For most of the last two years, "AI shopping" was a forward-looking narrative more than a measurable channel. That has changed. In 2026, it is a documented shift in consumer behavior, a live economic battle between platforms and retailers, and an infrastructure build-out involving nearly every major payments network. This is a look at the data behind that shift — where it stands today, what's driving it, and what it means for brands trying to be found.

The Data Behind AI Shopping in 2026

1. The scale of the shift

Consumer research behavior has moved faster than most channel transitions on record. According to L.E.K. Consulting, 46% of AI users now begin product research directly on a standalone AI platform — ChatGPT, Gemini, Perplexity, or Claude — up from just 25% in 2024, while the share starting on traditional search fell from 43% to 24% over the same period.

Adoption is broad, not niche. IBM's Institute for Business Value found 45% of consumers already use generative AI for some part of their buying journey, and per L.E.K., usage skews heavily by category: 71% of consumers use AI for travel research, 65% for consumer electronics, and 62% for financial products, with no tracked category falling below roughly half. MarTech's 2026 roundup puts U.S. figures similarly high: 43% of online shoppers used an AI assistant for product research in the prior 90 days, and among them, 20% used AI on their most recent purchase over $50.

The volume flowing through this channel is no longer marginal. Salesforce and Publicis Sapient estimate ChatGPT alone processes roughly 350 million shopping-related queries per week, against a base of more than 900 million weekly active users — making it, by user count, the largest AI shopping audience in existence. Adobe Analytics recorded 393% year-over-year growth in AI-referred traffic to U.S. retailers in Q1 2026, following 805% year-over-year growth in AI-driven retail traffic over Black Friday 2025.

2. Not all AI traffic converts the same way

Traffic volume and purchase intent are not the same thing, and the platform-level data makes that concrete. Per Elogic Commerce's 2026 analysis, Claude posts the highest conversion rate among major LLM platforms at roughly 16.8%, despite holding a comparatively small 2–10% share of generative AI referral traffic. ChatGPT converts in the 14–16% range, Perplexity around 10.5%, and Gemini trails at roughly 3%.

The practical implication: optimizing purely for which platform sends the most visits can be the wrong optimization target. A platform with a smaller audience can still be delivering a brand's highest-intent customers.

3. From research assistant to transaction layer

The more consequential shift in 2026 isn't that AI helps people decide — it's that AI is increasingly completing the purchase itself. Perplexity has rolled out a free conversational shopping experience with instant checkout powered by PayPal. OpenAI has built comparative shopping research into ChatGPT using a fine-tuned model with real-time retrieval across retail sites. Google has launched agentic checkout inside both Search's AI Mode and Gemini, with "buy for me" functionality already live with select U.S. retailers (commercetools, 2026). Walmart's Sparky assistant is now embedded not just in Walmart's own app but inside ChatGPT and Gemini directly (Genrise, 2026).

The dollar figures attached to this are no longer speculative. Salesforce reported that AI and agentic commerce influenced an estimated $67 billion in global sales during Cyber Week 2025 alone — about 20% of all orders placed worldwide that week — with traffic from third-party AI agent channels tripling year-over-year and converting at multiples of the rate of social media referrals (Genrise, 2026).

4. The payments infrastructure race

Making autonomous AI purchases actually work has triggered a genuine infrastructure buildout across the payments industry, and as of mid-2026 the landscape is still fragmented across at least six competing or overlapping protocols (Tom Wang, 2026):

Google's AP2 (Agent Payments Protocol) — an open, vendor-neutral standard announced in September 2025 with more than 60 launch partners including Mastercard, PayPal, Coinbase, American Express, and Salesforce. AP2 uses cryptographically signed "Mandates" to prove a transaction was actually authorized by the user, not an agent acting on its own (Google Cloud Blog, 2025). As of May 2026, governance of AP2 was handed to the FIDO Alliance for open, multi-stakeholder development (agentpaymentsprotocol.eu, 2026).
Visa's Trusted Agent Protocol (TAP) — Visa's network-specific implementation, commercially launched in Q1 2026 with more than 100 enrolled partners; Fiserv became the first major processor to adopt it at scale in January 2026 (Paz.ai, 2026).
Mastercard's Agent Pay — announced April 2025, using tokenized "Agentic Tokens" scoped to a specific agent, merchant, and consent policy, so a model can complete checkout without ever holding the underlying card number (eco.com, 2026).
Stripe's Shared Payment Tokens and Machine Payments Protocol, and OpenAI's Agentic Commerce Protocol (ACP) built with Stripe, round out the stack alongside Google's separate Universal Commerce Protocol (UCP), which counts Walmart, Target, Shopify, and Etsy among its partners (Opascope, 2026).

The overlapping standards mean most merchants will end up touching several of these indirectly through their payment processor rather than integrating each one directly (Paz.ai, 2026). The clearer signal for brands, per a 2026 fintech developer analysis, is that structured, machine-readable product data — specs, availability, return policy, schema.org markup — is becoming close to mandatory: every major protocol assumes it, and its absence effectively makes a product invisible to an agent (Tom Wang, 2026).

5. Not every retailer is opening the door

The buildout is not happening with universal cooperation. Amazon has blocked OpenAI's ChatGPT-User and OAI-SearchBot crawlers via robots.txt, meaning Amazon listings cannot currently surface in real-time ChatGPT shopping results — a defensive move widely read as protecting Amazon's roughly $4 billion advertising business (Elogic Commerce, 2026). In March 2026, a federal judge in San Francisco went further, issuing a preliminary injunction blocking Perplexity's Comet browser from making purchases on Amazon's platform, after Amazon argued the agent misrepresented automated traffic as human activity; Perplexity countered that its technology increased transaction volume (Elogic Commerce, 2026). It's the first significant legal precedent testing how far a retailer can go to resist agent-driven purchasing on its own platform.

The structural consequence, per Elogic Commerce's analysis: a brand selling through both Amazon and its own Shopify storefront may be fully visible to ChatGPT through the Shopify listing while its Amazon listing remains invisible to that same AI — meaning where a brand sells now partly determines whether AI can recommend it at all.

6. The trust gap that hasn't closed yet

Growth in usage hasn't been matched by growth in trust. Stellagent's 2026 research review cites Bain & Company findings that consumers trust a retailer's own on-site AI agent roughly three times more than a third-party agent like ChatGPT or Perplexity, and notes that 83% of consumers still say they don't fully trust AI-driven purchases. Attribution is also becoming genuinely harder to measure: Opascope's 2026 guide compares the current state to the early, murky years of podcast attribution, but notes agentic checkout removes even the partial signals podcasts had (promo codes, delayed measurement windows) — a one-click AI purchase can complete in seconds with almost no identity-capture step for the brand to observe.

7. It isn't only a consumer story

The B2B side is moving in parallel. Forrester projects that 20% of B2B sellers will face agent-led quote negotiations by the end of 2026 (Opascope, 2026), and Adobe's enterprise research found 78% of organizations expect agentic AI to handle a majority of customer interactions within 18 months — split across customer service (78%), post-purchase support (70%), and sales conversations (69%). The same research found a real capability gap behind that expectation: only 51% of organizations currently have the cloud infrastructure to support agentic AI, compared with 89% that already have generative AI infrastructure in place (Harro, 2026).

What this means in practice

The throughline across all of this data is the same: AI platforms are no longer a peripheral discovery channel to monitor occasionally. They're an active distribution and transaction layer with their own economics, their own infrastructure race, and their own legal fights over who controls access to the customer. Whether a brand shows up accurately — and where it's even eligible to be found at all — is now a function of decisions made well upstream of any single conversation: which platforms crawl your site, which payment protocols your processor supports, and whether your own product data is structured well enough for a model to use it in the first place.

References

L.E.K. Consulting, cited in "The AI shopping stats 2026," MarTech / Harro, 2026
IBM Institute for Business Value, "AI in the buying journey," Jan 2026
Adobe Analytics, Q1 2026 and Black Friday 2025 AI-referred traffic data
Adobe / Harro, enterprise agentic AI readiness research, 2026
Salesforce + Publicis Sapient, ChatGPT shopping query volume, May 2026
Elogic Commerce, "ChatGPT Commerce & Agentic Shopping Statistics 2026"
commercetools, "7 AI Trends Shaping Agentic Commerce in 2026," April 2026
Genrise, "AI Shopping Assistants in 2026," June 2026
Google Cloud Blog, "Announcing Agent Payments Protocol (AP2)," Sept 2025
Paz.ai, "Visa Trusted Agent Protocol: 2026 Guide" and "Agent Payments Protocol (AP2): What It Is in 2026"
eco.com Support, "What Is Mastercard Agent Pay?," June 2026
Tom Wang, "Google AP2 Completes the Agentic Payment Stack," April 2026
agentpaymentsprotocol.eu, "AP2 — Agentic payments, trust, and the EU perspective," May 2026
Stellagent, "AI Shopping Agent Showdown," April 2026
Opascope, "AI Shopping Assistant Guide 2026," April 2026