AI shopping assistants
are becoming the
new storefront.

AI shopping assistants are changing where buying decisions get made. A growing share of buyers now start with ChatGPT, Perplexity, or another assistant, ask what to buy, and arrive at your store with the consideration set already built by the assistant. If your products are not in the answer those assistants give, you lose the sale before the buyer ever touches your site. The discovery layer is working now and converting at measurably higher rates. In-chat checkout is still early and stumbling. The immediate work is catalog quality and structured data so your products show up when assistants field buying questions in your category.

For most of ecommerce history, your storefront was the first thing a customer saw. They found you through search or an ad, landed on your homepage or a product page, and the experience you designed did the selling. That assumption is breaking. A growing share of buyers now start with an AI assistant, ask it what to buy, and never touch your storefront until the decision is mostly made, if they touch it at all.

The AI assistant is becoming the new storefront, the layer where discovery, comparison, and increasingly the purchase itself happen. The numbers behind this are large enough to demand attention: McKinsey projects $900 billion to $1 trillion in US retail revenue flowing through agentic commerce by 2030, and an estimated $3 to $5 trillion globally (via Elogic). An IBM study in early 2026 found 45% of consumers already use AI for some part of the buying journey (IBM IBV).

And yet the same period saw OpenAI quietly kill its Instant Checkout feature and a federal judge block Perplexity's shopping agent from buying on Amazon. So which is it, the future of commerce or an overhyped feature that keeps stumbling? The honest answer is both, and understanding why is the difference between preparing intelligently and either ignoring the shift or betting the business on it too early. This is the operator's read.

The terms, before
the hype and the
backlash.

This space generated a thick layer of jargon fast, much of it from vendors with something to sell. Here is the plain-language map.

Keep the distinction between discovery and checkout in mind. Being recommended by an assistant and being purchased inside one are two very different milestones, and they are moving at very different speeds.

The numbers are
large enough to
take seriously.

Start with the scale of what is being projected, because it explains why every major platform is racing to build this. Beyond McKinsey's trillion-dollar 2030 figure, nearer-term estimates have AI platforms accounting for roughly $20.9 billion in retail spending in 2026, close to quadruple the prior year (Elogic). Adoption is already meaningful and skews younger: against a baseline where 23% of Gen X have recently used an assistant to search for products, the figure climbs to 32% for Millennials and 35% for Gen Z (Forrester data, via Elogic).

The behavior is real and the trajectory is steep. What is not yet clear is the timeline and the form it settles into. A brand reading these numbers should feel urgency about preparation and skepticism about anyone promising that in-chat checkout is about to become the dominant way people buy. Both things are true at once, which is the uncomfortable spot most genuinely new channels occupy for a few years.

The leaders pulled
back, and that
tells you something.

Here is the part the hype skips. In March 2026, OpenAI quietly killed Instant Checkout, its feature for buying inside ChatGPT, with the official explanation that it "did not offer the level of flexibility that we aspire to provide" (Forrester). In the same window, a US federal judge issued a preliminary injunction blocking Perplexity's Comet browser from making purchases on Amazon. The two companies positioned as leaders in agentic shopping both hit a wall in the same quarter.

That does not mean the trend is fake. It means the hardest part, an agent reliably and safely completing a real purchase across the messy reality of merchant systems, payments, returns, and platform politics, is genuinely hard and not solved. The discovery layer, where an assistant recommends your product, is working now. The transaction layer, where it buys for the customer, is still being fought over in courtrooms and rebuilt after failed launches.

Whoever owns the
first question owns
the customer.

The reason this matters even before in-chat checkout is settled is that the assistant is taking over the first touch. When a buyer asks an assistant "what is the best running shoe for flat feet under $150," the assistant, not your storefront, is doing the framing, the comparison, and the recommendation. By the time the customer reaches you, the consideration set was built somewhere you do not control, and you were either in it or you were not.

This is the same disintermediation that search engines and marketplaces did to brands, with a sharper edge, because a conversational answer names fewer options than a page of results. A search returns ten links and the buyer chooses. An assistant might name three products and the buyer trusts. Being one of the three is worth far more than being on page one ever was, and being absent is worse than ranking eleventh, because there is no second page to scroll to. Getting into that answer is the work I laid out in the answer engine optimization piece, and it is the foundation everything else in this post sits on.

Fewer of them,
but they arrive
ready to buy.

The volume through AI assistants is still small relative to total traffic, but the quality is not. Adobe's Q1 2026 data found AI-referred shoppers converted 42% better than other shoppers, and retailers with AI agent integrations reported outsized sales growth during the 2025 Cyber Week (Adobe, via Elogic). That tracks with the proprietary numbers I broke down in the AI-referred traffic piece: these buyers arrive further down the funnel, with the decision substantially made by the assistant before they ever reach you.

Conversion also varies sharply by platform, which is worth knowing as you decide where to focus. Across LLM shopping traffic measured by Shopify, Claude has shown the highest conversion at around 16.8%, ChatGPT in the 14 to 16% range, and Perplexity around 10.5%. The takeaway is not to chase one platform, since these numbers will move, but to recognize that assistant traffic as a category converts at rates that make even small volumes worth capturing.

Be the answer
now. Be buyable
soon.

There are two distinct goals here, and they deserve different levels of investment today. The first is being the recommended answer, getting your product named when the assistant responds to a buying question. This is available now, it is where the proven conversion is, and it is largely a function of the AEO work: clean product data, genuine reviews exposed as structured data, third-party corroboration, and a product that is legible to a machine as the right answer to a specific need.

The second goal is being transactable inside the agent, letting the assistant complete the purchase through emerging protocols like the Agentic Commerce Protocol or Google's Agent Payments Protocol. This is the frontier, and it is where the stumbles are happening. The right posture is to understand the protocols, make sure your platform and payment stack can support them as they stabilize, and not pour resources into a checkout integration that the ecosystem is still actively redesigning. Shopify has been positioning for this directly, which I covered in the Sidekick and agentic commerce piece, and being on a platform that is doing this work for you is a real advantage.

The thing you can
lose is the
relationship.

The strategic risk in agentic commerce is the same one marketplaces created, intensified. When a customer buys your product through an assistant, the assistant may own the relationship, the data, and the next recommendation, not you. You can become a fulfillment endpoint for someone else's interface, the way many brands became interchangeable listings on a marketplace. The buyer remembers the assistant that helped them, not necessarily the brand that shipped the box.

That risk is real, but it is not a reason to opt out, because opting out is not an option when the buyers are already there. It is a reason to fight for the brand relationship inside this channel: to be distinctive enough that the assistant names you specifically rather than a generic category, to capture the customer into your own owned channels once they buy, through the email and SMS flows I detailed in the Klaviyo flows piece, and to keep a direct relationship strong enough that customers come looking for you by name. The brands that survive marketplace dependence are the ones with real brand equity. The same will be true here.

Prepare for the
part that works.
Hedge the rest.

Here is the practical posture for a commerce brand in 2026, calibrated to where the channel actually is rather than where the projections say it will be.

Win the discovery layer now. This is the proven, available opportunity. Make your catalog legible to assistants with clean product and review data, build the third-party authority that gets you recommended, and treat being the named answer as a real channel to invest in. Everything in the AEO playbook applies directly. Get your data and feeds clean. Whether the transaction happens in-agent or on your site, an assistant can only represent your product accurately if your product data, pricing, availability, and reviews are accurate and structured. Messy catalog data is the most common reason a brand is misrepresented or skipped.

Understand the protocols, do not over-build on them. Know what ACP and AP2 are, confirm your platform is moving with them, and be ready to enable in-agent checkout when it stabilizes. But do not stake the quarter on a checkout integration the ecosystem is still rebuilding. Stand up measurement. Track which assistants are referring and converting, and watch branded and direct traffic for the dark-funnel lifts that assistant recommendations create even when they do not show up as a clean referral. And protect the relationship. Capture every customer you can into owned channels, and invest in the brand distinctiveness that makes an assistant name you rather than your category.

The AI assistant becoming the storefront is simultaneously overhyped in the short term and underestimated in the long term. The checkout features will keep stumbling and relaunching for a while. The discovery behavior, buyers asking an assistant what to buy and trusting the answer, is already entrenched and growing. Build for the part that is real now, get your catalog and your brand ready for the part that is coming, and keep your direct customer relationship strong enough that you are never just an endpoint in someone else's interface. The storefront is moving. Make sure you are still in the frame when it does.

Two context posts worth reading alongside this one: why some products win in ChatGPT and most stay invisible goes into catalog and data mechanics, and GEO vs SEO for commerce brands explains how the optimization game has fundamentally changed. Both connect directly to the discovery work this post is about. For the channel mix question, the 2026 channel mix framework shows where AI-mediated discovery fits alongside paid, organic, and social.

What brands
keep asking.

Getting ahead of the AI discovery shift is some of the highest-stakes work in commerce right now. It is part of the DTC brand consulting practice, and the form takes two minutes: start the conversation.

Questions I keep
getting asked.