AI Became the First Step in Shopping — Discovery Up 200%, Traditional Search Down 15%

Salesforce published its State of Commerce report, surveying 3,450 commerce professionals across 20 countries and 13 industries, including Korea. The conclusion is unambiguous: the starting point of product discovery is moving from the search box to the AI assistant.

What the numbers say

The share of global consumers using an AI assistant at the first stage of the shopping journey grew 200% year over year. Traffic arriving at commerce sites via AI chat rose 150% to 428% year over year in every quarter of the study period.

The other side shows the offset. Between August 2025 and May 2026, product discovery through traditional search fell 15%, while discovery through newer channels — AI assistants, social AI, delivery apps — rose 38%. Total discovery is not shrinking; the routes are being replaced.

What commerce teams have already done

Among Korean commerce companies, 48% already supply product data feeds to AI search platforms, and 44% have started optimizing content for conversational search. With close to half moving, the wait-and-see phase is over.

The real bottleneck is data, not visibility

The difficulty comes after visibility. Converting AI-driven discovery into purchase and retention requires integrated customer, product and inventory data, and that is where teams stall. Among Korean companies, 45% report duplicate or inconsistent customer data, 43% report price and promotion mismatches across channels, and 39% report inventory synchronization problems.

These three share a nature: they are errors a human shopper used to absorb. A price that differed by channel got shrugged off; an inventory mismatch got patched with an out-of-stock notice. An AI assistant instead summarizes the inconsistency and hands it to the user. Once a wrong price is baked into an answer, there is no touchpoint left to correct it.

What marketers should do now

First, promote the product feed to a marketing asset. Feeds have historically been plumbing owned by engineering, but the moment an AI reads that feed to compose a recommendation, its copy, attributes and images become creative. Feed attributes deserve the same care as detail-page copy.

Second, make cross-channel price and promotion consistency a recurring operational check. A quarterly audit is too slow; align the cadence with how often AI answers refresh.

Third, repair attribution. If traffic arriving from AI chat collapses into "other" or "direct" in your referrer taxonomy, a 150–428% increase exists but never appears in a report. For evidence that shoppers verify AI recommendations elsewhere rather than trusting them outright, see Half of U.S. Shoppers Verify AI Recommendations on Reddit; for planning around purchase confidence rather than channels, see Customers Assemble Confidence.