The AI Shopper Is Here: Retail Is Still Catching Up

A customer can arrive on a product page after using an AI assistant to compare alternatives and settle on a preferred option. The retailer may record the visit and eventual order with little context about how the choice was made.^[3]

That kind of research is becoming more common. In the 2026 Global Digital Shopping Index from PYMNTS Intelligence, commissioned by Visa Acceptance Solutions, 30% of respondents who used digital tools to research their latest purchase said they used ChatGPT, compared with 2% in 2024.^[1]

The study also found that 47% of respondents whose latest purchase was online used an AI platform or a retailer’s AI tool during that purchase. The survey covered the United States, Brazil and the United Arab Emirates. Those figures describe reported use among the surveyed shoppers, not purchases completed autonomously by AI.^[1]

For ecommerce teams, the challenge is understanding how this activity connects to their business. That includes the recommendations shoppers receive, the software accessing product information and the visits that eventually reach the store.

More of the shopping decision can happen before the click

Consider someone shopping for a backpack. They might describe their commute, laptop size and budget, then ask for something that also works on a weekend trip. An assistant can compare options against those requirements before the shopper opens a retailer’s website.

OpenAI’s shopping research supports that back-and-forth. It asks follow-up questions, compares products and lets people remove options or change their requirements. Its buyer’s guide presents a shortlist of recommended products alongside other options.^[3]

For a retailer, that shortlist may influence whether the shopper visits at all. The products shown depend on the request. A product left out of the recommendations may never get considered further by that shopper.

Shopify’s data gives a view of the traffic that does arrive. In the first quarter of 2026, AI-chatbot referral sessions on its platform grew more than eightfold year over year, while AI-referred orders grew nearly thirteenfold. Organic search still sent more sessions than all tracked AI platforms combined.^[2]

Where those visits began is particularly relevant. In Shopify’s analysis, 55% of AI-referred sessions started on a product page, compared with about 20% of organic-search sessions. That pattern is consistent with shoppers doing research before arrival, although the landing page alone cannot establish what happened in a particular conversation.^[2]

OpenAI says shopping research chats are not shared with retailers.^[3] The merchant can therefore receive the visit without the preferences, comparisons or objections discussed beforehand. The privacy boundary is appropriate, but it leaves retailers with less context for understanding the decision.

Not all AI-related traffic is the same

A person following a recommendation and software retrieving a product page are different kinds of activity. Combining them in one traffic total can make it harder to understand what is happening at a store.

Automated requests also serve different purposes. A crawler may collect pages for search or model training. A user-triggered fetch retrieves a page because someone asked an AI service to look something up. OpenAI documents separate systems for these jobs.^[5] An AI agent can go further, navigating a site or taking shopping actions on someone’s behalf.^[4]

HUMAN Security’s 2026 report found that 77% of the agentic AI activity it observed during 2025 occurred on product and search pages. It also recorded requests to account, sign-in and checkout pages. The finding covers HUMAN’s customer sample, and a checkout request does not by itself establish a completed purchase.^[4]

These distinctions matter to both measurement and access decisions. A service name in a request needs supporting evidence before it can be trusted. A referral tag alone does not prove a visitor is human, and knowing the source of a request does not establish an intention to buy.^[4]

Retailers still need to restrict unwanted automation. They also need to account for legitimate shopping activity that may resemble it. Uncertain traffic should remain clearly marked as unknown rather than being forced into a category.

Product information now has another audience

Product photography, descriptions and reviews help shoppers decide whether an item fits their needs. The underlying facts also need to be clear enough for software to retrieve and compare, including the product’s identity, available sizes and colors, current price, stock status and shipping and return terms.^[7][8]

Adobe’s April 2026 analysis gave U.S. retail product pages an average AI Content Visibility score of 66%. The score measures content readability under Adobe’s proprietary methodology; it is not a product’s probability of being recommended.^[6]

Some improvements may involve the product pages and feeds a retailer already maintains. A product feed is a structured list of product details supplied to another service. Google says Merchant Center data powers its AI shopping experiences. Shopify says its Catalog structures product details for AI channels and keeps them updated.^[7][8]

For the backpack in the illustration, the price, size and availability agree across both views. In a real catalog, missing attributes or conflicting prices could leave a shopper or an assistant working with incomplete information. Keeping those details accurate gives both a better basis for comparison, without guaranteeing a place in an AI recommendation.

Retailers are still catching up on measurement

The measurement finding is the one that stands out to me. In PYMNTS Intelligence’s reporting on its 2026 merchant research, only 23% of merchants surveyed said they could clearly identify both AI-driven traffic and purchases. Another 21% could recognize the traffic but could not connect it to completed purchases. These are self-reported capabilities.^[9]

A retailer might recognize visits from an AI service and still struggle to connect them to orders. Another might see automated product requests without being able to verify their source. Those are different problems, and a single “AI traffic” number can obscure them.

This is the problem we are exploring at Cartograph AI. We call it the Agent Visibility Gap: the difference between AI’s involvement in shopping and the evidence a retailer has to understand it. The term describes our view of the problem, not an industry benchmark.

The useful starting point is the activity a retailer is authorized and able to observe: requests reaching its systems, the product information served, identifiable referrals, failed shopping steps and completed orders. Connecting those records requires evidence. Events occurring close together are not necessarily part of the same shopping journey.

Some parts will remain unknown. Private AI conversations should remain private, and unidentified traffic should not automatically be labeled AI. The absence of agent evidence does not prove that every visitor was human.

For a retail team, the value of better measurement is being able to decide what needs attention. An increase in AI-related traffic becomes more useful when the team can distinguish product-information requests from shopping activity and identify where something failed. That is work retailers can begin while the broader shopping experience continues to develop.

Sources

[1] PYMNTS Intelligence, commissioned by Visa Acceptance Solutions. The AI-Powered Shopper Has Arrived: Global Digital Shopping Index, 2026 edition. Figures 9 and 10, printed pages 22–25. March 2026 survey across the United States, Brazil and the United Arab Emirates. The online-purchase finding uses N=2,475; the product-research finding uses N=3,013 respondents who used a computer, mobile device or AI platform to research their latest purchase. Read the report.

[2] Kyle Risley, Shopify. “AI-referred shoppers convert better and spend more: What Shopify’s early data shows.” May 11, 2026. Q1 2026 platform data on referral growth, orders and product-page landings. Read the analysis.

[3] OpenAI. “Using shopping research in ChatGPT.” Help Center, accessed September 11, 2026. Product comparisons, buyer’s guides and the retailer privacy boundary. Read the documentation.

[4] HUMAN Security. The 2026 State of AI Traffic & Cyberthreat Benchmark Report. Figure 6 and the methodology section. The agentic AI category includes AI agents and agentic browsers; the findings reflect activity observed in a subset of HUMAN’s customers during 2025, not all internet traffic. Read the report.

[5] OpenAI. “Overview of OpenAI Crawlers.” Developer documentation, accessed September 11, 2026. Distinct search, training and user-triggered request roles. Read the documentation.

[6] Vivek Pandya, Adobe Digital Insights. “Adobe report: U.S. retailers see surge in AI traffic, but many websites are not entirely readable by machines.” April 16, 2026. Adobe’s AI Content Visibility methodology and U.S. retail page scores. Read the analysis.

[7] Google. “AI is changing retail. Here’s how businesses can keep up.” April 8, 2026. Merchant Center product information in AI shopping experiences. Read the article.

[8] Shopify. “Shopify Catalog and product discovery for agentic storefronts.” Help Center, accessed September 11, 2026. Structured product details and updates for AI channels. Read the documentation.

[9] PYMNTS. “How 23% of Merchants Captured Retail’s Next Agentic Commerce Advantage.” September 2, 2026. Publisher analysis of its Global Digital Shopping Index: Merchant Edition, commissioned by Visa Acceptance Solutions. The 23% and 21% figures are reported merchant capabilities. Read the analysis.

Methodology note: Consumer surveys, merchant self-reports, platform referrals, automated-traffic observations and a proprietary readability score measure different things. Their percentages should not be combined into an estimate of AI-commerce market share.