AI Traffic Converts 3x Better, Yet Agentic Checkout Is Still Stalling
AI-referred traffic converts at least three times better than non-AI traffic, but in-chat checkout is not moving. Here is the real brake, which is first-party data ownership rather than technology, and what e-commerce operators should prepare now.
Key Takeaways
- AI-driven product discovery keeps growing. India's Flash AI reports that non-AI traffic converts at 6-10%, while queries arriving through ChatGPT, Gemini and similar assistants convert at least three times higher.
- The reason checkout alone is stuck is not technology. Retailers still have no clarity on how much access they get to first-party data such as purchase records and contact details, and that is the biggest brake.
- The workable line for now is to open discovery to AI while keeping checkout on your own stack. Investing in product feeds and AI-readable product information comes first, and delaying in-chat checkout costs you little.
Discovery and checkout have split apart
AI's ecommerce ambitions are hitting a reality check as trust and changing buying habits are holding back chatbot-led commerce
m.economictimes.comThe Economic Times piece published on August 18, 2026, frames OpenAI's retreat from Instant Checkout not as an isolated failure but as a structure common to AI commerce as a whole. The Indian founders and executives interviewed all pointed at the same asymmetry: AI works reliably as an entry point for product discovery, yet checkout refuses to move.
That asymmetry shows clearly in the numbers. Flash AI, an AI-led discovery platform deployed with direct-to-consumer brands, reports that non-AI traffic converts at 6-10% while queries routed through ChatGPT, Gemini, Flash or Amazon convert at least three times higher. Buyers arriving via AI land with most of their comparison work already done.
Founder Ranjith Boyanapalli splits the relationship between shopping and AI into two layers, discovery and checkout, and argues that almost all innovation is happening on the discovery side, with checkout still a few years away. ChatGPT launched checkout first and rolled it back, in his reading, because it realised the feature needs both consumer and merchant adoption.
It is not an issue of whether there is enough technology ready; it is a matter of whether the ecosystem, regulators are ready, and of course merchants and consumers.
The sequence of events, from OpenAI enabling direct purchases of Etsy, Walmart and Shopify items inside ChatGPT in September 2025 to reversing course roughly six months later, along with the conversion gap Walmart observed, is covered with the numbers in why ChatGPT Instant Checkout failed. The pivot by retailers toward integrating their own assistants instead of in-chat checkout is best illustrated by Walmart bringing Sparky into ChatGPT. What this article takes up is the layer beyond that: the reason it is stuck.
Retailers cannot hand over checkout because first-party data ownership is unsettled
The most easily missed thread in the ET piece is the first-party data point raised by industry executives. First-party data means information a business collects directly from its customers, such as purchase records, contact details and site usage. Raviteja Dodda, founder of the agentic customer engagement platform MoEngage, says brands need first-party customer understanding rather than dependence on a third party, and that it is not yet clear how that works.
Reading the technical spec makes the situation more concrete. In OpenAI's Agentic Commerce Protocol checkout spec, the Buyer object passed to the merchant defines name and email address as required fields, with the email described as being for communication. Phone number is optional. On payments, OpenAI is not the merchant of record; the merchant runs risk analysis and charges through its own PSP. So the minimum customer information needed to complete an order does technically reach the merchant.
The unresolved part is everything after that. Can the merchant load that email into a CRM and use it for repeat-purchase campaigns or newsletters? Can a customer acquired through ChatGPT be connected to the merchant's own loyalty base? OpenAI's public documentation is silent on this point, so whether the data may be used for marketing purposes remains undisclosed. The merchant handles returns and support, yet has no way of knowing where that customer will come back from. Suspecting an arrangement where the cost sits with you and the relationship sits with the platform is a reasonable retail instinct.
Zave cofounder Hiren Patel explains the same wall from the consumer behaviour side. A shopper's primary discovery destination remains the commerce platform, and changing that is hard for reasons much deeper than we tend to think. The trust these platforms accumulated over years is itself the barrier to entry.
In India, the land grab is happening on the discovery side
While checkout stalls, investment on the discovery side is accelerating. On April 7, 2026, Google announced on its India blog a simultaneous shopping update across the Gemini app, AI Mode and Circle to Search. Gemini chats now surface shoppable product listings, comparison tables and prices, while AI Mode in Search bundles inventory and review information into its answers.
The scale figures vary by source. That blog describes the Shopping Graph as more than 50 billion product listings, while the Google statement quoted by ET puts it at more than 60 billion. Both refer to the same catalogue at different points, since it has kept expanding from 35 billion in early 2023. The statement quoted by ET adds that 87% of people reported making more confident decisions using AI Mode, and that Indian shoppers using AI in their purchase journey interact with 13 touch points on average, against 5.5 for those who do not.
More than doubling the number of touch points rewards careful reading. AI is not shortening consideration; it is thickening it. That is precisely why conversion at landing runs high, and it means there is no inherent need to finish the payment inside the chat.
Indian players are concentrating on discovery too. Flipkart shipped its conversational AI commerce platform SLAP (Shop Like a Pro) as a standalone application rather than an in-app feature. It succeeds Flippi, the generative AI assistant launched in 2023, and wires the Flipkart Minutes quick-delivery service directly into the chat. Meesho unveiled PRISM, a discovery system running more than 100 ranking models over behavioural signals, and says over 75% of orders now come through AI-powered personalised feeds. It also runs Vaani, a voice conversational assistant covering more than 10 Indian languages.
What all three share is that almost nothing in their announcements concerns completing payment inside the chat. Conversation is designed as an entrance, and checkout returns to the company's own flow. A senior Indian ecommerce executive quoted by ET said AI will play a significant role in how people find and discover products, while stating plainly that end-to-end agentic checkout is a few years away.
How to read the gap between forecasts and the ground
The expectation side still carries big numbers. A Morgan Stanley report published in December 2025 estimates that AI-powered shopping assistants could drive up to $385 billion in US ecommerce spending by 2030, equal to 10-20% of the US ecommerce market. Its AlphaWise consumer survey found that nearly 23% of Americans made at least one AI-assisted purchase in the past month, with large language model adoption approaching 50% of US consumers. ET summarised the report as showing 30-40% of LLM users making purchases.
Lined up, the figures look bullish, but the forecast does not assume a world where agents buy autonomously. The categories Morgan Stanley flagged as fastest-growing were grocery and consumer packaged goods, meaning repeat replenishment. Put the other way, for higher-ticket goods that require comparison, the value stays concentrated in better discovery for now.
That is where the caution from the Indian ecommerce executive lands. The technology may be in place, but regulators, the payments ecosystem and consumers are not ready at the same moment. Boyanapalli's "ecosystem problem" is exactly this mismatch in timing.
What e-commerce operators should prepare for
Open discovery to AI, keep checkout in-house. Take that as the working assumption and investment priorities fall out naturally.
Product data comes first. Rashida Bohra, cofounder of the AI-led fashion discovery platform ALT Fashion, notes that only a few years ago she had to explain AI-led alternative marketing channels, whereas brands are now switching to product catalogues, descriptions and images that are AI readable. Structured attributes, life-context vocabulary, machine-readable inventory and delivery terms. This work pays off regardless of where checkout ends up.
If AI-referred traffic stays blended with everything else, a three-times conversion gap disappears into the average. Measurement is the next thing to fix. Split the referrers and track conversion rate, average order value and return rate for AI-referred sessions separately. Without that separation you cannot make an investment call on discovery at all.
The landing experience matters too. Buyers arriving via AI have finished comparing, so there is no need to lay out comparison material again on the product page. Show stock and delivery date, and make the path to payment as short as possible. In a world where AI does not own checkout, the design of that landing surface is your conversion rate.
Bring your customer-registration design forward as well. Even if you consider in-chat checkout later, as long as the terms for first-party data remain undisclosed you need your own path to pull the relationship in-house right after purchase. Consolidate order confirmations, shipping notifications and review requests on your own domain, and the customer relationship survives whatever happens to the route.
Finally, split the internal decision. Investing in discovery and implementing checkout are separate agenda items. The first pays off immediately; the second can wait until specs and terms settle. Bundle them into one "agentic commerce readiness" project and the items you could start today get held hostage by the ones you should be waiting on.
Conclusion
AI commerce today is a picture of discovery running ahead while checkout waits. What is holding it is not technology but an unresolved question about who owns the customer relationship. OpenAI's spec passes merchants the information needed to complete an order, but says nothing about how far that information may be used.
The thing to watch next is whether the platforms put those terms in writing. Two points matter: whether marketing use is permitted, and who owns customers acquired through AI. The moment both are published, retail caution moves a notch. Until then, invest in discovery, sharpen the landing, and keep checkout on your own stack. It is unglamorous, and it is the position that loses the least.
