The Maze: Ask an AI assistant to compare two products and it can help you choose. Ask it to place the order and you hand over a different job. EMARKETER’s April 2026 figures put product comparison first and completing a purchase last among the six tasks shown. The survey base was 823 US residents aged 15–80 who used generative AI weekly. The percentages describe stated preferences. A respondent could favor AI for several tasks, so these shares cannot tell us how many shoppers progressed from one task to another.
The details
🔎 Choosing what to buy. In EMARKETER’s survey, 44% preferred AI for comparing products, 38% for comparing prices and 34% for getting recommendations. Those tasks can answer different questions about the same purchase. Product comparison helps a shopper weigh features across models, price comparison helps assess the offers, and a recommendation can narrow the list in the first place.
🛒 Looking up an item and placing an order. The same exhibit shows 28% preferred AI for looking up a specific product and 9% for completing a purchase. Knowing the item still leaves a decision about which seller to use. EMARKETER’s keynote recap describes shoppers working through decisions with a chatbot, doing more research and often buying from retailers. That is the keynote’s interpretation of the journey. These results describe preferences and give no order counts by seller or channel.
Our take
A retailer has two decisions to make. It can compete to be the store an assistant recommends and decide separately whether the assistant should handle the order.
🏪 Give the assistant a reason to name your store. 31% preferred AI for finding the best place to buy. Imagine an assistant recommending a laptop. To recommend a seller too, it would need to match the exact configuration to a live offer. A retailer could test whether keeping model identifiers, current prices and availability aligned across the data it shares and its product pages helps the assistant cite the right offer. If the assistant can use those facts, the retailer could compete for a visit even when another seller carries the same model. The survey does not establish which offers assistants recommend.
📦 Measure the path to the retailer’s checkout. Preference for AI to find the best place to buy was 22 percentage points higher than preference for AI to complete a purchase. If the assistant can send people to the exact item page and the retailer can identify those referrals, it could count arriving visits and completed orders separately. Few visits would tell the retailer to inspect whether its offers show up and earn clicks. Visits without orders would prompt a look at the landing page and purchase flow. Any later test of checkout inside the assistant could compare completed orders and what the retailer earns after fees. The survey supplies neither result.
