The Maze: AI shopping is moving from an experiment to a mass-market habit. EMARKETER expects 104.1 million Americans to use generative AI for shopping each month in 2029. That equals 30.3% of the US population. By 2030, the share reaches 31.7%. A new discovery layer is forming in front of roughly one in three consumers.
The audience is already large enough to change road maps. EMARKETER says nearly 80 million US consumers will use AI shopping tools in 2026. Growth then falls from 13.1% in 2027 to 5.0% in 2030. Applying the published rates around the exact 2029 value produces approximate monthly audiences of 79.6 million, 90.0 million, 97.4 million, and 109.3 million for the other years. The pace cools, but the base compounds.
Discovery is advancing faster than transaction ownership. In its state of AI shopping assistants, EMARKETER argues that consumers are increasingly using these tools for research and product discovery, while conversion remains less mature. That split matters. A retailer can lose the first conversation without losing the checkout—or win a recommendation and still hand the purchase to someone else. Search visibility, product-feed quality, merchant authority, and checkout conversion are becoming related but separate operating jobs.
The traffic is getting commercially better, not just bigger. Adobe analyzed more than 1 trillion US retail-site visits and reported that traffic from generative AI sources rose 393% year over year in the first quarter of 2026. In March, those visits converted 42% better than non-AI traffic; a year earlier, they converted 38% worse. Adobe also found 12% higher engagement, 48% more time on site, and 13% more pages viewed. Early curiosity is starting to behave more like qualified intent.
Retail infrastructure is still written for people and search crawlers. Adobe found only 66% of sampled product pages were machine-readable, leaving a third difficult for AI systems to interpret. That makes the first competitive move unglamorous: structured product data, accurate availability, clear pricing, trustworthy reviews, consistent variants, and pages that agents can parse. The retailers that make their catalog legible can earn distribution before anyone has agreed what “AI optimization” should be called.
The forecast is directional, not a revenue guarantee. The public EMARKETER exhibit discloses population shares and annual growth, but not the full model or exact definition of shopping assistance. Monthly use can include research, comparison, and purchase support. It does not mean 104.1 million people will buy through an AI agent. Operators should track referred sessions, assisted conversion, product inclusion, margin, and repeat purchase.
Why it matters: By the time a channel reaches one-third of US consumers, “wait and see” is no longer a strategy. Retailers should treat AI assistants as a new discovery counterparty: make the catalog understandable, measure the traffic separately, and protect the economics at checkout. The interface may answer the question. The retailer still needs to win the basket.
