Will AI Shopping Agents Change How People Find and Buy Products?

Share

13 min

The short version

AI shopping agents already influence hundreds of billions in holiday sales, but checkout is another story. This evidence-backed look at agentic commerce in November 2026 covers Universal Cart, Instant Checkout's pullback, payment rails, merchant blocking, and concrete steps for shoppers and store owners.

Summarize with AI

ChatGPT

OpenAI

Gemini

Google

Perplexity

AI Search

Editorial Disclosure & Affiliate Notice

This content is published for informational and educational purposes only. It is not intended as a substitute for professional, legal, financial, or medical advice. AIUnpacker is funded through clearly labeled sponsored articles and reviews, relevant paid link placements, and affiliate commissions where disclosed. Commercial partnerships help fund the free editorial library. Sponsored work is labeled, and a payment does not change the verdict.

  • For educational purposes only. Nothing here should be taken as a guarantee, recommendation, or professional recommendation.
  • AI-assisted editing. Drafts are produced with AI assistance and get a human pass before they go up.
  • Opinions are our own. Also, we are not affiliated with most tools we cover unless explicitly stated.
  • Information may be outdated. Verify pricing, features, and policies directly with the vendor.
  • Last reviewed: . Published .

Read more on our About page, Terms and Editorial Policy.

Yes: AI shopping agents are already changing how people find products, and the 2026 holiday data proves it. But they’re changing how people buy far more slowly, because most shoppers still won’t hand an AI their wallet. Salesforce measured $262 billion of AI-influenced online sales in the 2025 holidays, yet Deloitte’s October 2026 survey found just 16% of consumers would let AI complete a purchase for them. Discovery has moved. The register, mostly, hasn’t.

What are AI shopping agents, and how do they work?

AI shopping agents are AI assistants that search products, compare options, and (in some cases) complete a purchase on your behalf. Three layers power them, mostly built between April 2025 and October 2026.

First, discovery: Amazon’s Rufus, ChatGPT’s shopping results, and Google’s AI Mode answering “what should I buy.” Second, transaction protocols. The Agentic Commerce Protocol (ACP) is an open standard from OpenAI and Stripe that passes order details to a merchant’s backend, where the merchant accepts or declines and processes payment. Google’s answer, announced at NRF on January 11, 2026, is the Universal Commerce Protocol (UCP), co-developed with Shopify, Etsy, Wayfair, Target, and Walmart to cover discovery through after-sales support.

The third layer is payments. The Agent Payments Protocol (AP2), announced by Google with roughly 60 payments partners in September 2025, creates cryptographically signed “mandates”: verifiable proof of what a user authorized an agent to buy. Mastercard’s Agent Pay (April 2025) tokenizes card credentials so agents never see your raw card number.

Agentic commerce is the umbrella term: buying where an AI agent, not a human clicking “add to cart,” drives some or all of the transaction.

What’s actually live in November 2026?

The biggest live change is Google’s Universal Cart, live in the US since October 15, 2026. it pulls items from multiple retailers into one cart, tracks price drops and restocks, and recommends payment methods based on card benefits. It works inside Google Search (including AI Mode) and the Gemini app, on Google’s 60+ billion-listing Shopping Graph, with Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and many Shopify merchants named as partners.

Checkout runs two ways: pay through Google Pay inside the cart, or move items to the merchant’s site. Either way the merchant stays the seller of record and handles support, per Google. The aisle isn’t the store.

DevelopmentStatus November 2026What it changesWho it affects
OpenAI Instant Checkout + ACP (launched Sept 29, 2025)Pulled back March 2026; refocused on discovery; ~12 merchants went liveChat stays a research aisle, not a register, for nowChatGPT users; small Etsy and Shopify sellers
Google Universal Cart + UCP (announced Jan 11, 2026; live Oct 15, 2026, US)Live in US Search, AI Mode, and Gemini; agentic checkout rolling out via UCPOne cross-retailer cart; price tracking; in-Google checkoutUS shoppers; merchants on Google Merchant Center
AP2 protocol (Sept 16, 2025; FIDO Alliance April 28, 2026)Vendor-neutral under FIDO; v0.2 adds autonomous paymentsCryptographic proof of what an agent was authorized to buyBanks, card networks, agent developers
Mastercard Agent Pay (April 29, 2025)Agentic Tokens in market; Agent Pay for Machines launched June 2026Tokenized credentials so agents never expose card numbersCardholders; merchants accepting agent payments
Visa Intelligent CommerceFirst live pilot with BILL/Billtrust (Oct 20, 2026); DBS consumer pilotsTrusted-agent payments reaching productionBusinesses first, consumers next
Perplexity Buy with Pro (November 2024)Live for US Pro subscribers (PayPal rails); reportedly the merchant of recordPlatform takes on sale liability, not just referralsPerplexity Pro subscribers; participating merchants

Amazon deserves its own row even though it blocks external agents: its Rufus assistant reached 300 million customers in 2025 and drove roughly $12 billion in incremental annualized sales, per Amazon figures reported by Modern Retail. Agent-assisted discovery at massive scale: entirely inside Amazon’s walls.

What happened to ChatGPT Instant Checkout?

OpenAI launched Instant Checkout on September 29, 2025 with US Etsy sellers and promised access for over a million Shopify merchants. results were organic and unsponsored, merchants kept control of fulfillment, returns, and the customer relationship, and only merchants paid a per-purchase fee.

It didn’t stick. Only about a dozen merchants ever went live; The Information reported the pullback on March 4, 2026, and OpenAI confirmed checkout would move to its Apps model while product discovery became the priority. Forrester called it a meaningful setback for the hype cycle.

That is the most honest market signal of the year. People happily use ChatGPT to figure out what to buy; they don’t yet want a chatbot holding their credit card. That’s not a category failure: it’s a scope correction, and Google is betting the other way with checkout inside Search.

Who’s building the payment rails underneath?

Agents change who (or what) clicks “buy,” so payments moved fast. In April 2026, Google donated AP2 to the FIDO Alliance jointly with Mastercard, which co-developed a companion standard called Verifiable Intent; AP2 v0.2 added autonomous payments. Neutral governance is how competitors adopt a protocol.

Mastercard shipped Insight Tokens in September 2025 and Agent Pay for Machines in June 2026, while Visa went live on October 20, 2026 with the first trusted-agent payment pilot alongside BILL and Billtrust: a B2B start, but a live one.

Why does this matter to you? Mandates (those signed records of what you authorized) stop an agent from buying 500 units of something you wanted one of. It’s also why any agent asking for your raw card number is a red flag.

How big will this get, and how fast?

Forecasts disagree wildly. Juniper Research projects agentic commerce spend growing from about $8 billion in 2026 to $1.5 trillion globally by 2030. eMarketer, tracking AI-platform-driven US e-commerce only, forecasts about $144 billion by 2029 (8.8% of US e-commerce) and 15.8% by 2030. The range depends on how much “agentic” behavior you count.

The behavioral data says more. Worldpay’s Agentic Commerce Report 2026 found 74% of consumers would let an AI agent compare prices and 69% would let it research products, but only 21% would let an agent complete the purchase, and 59% still want to approve every transaction. Deloitte’s 2026 holiday survey, released October 15, 2026, agrees: 39% of consumers use generative AI as their primary tool for at least one holiday shopping task, but only 16% would permit AI to purchase on their behalf.

The whole story of agentic shopping in two numbers: 74% of consumers will let an AI agent compare prices, but only 21% will let it complete the purchase. (Worldpay Agentic Commerce Report 2026)

Meanwhile, discovery is compounding. Adobe Analytics measured AI-driven traffic to US retail sites up 127% year over year in August 2026, and forecasts roughly 130% growth through the holidays. Sharper still: in Q1 2026, AI-referred traffic converted 42% better than regular visitors: a reversal from being 38% worse a year earlier.

NRF forecasts holiday sales up 4.2% to $1.013 trillion; Adobe forecasts a record $275.1 billion in US online holiday sales, up 6.7%. Salesforce’s 2025 data showed AI and agents influenced about 20% of global online sales ($262 billion) and its 2026 predictions, reported by Quartz, expect 20% of holiday e-commerce traffic from AI chat agents. The money already flows through AI-assisted paths; it mostly completes on merchant sites.

What’s holding agentic shopping back?

Three things: retailer blocking, unsettled courts, and consumers who don’t trust agents with money.

The blocking problem is bigger than it looks because most of it is accidental. An August 2026 audit of 51 large US retailers’ published access rules, by identity firm Billions and reported by CNBC on November 1, found 37 of 51 block AI shopping agents: mostly through legacy anti-bot rules written before agents existed. Only five, including Amazon, explicitly restrict agents; only 11 have a deliberate position. Amazon and Kohl’s block on purpose; Walmart has opened up, saying agents see the same prices humans do.

The legal fight is Amazon v. Perplexity. Amazon sued in November 2025, arguing the Comet browser agent violated the Computer Fraud and Abuse Act by shopping behind users’ logins. A judge granted Amazon a preliminary injunction in March 2026, but on August 4, 2026, the Ninth Circuit vacated it, holding that an AI agent is a “tool, not a person” under the CFAA, so access liability sits with the users who direct the agent. The case continues. This isn’t legal advice. The “can agents legally access my store?” question is still live in court.

Then there’s the trust gap. Worldpay found the biggest worries are privacy, errors, and loss of control: exactly why 59% want per-purchase approval. Consumer behavior changes slower than infrastructure.

What stays human?

More than the hype suggests. Agents take over the tedious parts (comparison, spec-checking, price monitoring, restock alerts) while humans keep taste, trust, and accountability. Nobody wants an agent choosing a gift for their mother based on a feed file.

Brands shouldn’t panic, either: Adobe’s conversion data shows AI-referred shoppers are now more valuable than average, making discovery agents a customer-acquisition channel, not a brand replacement. The stores that win will be ones agents can read easily (clean data, clear prices) that still give humans a reason to return directly.

What shoppers should do now

  1. Use agents for research; keep checkout manual for now. Let AI do the comparison grind; you click the final button. The Worldpay split is a sensible default.
  2. Try Universal Cart if you’re in the US. Useful for cross-retailer price tracking before the holidays; moving items to the merchant’s site keeps your usual protections.
  3. Never give an agent your raw card number. Legit flows use tokens or wallets like Google Pay; a chatbot asking for your card digits is a scam pattern.
  4. Cross-check recommendations. AI results are only as good as merchant feed data, and stale prices happen. Verify on the merchant’s site before anything expensive.
  5. Give agents constraints, not just a query. Budget, size, deadline, dealbreakers. “Best running shoes” gets generic answers; “under $120, wide fit, arrives before December 20” gets useful ones.

What merchants should do now

  1. Fix your product feed first. Complete, current Google Merchant Center data is table stakes for Universal Cart, and OpenAI’s ACP feed spec (developers.openai.com/agentic-commerce) defines the fields agents expect: IDs, titles, prices, availability, shipping, returns.
  2. Add structured data and keep it consistent. Schema.org markup matching your feed and APIs. Adyen’s 2026 guide notes: agents filter and recommend on machine-readable attributes; inconsistent data makes you invisible or wrong.
  3. Audit your bot rules today. Remember, 37 of 51 large retailers block agents mostly by accident. Check robots.txt and your CDN’s bot management so a 2021-era rule isn’t turning away 2026 customers.
  4. Pick your UCP level. Agentic checkout in AI Mode requires UCP integration via Merchant Center; Google’s docs lay out multiple access levels. Universal Cart is the low-effort entry; full agentic checkout is the deep one.
  5. Settle merchant-of-record questions before signing. Merchant of record is the entity legally liable for the sale: taxes, fraud, chargebacks. OpenAI’s ACP and Google’s Universal Cart keep merchants in that seat; Perplexity reportedly takes it for Buy with Pro. Your seat changes your tax exposure: run agreements past your payments provider and, for anything material, a lawyer.
  6. Measure AI-referred traffic separately. Adobe’s numbers show it converts better than average; lumping it into “organic” underinvests a channel growing 127% a year.

The bottom line

Will AI shopping agents change how people find and buy products? They already have: the finding half. One in five dollars of last year’s holiday e-commerce was AI-influenced, AI-referred traffic is growing triple digits, and Amazon built a 300-million-user assistant most people never call an “agent.”

The buying half changes on a years-long timeline: trust, law, and retailer incentives all lag infrastructure. Shoppers: use agents to shop, keep your card to yourself for now. Merchants: the unglamorous work (feeds, structured data, bot rules) is where this season’s money is. And hold trillion-dollar forecasts loosely: the gap between Juniper’s $1.5 trillion and eMarketer’s 8.8% is the gap between what’s possible and what people will actually permit.

FAQ

What are AI shopping agents?

AI shopping agents are AI assistants that find, compare, and sometimes buy products on your behalf. They combine product feeds, shopping graphs, and payment protocols. Today they’re strongest at research and price tracking (Google’s Universal Cart and Amazon’s Rufus are the biggest examples) while autonomous checkout remains early and optional.

Can AI agents actually buy things for you today?

Yes, in limited cases. Google’s Universal Cart lets US shoppers pay via Google Pay inside Search, AI Mode, and the Gemini app, with the merchant staying the seller of record. Perplexity’s Buy with Pro checks out for US Pro subscribers, and Amazon’s Rufus has agentic features like Buy for Me. Most shoppers still approve each purchase themselves.

What is Google Universal Cart?

Universal Cart is Google’s cross-retailer shopping cart, launched in the US on October 15, 2026. You can save items from retailers like Nike, Target, Walmart, Wayfair, Sephora, and Ulta Beauty in one place, track price drops and restocks, and check out via Google Pay or on the merchant’s own site, powered by the Universal Commerce Protocol.

What happened to ChatGPT Instant Checkout?

OpenAI launched Instant Checkout on September 29, 2025, starting with US Etsy sellers and promising Shopify support via the Agentic Commerce Protocol. Only about a dozen merchants went live. In March 2026 OpenAI pulled back, moving checkout to its Apps model and refocusing ChatGPT on product discovery: shoppers buy on merchant sites.

Are AI shopping agents safe for payments?

They can be, when credentials stay tokenized. Mastercard’s Agentic Tokens and the AP2 protocol, now governed by the FIDO Alliance, cryptographically record what an agent was authorized to buy. Worldpay found 59% of consumers still want per-purchase approval: sensible today. Never type raw card numbers into a chatbot, and review agent-initiated charges.

What is agentic commerce?

Agentic commerce is commerce in which AI agents, not just human shoppers, drive discovery, comparison, and transactions. Juniper Research estimates about $8 billion in 2026, forecasting $1.5 trillion globally by 2030. The plumbing (ACP, UCP, AP2, and card-network token programs) was mostly built between April 2025 and October 2026.

How will AI shopping agents affect online merchants?

Merchants face a discovery shift first: Adobe measured AI-driven traffic to US retail sites up 127% year over year in August 2026, and those visitors convert better than average. You need clean product feeds, structured data, and consistent pricing across surfaces. Also audit bot rules, 37 of 51 audited US retailers block agents, mostly by accident.

Which retailers block AI shopping agents?

An August 2026 audit of 51 large US retailers reported by CNBC found 37 blocking AI shopping agents, mostly through legacy anti-bot rules rather than deliberate policy. Only five, including Amazon, explicitly restrict agents; Amazon and Kohl’s are deliberate blockers, while Walmart has opened up. Only 11 retailers had a stated AI-agent position.

How do I get better results when shopping with AI?

Be specific: give the agent your budget, size, use case, and dealbreakers, then ask for a comparison with prices and return policies. Cross-check recommendations against reviews and the merchant’s site before buying, since AI results can be stale or shaped by whichever feed data is cleanest. Keep checkout manual until you trust the flow.

What should merchants do first?

Start with data, not gadgets. Get your Google Merchant Center feed complete and current, add schema.org structured data to product pages, and make prices, availability, and shipping match everywhere. Audit your bot rules so agents aren’t blocked by accident, then review the OpenAI ACP feed spec. Merchant-of-record and tax questions deserve professional advice.

Sources

Weekly digest

Get our weekly AI digest

The latest AI tools, prompts, and insights — short, useful, every week.

No spam. Unsubscribe anytime.

AIUnpacker Editorial

Articles are written for a specific question. Sponsored work is labeled.

About us ·More articles