answer-and-citation contest.
The shopping line is smaller, but more dangerous than it looks. EMARKETER defines generative AI shopping users as people using third-party AI platforms such as ChatGPT and retailer-native assistants such as Amazon Rufus, while excluding AI-powered search summaries. That makes the shopping series narrower than the search series. It still climbs from an inferred ~25.6% in 2026 to ~33.9% in 2030. The implication is blunt: commerce brands will face two AI surfaces, not one. First, broad AI search decides which brands enter consideration. Then retail assistants shape comparison, substitution, and confidence closer to purchase.
GEO is really a credibility system wearing an optimization name tag. Kelsey Voss's LinkedIn post frames the practical tension well: CMOs are being pushed to act while attribution models and best practices remain immature. Her strongest point is that GEO should not become another attempt to game the system. The durable work is less glamorous: stronger product pages, consistent claims, structured data, credible third-party mentions, and source material that models can cite without mangling the brand. In classic SEO, ranking first could hide weak content for a while. In AI answers, weak source material can turn into a wrong summary, a missing citation, or a competitor recommendation.
Measurement will lag the budget request. A 2026 paper on AI brand recommendations found that when an assistant recommends a brand to an observably unengaged user, same-name Google search rises 4.3 percentage points, own-site visits rise 2.4 points, and brand-specific retailer-page visits rise 1.0 point. The authors stress that the design is observational and does not observe transactions. Still, the direction is useful: assistants can create brand exposure that later shows up as ordinary search or retail navigation. Last-click dashboards will call that organic. The model did some of the selling. The dashboard missed the handshake.
Retailers should treat AI search as a new shelf, not a new ad unit. A separate Ctrip study found that platform AI chat appears in the same broad purchase-journey phase as traditional search, with users moving back and forth between chat and search. That is the operator lesson. AI discovery will not neatly replace the old funnel. It will sit on top of it, interrupt it, and sometimes route shoppers around it. Brands that wait for perfect attribution will arrive with clean dashboards and dirty shelves.
Why it matters: The EMARKETER forecast makes GEO feel less like a search-industry fad and more like brand infrastructure. By 2030, AI search is projected to be a monthly behavior for 45.3% of US internet users. For ecommerce teams, that means product data, reputation signals, content authority, and retailer-native assistant readiness are no longer separate chores. They are the new shelf discipline.
