AI Shopping Is Moving From Discovery to Checkout

For the last few years, AI shopping has largely been about discovery.

Customers could ask ChatGPT what to buy, use Google’s AI-powered search experiences to compare products, or turn to AI for recommendations.

Now the technology is moving further into the transaction itself.

Google is building shopping experiences that can support product research, basket building and checkout across Search and Gemini. OpenAI is expanding shopping inside ChatGPT, giving merchants new ways to surface their catalogues while customers research and compare products conversationally.

For ecommerce businesses, that matters because the website may no longer be the place where every stage of the purchase journey happens.

From search assistant to shopping assistant

The traditional ecommerce journey has usually involved several distinct stages.

A customer discovers a product through Google, social media, advertising or another channel. They visit the retailer’s website. They browse. They compare. They add something to their basket. They check out.

The website sits at the centre of almost the entire transaction.

AI shopping starts to blur those boundaries.

Google’s Universal Commerce Protocol — or UCP — is a new open standard designed to let agents and commerce systems work together across discovery, buying and post-purchase support. Google’s current UCP-powered checkout allows participating merchants to offer checkout directly on eligible product listings in AI Mode and Gemini while remaining the seller of record.1

Google has also introduced Universal Cart, an intelligent shopping cart designed to work across retailers and services including Search and Gemini. Google says shoppers will be able to add products while browsing its services and, for participating brands, either check out on Google with Google Pay or transfer items to the retailer’s site.2

That is a significant change.

The shopping journey no longer necessarily has to move:

That does not mean retailer websites are about to disappear. But it does mean parts of the shopping journey that have traditionally happened on the website could increasingly happen somewhere else.

ChatGPT is moving deeper into ecommerce too

OpenAI is developing along a similar trajectory, although its current shopping experience places particular emphasis on product discovery.

In March 2026, OpenAI introduced richer shopping experiences in ChatGPT that allow people to browse products visually, compare options side-by-side and access more up-to-date product information without repeatedly moving between websites. OpenAI says merchants can share product feeds and promotions through the Agentic Commerce Protocol, creating a foundation for broader AI-native commerce experiences.3

For Shopify merchants, OpenAI currently says relevant stores may appear in ChatGPT shopping experiences, with checkout taking place on the merchant’s own online store.4

That distinction matters.

This is not yet a world where every ecommerce transaction disappears into an AI platform.

But we are clearly moving towards an environment where a greater proportion of the research, comparison and decision-making may happen before a customer reaches the retailer’s traditional storefront.

For brands, that makes ecommerce strategy and development increasingly inseparable from how products are represented across the wider digital ecosystem.

Product data is becoming part of the storefront

This shift changes the importance of something many businesses still treat largely as a technical requirement:

the product feed.

For years, feeds have powered channels such as Google Shopping, Performance Max, Meta catalogues and marketplaces.

They are often thought of as something the advertising platform needs in the background.

Agentic commerce makes that data much more strategically important.

If an AI system is going to recommend, compare or transact with a product, it needs to understand it.

Google’s UCP documentation specifically emphasises the need for robust, up-to-date Merchant Center data, while newer capabilities allow agents to work with information including variants, inventory and pricing.5

That makes the product feed something more than an advertising feed.

It starts to become a machine-readable version of the shop itself.

For ecommerce businesses, poor product data may therefore create problems well beyond Shopping Ads. A product with a vague title, weak description, missing attributes or inconsistent information may simply be harder for automated systems to interpret confidently.

SEO and ecommerce are becoming harder to separate

There is another consequence.

The line between SEO, product content and ecommerce technology is becoming increasingly blurred.

A traditional SEO strategy might focus on helping a product or category page appear when somebody searches:

“luxury mahogany console table”

That remains important.

But AI-led shopping introduces much richer questions:

“I’m looking for a statement console table for a Georgian hallway. I want something traditional, handmade and substantial, but not overly ornate.”

or:

“Show me a hand-finished mahogany console table around 120cm wide that would suit a classic English interior and can be delivered in the UK.”

These are not simply keyword searches.

They combine intent, context, product attributes and personal preferences.

For a retailer to appear appropriately, the systems answering those questions need enough information to understand both the product itself and the space, style and use case it is suitable for.

That reinforces many of the fundamentals we already associate with good SEO and search optimisation: clear product titles, comprehensive descriptions, strong category architecture, structured information, accessible websites and consistent information across the wider web.

But the objective becomes broader than ranking a page.

It is increasingly about making sure a business and its catalogue can be understood by the systems mediating the customer’s research.

Does this mean ecommerce websites become less important?

Possibly the most obvious concern is whether AI shopping platforms eventually reduce the importance of the retailer’s own website.

There is certainly potential for some parts of the journey to move elsewhere.

If a customer can discover a product, compare it with competitors, add it to a basket and potentially pay without following the traditional route through the retailer’s site, fewer interactions may happen directly on the storefront.

But that does not make the website redundant.

It changes its job.

The website remains where a retailer controls its brand, merchandising, storytelling, complete product information, customer accounts, loyalty, cross-selling, editorial content, customer service and much of the post-purchase relationship.

Google’s UCP model is designed to keep participating retailers as the seller of record, while allowing some journeys either to complete on Google or transfer to the merchant’s own checkout.1

So the better question is not:

“Will AI replace ecommerce websites?”

It is:

For brands, maintaining a compelling destination becomes more important if the customer arrives later in the buying process and with stronger expectations.

The customer relationship could become the real battleground

There is a deeper commercial question underneath all of this.

Who owns the customer relationship?

Retailers have already faced this challenge with marketplaces.

Social commerce presents something similar.

Agentic commerce potentially pushes that tension further.

If an AI assistant understands what products somebody likes, their budget, previous purchases, delivery preferences and potentially even payment preferences, the platform could become one of the most influential intermediaries between customer and retailer.

The ability to make shopping easier is clearly valuable.

But retailers will need to think carefully about what they gain — and what they potentially give up — when more of the customer journey takes place on somebody else’s platform.

Paid advertising is likely to change with it

Agentic shopping also creates an interesting question for paid media.

Retail advertising has traditionally worked by getting a customer’s attention and encouraging them to click.

But if an AI assistant is already helping the customer make a decision, advertising may increasingly need to operate inside that decision process.

That could gradually move paid shopping away from:

keyword → advert → click

towards something closer to:

customer intent → AI recommendation → relevant commercial offer → transaction

The implications for bidding, product feeds, creative assets and attribution could be substantial.

I would therefore expect the boundaries between paid search and shopping advertising, Merchant Center optimisation, ecommerce data and AI visibility to become progressively less distinct.

Attribution could become more complicated again

If more of the shopping journey happens inside an AI platform, marketers also face another familiar problem:

measurement.

Imagine:

ChatGPT → product research → brand discovered → Google search → website → purchase

Google Analytics may attribute the final transaction to Organic Search.

Or:

Gemini → product comparison → platform cart → merchant checkout

Depending on how those systems pass attribution data, the journey could look very different inside the retailer’s analytics platform.

The underlying challenge remains:

The place where somebody converts is not necessarily the place that created their preference.

That is already true across social, search, email and offline marketing.

AI simply adds another layer.

This could make it increasingly important for ecommerce teams to combine channel attribution with wider analytics and performance measurement, including brand search, new customer acquisition, customer surveys, incrementality testing and overall commercial performance.

This is not happening everywhere overnight

There is an important caveat.

Agentic commerce is still developing.

Google’s UCP onboarding is currently rolling out gradually in the United States, Canada and Australia, and the UCP-powered checkout feature is available only to selected merchants at this stage.5

Similarly, the exact capabilities available through ChatGPT vary by merchant and integration.

So UK retailers do not need to rebuild their ecommerce strategy tomorrow because customers are suddenly bypassing every website.

That is not what is happening.

But the direction of travel is becoming clearer.

The major technology platforms are building infrastructure intended to allow AI systems to participate in commerce from discovery through to transaction.

And infrastructure changes usually matter before mass customer behaviour completely catches up.

What should ecommerce brands do now?

There is no need to chase every new protocol or rush into speculative integrations.

Most retailers would benefit far more from strengthening the fundamentals they already control.