AI shopping agents now decide which stores appear in purchase recommendations. They bypass the vast majority of online retailers. A new study exposes the scale of this shift.
Recomaze tested six purchase-intent queries across 9,720 ecommerce stores using Google Gemini. The AI surfaced products from just 19 percent of those sites. The rest drew a blank. Most stores simply do not exist to the models that increasingly guide consumer spending. (The Next Web, June 2026)
The finding arrives as major tech companies push autonomous agents deeper into commerce. OpenAI, Google, Perplexity and Amazon have rolled out features that let chatbots research, compare and sometimes buy products on users’ behalf. Shoppers no longer scroll endless grids. They ask questions. The agents answer with a handful of options drawn from structured data they can parse quickly.
But the data those agents rely on remains patchy. Product catalogs often sit behind poor formatting, inconsistent descriptions or thin metadata. Agents favor sites that speak their language – clean, machine-readable attributes, detailed specifications, explicit answers to common objections. Everything else fades into noise.
Consumer surveys reflect growing yet cautious adoption. Only 13 percent of people say they mostly or completely trust AI for shopping advice. Personal recommendations from friends still command 53 percent trust. Just 4 percent would let an AI complete a purchase without final human review. (YouGov, July 2025)
Privacy worries top the list of concerns. Thirty-four percent cite data security. Thirty percent fear aggressive upselling. Twenty-four percent doubt the accuracy of suggestions. The hesitation is real. Yet usage climbs. Sixty-six percent of frequent online shoppers now consult AI tools like ChatGPT before buying. Thirty-four percent of those turn to it first for product discovery. (Yotpo, January 2026)
Brands that optimize for these agents see different results. Recomaze builds tools that rewrite product content for AI readability. It generates structured summaries, Q&A snippets and knowledge layers that agents can query directly. Early users report longer conversation times with on-site agents – three minutes 42 seconds on average, 18 seconds longer than before. The extra time translates into higher conversion.
The gap widens when agents act with greater autonomy. Harvard Business Review researchers note that traditional marketing tactics – brand storytelling, emotional appeals, paid placements – often fail with non-human shoppers. AI agents optimize for price, features, availability and verifiable claims. They ignore image-heavy pages or vague copy. They scan for structured signals instead. (Harvard Business Review, May 2026)
And the stakes rise fast. Protocols like Google’s Universal Commerce Protocol and OpenAI’s agentic tools let assistants transact across retailers without forcing users into any single walled garden. Amazon has responded by tightening its robots.txt rules and blocking certain crawlers. It removed hundreds of millions of its own product listings from ChatGPT search results earlier this year. The move signals how seriously platforms view the threat to their discovery monopoly.
Smaller merchants feel the pressure first. A brand that ranks well in Google search can still receive zero mentions from AI recommendation engines. Competitors with cleaner data structures capture disproportionate share of AI-driven traffic. One analysis found that missing the top three AI recommendations can cost a mid-sized store more than $23,000 monthly in lost sales. (Hexagon, January 2026)
Yet the solution isn’t simply adding more keywords. Agents learn from interactions. They cross-reference reviews, pricing history, return policies and stock accuracy. Inconsistent data across a catalog leads them to dismiss entire sites. One wrong variant description and the agent skips the product to avoid risk.
Companies such as Rep AI and Alhena position themselves as sales-first agents. They don’t wait for queries. Behavioral triggers prompt them to engage shoppers heading for the exit. They pull real-time inventory and craft personalized arguments on the spot. Early results show lifts in average order value and conversion where traditional chatbots only handled support tickets.
The shift rewards data discipline. Merchants must maintain unified product graphs that feed every channel – their own site agent, external models, comparison tools. Fragmented systems create blind spots. Agents notice those blind spots immediately.
Trust remains the bottleneck. Shoppers want AI to narrow choices and explain trade-offs. They hesitate to surrender final control. Four percent willing to let agents buy outright represents a tiny beachhead. That number will grow as reliability improves and safeguards multiply. But brands cannot assume passive discovery will carry them forward.
Executives at Recomaze argue the industry has reached an inflection. AI agents already influence billions in purchases. Visibility inside those agents has become table stakes, not a nice-to-have. Stores that treat product data as code – structured, versioned, queryable – gain an edge. Those that treat it as marketing collateral fall out of consideration.
The research from Recomaze carries a blunt warning. If current patterns hold, more than 80 percent of ecommerce inventory risks becoming invisible to the next generation of shoppers. The stores that adapt will capture a larger slice of an expanding pie. The rest risk watching their traffic migrate to competitors that speak fluent AI.
Retailers have faced platform shifts before. This one moves faster. The models improve weekly. Consumer habits follow. Data readiness determines who gets seen when a user says, “Find me the best running shoes under $120 that don’t hurt my knees.” The answer comes back with three options. Everything else disappears.
WebProNews is an iEntry Publication