Search Box Under Pressure: Agentic Discovery Forces Retailers to Rebuild Onsite Search

Onsite search is undergoing its biggest shake-up since faceted navigation became standard – and the pressure is coming from outside the shop. According to the fourth edition of Salesforce's State of Commerce report, published in late July 2026, the use of agentic search as the first step in the shopping journey grew 200% year over year. The study draws on surveys of 3,450 commerce professionals and roughly 4,690 consumers, plus behavioural data from more than 1.5 billion shoppers across 37 countries.

The direction of travel is uncomfortable for retailers. Between August 2025 and May 2026, Salesforce found that product discovery via brand-owned properties fell 7% and traditional search fell 15%, while new channels – AI assistants, social AI and delivery apps – grew 38%. Traffic referred from AI chats grew between 150% and 428% year over year in every quarter measured. Caila Schwartz, who leads agentic commerce shopper insights at Salesforce, warns that brands absent from that first conversation are "invisible before the journey even begins."

Vendors are responding on two fronts. First, search inside the shop is becoming conversational. Salesforce's Agentic Commerce Search trains a small custom model using synthetic searches, clicks and purchases at roughly a 10:1 ratio to real behavioural data; the company reports around 33% more clicks per user, 17% more add-to-carts and a sharp reduction in hand-written merchandising rules. Second, retailers are making catalogues machine-readable. The protocol stack now includes the Agentic Commerce Protocol (ACP) built by OpenAI and Stripe, Google's Universal Commerce Protocol (UCP) unveiled at NRF 2026, and Anthropic's Model Context Protocol as a data connectivity layer. Shopify opened UCP and Catalog API access to developers in June 2026, and Mastercard brought a sandbox live in the UK in August so merchants can test agent discoverability, trusted payment flows and dispute handling before go-live.

The vendor landscape is reshuffling too. Gartner published an updated Magic Quadrant for Search and Product Discovery on 22 June 2026, again naming Constructor a Leader, while Forrester's Wave on commerce search widened the frame from keyword retrieval to broader discovery tooling: product listing pages, facets and attribute governance. Consolidation continues – Athos Commerce has combined Searchspring, Klevu and Intelligent Reach since 2025 – and Netcore Unbxd is pushing an "Enrichment for Agentic Commerce" offering aimed squarely at visibility inside AI shopping channels.

Europe offers a concrete example. German retailer Otto launched an AI assistant built on Google Gemini in March 2026 that handles text and voice input, asks clarifying follow-up questions and surfaces matches from roughly 18 million items, with the stated goals of easier purchase decisions and lower return rates. The company has also added camera-based visual search.

Enthusiasm has limits, though. ChannelEngine's Marketplace Shopping Behavior Report found that 58% of 4,500 surveyed online shoppers had used AI tools for product search, yet only 17% felt comfortable buying directly through AI; 53% still compare across an average of three marketplaces, and missing reviews make 60% hesitate. A Criteo survey of more than 6,000 consumers in six countries put chatbot use for product search at 39% and for product comparison at 47%.

Then there is measurement. In agent-mediated journeys, the behavioural data stream often begins only at add-to-cart, leaving attribution, personalisation and retail media partially blind. For search teams, the 2026 priority list is therefore unglamorous but clear: structured product data, disciplined attributes, agent-ready APIs and tested fallbacks for zero-result queries.