Google is testing checkout inside AI search. What it means for Amazon, Shopify and every DTC brand

TechCrunch reported on 26 September that Google was testing a new shopping path for some users in India: selected Flipkart listings in Gemini and Google Search's AI Mode displayed a Buy button that opened a Flipkart-branded checkout without taking the shopper out of the AI experience.

The reported test covered a limited selection of smartphones, electronics and mobile accessories. TechCrunch also reported that a source expected a broader rollout in October, but Google has not publicly confirmed that timetable, the test allocation, the number of eligible products or any transaction results.

The significance is not that Google has suddenly solved agentic commerce. It is that product discovery, comparison and a merchant-branded purchase path can now sit within one AI interface. That shortens the distance between recommendation and transaction, while raising a harder question for brands: who controls the alternatives a shopper sees before the Buy button appears?

<h2>What makes the Flipkart experiment different?</h2>

The reported experience combines three commercially important moments. A shopper can ask Gemini or AI Mode about products, Google's AI can present alternatives, and selected Flipkart listings can expose an immediate purchase path.

In the experience seen by TechCrunch, Amazon listings appeared beside Flipkart products, but the direct-purchase option was available only for selected Flipkart listings. That observation does not show that Google favoured Flipkart in its recommendations. It shows that recommendation and transaction availability are becoming separate competitive advantages.

The implementation boundary matters. Google has not disclosed what technology powered the reported Flipkart flow, who processed payment, who was merchant of record or whether the Universal Commerce Protocol powered it. TechCrunch described a Flipkart-branded checkout and said the underlying technology was unclear. Google's separately documented Universal Cart is currently U.S.-only, so it should not be used as a label for the India test.

The safest conclusion is narrower: Google is experimenting with a way to connect AI-assisted product discovery to a merchant-branded transaction path.

<h2>Three different approaches to the future of AI shopping</h2>

<h3>Google: extend AI search into commerce</h3>

Google's strategy extends beyond the reported Flipkart experiment. In January, it introduced the Universal Commerce Protocol, or UCP, as an open standard through which AI agents, merchants and payment providers can coordinate discovery, buying and post-purchase support. Its September merchant update said UCP-enabled direct checkout supported hundreds of thousands of brands and retailers across Google services.

Those announcements provide adjacent infrastructure context, not proof of the Flipkart test's technical stack. Google's current Merchant Center documentation describes UCP onboarding as controlled and U.S.-first. The company also distinguishes native checkout on Google from cart transfer to a merchant site.

The strategic opportunity for Google is to extend its role from identifying products into later stages of the journey. A shopper who previously used Google to find a product might then visit several websites to compare alternatives and complete a purchase. AI interfaces can increasingly consolidate those activities.

For retailers, participation offers access to shoppers who are already researching products. It also creates a dependency on the interface that determines which alternatives enter the conversation.

<h3>Amazon: build the shopping assistant and control access</h3>

Amazon is extending its shopping ecosystem beyond products sold directly on its marketplace. The company reported in March that Shop Direct covered more than 100 million products from more than 400,000 external merchants. It also said that tens of millions of products were eligible for Buy for Me and that Shop Direct had generated millions of referrals to external stores.

Those are company-reported catalogue and referral counts, not audited sales or conversion results. Shop Direct and Buy for Me also describe different transaction paths. Shop Direct sends the shopper to the merchant site. For eligible Buy for Me products, Amazon's AI can complete the external merchant checkout after the shopper confirms details.

Amazon's current merchant and consumer FAQs say the listed merchant remains seller of record and manages fulfilment, returns and customer service. A Buy for Me purchase is therefore not an Amazon retail sale in the conventional sense, even though Amazon facilitates the experience.

Amazon is also enforcing control over third-party agents that attempt to shop its marketplace. GeekWire reported on 20 September that Amazon had blocked shopping attempts by Meta's Muse under an unauthorised-agent warning. Amazon attributed the decision to its access rules and raised transparency, credential and security concerns. Those technical claims remain Amazon's allegations, not independently verified findings.

The tension is clear. Amazon wants its own interface to help customers shop beyond Amazon while retaining contractual and technical control over outside agents that use Amazon.

<h3>Shopify: supply commerce infrastructure across competing AI platforms</h3>

Shopify is pursuing a different position. Rather than requiring every shopping journey to begin in a Shopify-owned assistant, it is connecting merchants to multiple AI environments through Shopify Catalog, Agentic Storefronts, checkout and order infrastructure.

Its documentation lists ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta as supported third-party channels, with material differences between them. ChatGPT is documented as a discovery and referral surface for eligible U.S. buyers, with purchase completed in the merchant's online-store checkout. Google AI Mode and Gemini availability and direct checkout are rolling out for eligible U.S.-based stores selling to U.S. customers. Copilot supports direct checkout for eligible U.S. customers. Meta documents direct checkout for eligible merchants and customers in the United States, Canada and Mexico, with Muse qualified as available when available.

Shopify says the merchant remains seller or merchant of record and retains fulfilment, returns and customer-service responsibilities. Merchants can manage channel access and, where supported, direct checkout.

For DTC brands, this reduces the need to build a separate commerce integration for every AI platform. It does not give Shopify control over every recommendation. Shopify explicitly says participating AI channels may re-rank products using their own logic and that its search preview is directional rather than a prediction of what shoppers will see.

<h2>The commercial stakes: AI-referred shoppers are behaving differently</h2>

Adobe Digital Insights analysed more than one trillion visits to U.S. retail sites and more than 100 million SKUs. For July 2026, it reported that AI-driven retail traffic was 62% higher year over year.

Adobe also reported the following relative differences between AI-referred and non-AI U.S. retail visits:

<div class='overflow-x-auto not-prose my-8'><table class='w-full min-w-[760px] border-collapse text-left font-sans text-sm'><thead><tr class='bg-[#15303C] text-[#F7FAF8]'><th class='px-4 py-3'>July 2026 measure</th><th class='px-4 py-3'>Reported difference</th><th class='px-4 py-3'>Boundary</th></tr></thead><tbody><tr class='border-b border-[#CAD7D3] bg-[#EAF1EF]'><td class='px-4 py-3 font-semibold'>Conversion rate</td><td class='px-4 py-3'>60% higher</td><td class='px-4 py-3'>Relative rate, not percentage points</td></tr><tr class='border-b border-[#CAD7D3]'><td class='px-4 py-3 font-semibold'>Revenue per visit</td><td class='px-4 py-3'>53% higher</td><td class='px-4 py-3'>Observed association, not causal lift</td></tr><tr class='border-b border-[#CAD7D3] bg-[#EAF1EF]'><td class='px-4 py-3 font-semibold'>Add-to-cart rate</td><td class='px-4 py-3'>28% higher</td><td class='px-4 py-3'>Absolute rates undisclosed</td></tr><tr><td class='px-4 py-3 font-semibold'>Time on site</td><td class='px-4 py-3'>59% higher</td><td class='px-4 py-3'>Why engagement differs was not isolated</td></tr></tbody></table></div>

These are relative observational differences, not absolute rates or causal effects. Adobe did not establish that AI referrals caused the higher performance. One plausible interpretation is that some AI-referred shoppers arrive after narrowing their alternatives, but the public report does not isolate that mechanism.

The reported Flipkart experiment introduces a second measurement problem. Adobe measures consumers arriving at retail websites. Transactions completed inside an AI interface may not generate the conventional website visits on which many marketing attribution systems depend.

DTC brands therefore face two distinct questions: how AI influences which products enter the choice set, and how that influence connects to orders regardless of where checkout occurs.

<h2>A new competitive advantage: product information AI can use</h2>

Commerce infrastructure is only one part of the problem. As AI assistants become more involved in purchasing decisions, brands need to understand how those systems evaluate products.

Google reported that, in testing with lululemon, retailer-supplied conversational attributes were incorporated in 50% of relevant product recommendations in AI Mode. This is a product-information incorporation result. It is not a 50% increase in recommendations, recommendation position or sales, and Google did not disclose the test's sample size, dates or control condition.

AI shopping assistants need information that connects products with customer requirements. A shopper might ask for running shoes suited to a particular surface, skincare appropriate for specific needs or an appliance meeting several constraints. Accurate specifications remain necessary, but assistants may also need to understand benefits, suitability, limitations and supporting evidence.

Brand-owned product pages are not the entire decision environment. AI systems may also encounter customer reviews, expert assessments, independent comparisons, retailer listings and competing claims. Product information establishes what a brand says about itself. The surrounding evidence environment can reinforce, qualify or challenge those claims.

<h2>Five things every DTC brand should do now</h2>

<h3>1. Make products accessible to relevant AI shopping platforms</h3>

Start with existing commerce infrastructure. Shopify merchants should review Agentic Storefronts, confirm eligibility and check which AI channels and checkout options are enabled. Brands operating outside Shopify should maintain reliable product feeds, structured product information and relevant merchant integrations.

Amazon's Shop Direct may offer another route for participating external merchants. Google's Merchant Center and UCP materials provide additional paths, but access remains eligibility- and market-dependent. The objective is not to enable purchasing everywhere. It is to know where products are discoverable, recommendation-ready, transaction-ready and commercially relevant.

<h3>2. Improve the information AI uses to evaluate products</h3>

Review product information from the perspective of a customer asking an AI assistant for advice. Can the system identify the intended customer, relevant benefits, meaningful differences from alternatives, applicable constraints and substantiation? Specific, verifiable information is more useful than generic promotional language.

Google's lululemon finding gives this work urgency. Shopify's product-discovery guidance likewise emphasises detailed descriptions, images, variants, reviews and store policies. Brands should also test whether their most important claims are consistently supported across credible third-party evidence, not only on their own sites.

<h3>3. Measure awareness, consideration, recommendation and transaction separately</h3>

Being mentioned is not equivalent to being considered. Consideration is not equivalent to an affirmative recommendation. Recommendation does not guarantee an available checkout path. Checkout availability does not prove a completed or incremental sale.

<div class='overflow-x-auto not-prose my-8'><table class='w-full min-w-[820px] border-collapse text-left font-sans text-sm'><thead><tr class='bg-[#15303C] text-[#F7FAF8]'><th class='px-4 py-3'>Stage</th><th class='px-4 py-3'>Commercial question</th><th class='px-4 py-3'>Example measure</th></tr></thead><tbody><tr class='border-b border-[#CAD7D3] bg-[#EAF1EF]'><td class='px-4 py-3 font-semibold'>Awareness</td><td class='px-4 py-3'>Does the AI identify the brand?</td><td class='px-4 py-3'>Visibility and brand appearances</td></tr><tr class='border-b border-[#CAD7D3]'><td class='px-4 py-3 font-semibold'>Consideration</td><td class='px-4 py-3'>Does it present the brand as a relevant alternative?</td><td class='px-4 py-3'>Competitive inclusion and associations</td></tr><tr class='border-b border-[#CAD7D3] bg-[#EAF1EF]'><td class='px-4 py-3 font-semibold'>Recommendation</td><td class='px-4 py-3'>Does it actively recommend the product?</td><td class='px-4 py-3'>Recommendation rate and position</td></tr><tr class='border-b border-[#CAD7D3]'><td class='px-4 py-3 font-semibold'>Transaction availability</td><td class='px-4 py-3'>Can the shopper start an appropriate purchase path?</td><td class='px-4 py-3'>Checkout coverage by platform and market</td></tr><tr><td class='px-4 py-3 font-semibold'>Commercial outcome</td><td class='px-4 py-3'>Did the journey produce incremental value?</td><td class='px-4 py-3'>Orders, net revenue, margin, cancellations and returns</td></tr></tbody></table></div>

These are distinct outcomes rather than guaranteed stages in every journey. A brand can be frequently recommended yet offer integrated checkout in few environments. It can have broad checkout availability yet rarely be recommended.

<h3>4. Test embedded checkout while protecting customer economics</h3>

Integrated checkout may reduce friction, but it can also change attribution, data access and customer economics. Shopify's Meta integration, for example, allows eligible merchants to support direct checkout while retaining fulfilment and customer-service responsibilities. Shopify says conventional third-party analytics pixels do not fire when purchases complete directly within Meta, and some checkout customisations are unavailable.

Brands should evaluate incremental orders, contribution margin, acquisition cost, customer data available, cancellations, returns and repeat purchasing. Where feasible, controlled tests should distinguish additional demand from transactions that might otherwise have occurred on the brand's website.

<h3>5. Avoid dependence on one AI shopping ecosystem</h3>

Amazon's dispute with Meta illustrates the risk of relying on a single platform. Technical integrations, commercial agreements and access policies can change. Brands should maintain reliable product data, credible supporting evidence, direct customer relationships and purchasing capabilities that remain useful across multiple AI environments.

Shopify's multi-platform approach can reduce integration work, but it does not remove differences in how individual assistants evaluate and recommend products. AI shopping platforms are additional routes to customers, not substitutes for independent brand competitiveness.

<h2>Connect AI recommendations with human purchasing decisions</h2>

AI systems increasingly participate in decisions that consumers once made through traditional searching and comparison. The criteria an assistant emphasises may not match the attributes that drive human preference.

A product could be visible and recommended in AI yet underperform commercially if the system highlights benefits consumers value less. A brand with strong consumer equity could also be poorly represented in AI answers.

Dig Insights has introduced its System 0 framework to examine AI's role in decision-making. Its public page reports that 65% of an undisclosed quick study preferred to use AI in some way when making a decision and 50% liked AI suggestions for purchase options. The page does not disclose the sample, market, field dates, weighting or questionnaire, and the figures do not measure completed purchases.

The useful idea is therefore a research agenda rather than a proven causal result. Combining AI visibility and recommendation tracking with consumer research and transaction data could help DTC businesses distinguish weak visibility, poor AI competitive positioning, checkout friction and differences between AI recommendations and human demand.

Commercial neutrality is another open question. The reported Flipkart experiment does not show that integrated checkout changes recommendation behaviour. Whether transaction availability affects organic inclusion or position is a separate, testable question.

<h2>AEO Updates Takeaway</h2>

The reported Flipkart test offers an early example of how AI shopping could change the relationship between brands, retailers and consumers. Amazon is extending its own external-shopping layer while enforcing rules for outside agents. Shopify is supplying product and transaction infrastructure across several AI platforms. Google is moving product discovery and purchasing closer together.

The company that processes a transaction will not necessarily control the AI interaction that determines which product is chosen. DTC brands therefore need two capabilities: evidence that helps AI systems evaluate and recommend their products, and commerce infrastructure that lets customers purchase through appropriate channels.

A brand can be widely recommended yet difficult to purchase. It can offer frictionless checkout yet fail to enter the recommendation set. It can perform well on both without proving incremental sales.

<blockquote><strong>The next competitive advantage in e-commerce will not simply be getting customers to visit a website. It will be earning a place in AI-mediated consideration and making the product purchasable when that recommendation occurs.</strong></blockquote>

<h3>References</h3>

[1] <a href='https://techcrunch.com/2026/09/26/google-tests-buying-from-walmart-owned-flipkart-through-gemini-and-ai-mode-in-india/' target='_blank' rel='noopener noreferrer'>Ivan Mehta, Google tests buying from Walmart-owned Flipkart through Gemini and AI Mode in India</a>, TechCrunch, 26 September 2026.

[2] <a href='https://blog.google/products/ads-commerce/agentic-commerce-ai-tools-protocol-retailers-platforms/' target='_blank' rel='noopener noreferrer'>Google, New tech and tools for retailers to succeed in an agentic shopping era</a>, 11 January 2026.

[3] <a href='https://blog.google/products-and-platforms/products/shopping/google-shopping-updates-holiday-shopping/' target='_blank' rel='noopener noreferrer'>Google, Boost your holiday sales with these agentic commerce updates</a>, 16 September 2026.

[4] <a href='https://www.aboutamazon.com/news/retail/amazon-shop-direct-external-stores' target='_blank' rel='noopener noreferrer'>Amazon, Amazon introduces feeds to make it easier for merchants to reach more customers through AI-powered Shop Direct</a>, 11 March 2026.

[5] <a href='https://www.geekwire.com/2026/amazon-blocks-metas-muse-ai-assistant-in-new-standoff-over-agentic-shopping/' target='_blank' rel='noopener noreferrer'>Todd Bishop, Amazon blocks Meta's Muse AI assistant in new standoff over agentic shopping</a>, GeekWire, 20 September 2026.

[6] <a href='https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts' target='_blank' rel='noopener noreferrer'>Shopify, Agentic Storefronts documentation</a>, accessed 27 September 2026.

[7] <a href='https://business.adobe.com/assets/pdfs/resources/sdk/ai-traffic-trends-report-august-2026/ai-traffic-trends-report-2026-08-19.pdf' target='_blank' rel='noopener noreferrer'>Adobe Digital Insights, AI Traffic Trends Report, August 2026</a>, 19 August 2026.

[8] <a href='https://diginsights.com/resources/system-0/' target='_blank' rel='noopener noreferrer'>Dig Insights, Introducing: System 0</a>, publication date not disclosed.

<p class='text-sm italic'>The source families are not pooled because they measure different units: observed product surfaces, platform capabilities, catalogue scale, website traffic, stated preferences and editorial measurement stages.</p>