The end of the search box: Winning product discovery and checkout in AI channels

The product search shift: From box to conversation

For twenty years, the search box shaped how people found products online. You typed a few keywords, skimmed pages of results, clicked, compared, clicked again, and, eventually, made a decision. That model is fading faster than anyone expected.

Here’s what’s really happening: People aren’t typing keywords anymore; they’re expressing their search as natural conversations.

They’re describing needs in natural language. They’re asking questions the way they’d ask a salesperson. And large language models (LLMs) aren’t responding only with lists of links; they’re responding with answers, summaries, recommendations and actions.

Several signals make this shift undeniable:

• Google states that modern users “interact using conversational language, not keywords,” and expect systems to understand intent, context and full-phrase meaning, not just literal terms.
• McKinsey’s 2025 analysis reveals that half of all consumers now utilize AI-powered search tools, and AI-powered search surfaces could redirect 20–50% of traffic that previously entered through traditional search engines.

In short, the front door of commerce is shifting from your website. It’s the AI conversation happening elsewhere.

But this shift isn’t just about how people search. It’s about what they expect from search. And that’s where the story gets bigger.

White Paper

Navigating the Agentic Commerce Era On Your Terms

The practical guide to help you approach agentic commerce strategically — not out of FOMO.

Download Now

When search evolves, shopper expectations evolve faster

Once shoppers experience conversational discovery, their expectations are permanently altered. The bar rises, not slightly, but fundamentally, in the following ways: 

They expect search to understand intent, not keywords

The old query: men’s red running shoes uk 9.

The new query: I need durable running shoes for winter, size 9, budget around £120.

That’s not “search.” That’s problem-solving. A request for reasoning. 

They expect comparison and decision-making, not filtering

Shoppers don’t think in faceted filters. They think in trade-offs:

• Which works better for flat feet?
• Is there a warmer option?
• What’s the best option for trail running?

This requires inference and judgment, not a drop-down menu.

They expect one smooth flow, not a long sequence of steps

The traditional online shopping journey is full of small hurdles. A shopper starts with a search box, lands on a list of products, clicks on one, adds it to the cart, goes through checkout, enters details, pays and finally gets a confirmation.

It’s familiar, but it’s slow. And every step adds a chance for the shopper to drop off.

The conversational flow is completely different.

A shopper asks a question.

• The system recommends the most relevant option.
• The shopper says, “yes.”
• The purchase is complete.

That’s it.

Ask → decide → buy.
One moment of intent.
One place where everything happens.
One clean outcome.

Shoppers are getting used to this simplicity in AI tools, and once they experience it, the traditional multi-step flow feels outdated and heavy.

They expect outcomes, not just options

“I’m going hiking next weekend.”
“I’m traveling to Iceland in December.”
“I want to start weight training.”

These are not product queries. They are goals. AI must make decisions on behalf of the shopper.

This creates a gap that traditional commerce systems can’t close.

The old commerce stack was built for a world where shoppers did all the work. They typed the keywords, clicked the filters, opened the product pages, added items to the cart and pushed the order through checkout.

Everything depended on the shopper driving the process step by step. But AI flips that model.

When a shopper says: “I need a good winter jacket for Iceland—keep it under £150.”

They’re not asking for a list of pages to click through.
They expect the system to:

• Understand what “good” means in that context.
• Filter the products automatically.
• Compare options on their behalf,
• Pick the right size or variation.
• Build the cart.
• And complete the purchase.

All of that without forcing them through a long, manual flow. 

Traditional commerce engines weren’t designed for this kind of behavior. They expect predictable inputs, clear steps and a human to make every decision along the way.

AI doesn’t work like that. It needs a way to “talk” to the commerce system directly, ask for data, take actions and complete tasks safely.

That’s why the industry needed a new standard. A clean, predictable way for AI agents to interact with product data, carts and checkout flows. And that standard is the Agentic Commerce Protocol (ACP).

ACP: The layer between commerce and AI agents

The ACP created by OpenAI in partnership with Stripe, commercetools and other commerce leaders, is becoming the connective tissue between AI agents and digital storefronts.

What ACP actually does

At its core, ACP gives AI agents three abilities:

1. Understand products precisely
   Structured product data, attributes, variants, inventory and pricing that an AI can reason over without guesswork.
   
   
2. Take real actions
   Standardized functions that let an agent:
   • Search products.
   • Compare options.
   • Build a cart.
   • Add line items.
   • Calculate shipping and tax.
   • Initiate checkout.
     
     
3. Operate with guardrails
   Permission controls ensure agents can only perform actions that merchants allow — no hallucinated actions, no unsafe transactions.

ACP is not a plugin or scraping. It’s a universal protocol, much like HTML was for browsers.

How ACP works behind the scenes

Think of ACP as three cooperating layers:

1. Discovery layer
Retailers submit structured product data, either through a feed, connector or platform like commercetools AI Hub. The key is clarity. AI agents perform best when working with catalog data that’s trustworthy, enriched and consistent.

2. Action layer
ACP introduces a set of standard functions that all agents are familiar with. Instead of building custom APIs for each AI platform, merchants expose one uniform set.

3. Trust layer
Every action is governed by policies:

• What the agent can do.
• What data can it request.
• What operations require confirmation.

This is how AI becomes both powerful and safe. ACP is the missing link that enables AI to transition from answering questions to completing purchases.

Instant checkout within conversations

The most transformative aspect of ACP is Instant Checkout, a payment flow that’s handled entirely within ChatGPT, eliminating the need to redirect shoppers to a website.

Here’s the end-to-end flow:

• The user expresses intent
  “I want a warm running jacket for under £150.”
• The model uses ACP functions to fetch products
  It evaluates material, warmth, price, fit — all the data points.
• The shopper makes a choice inside ChatGPT
  “I’ll take the Nike Shield Jacket, size M.”
• ACP assembles the cart
  Behind the scenes, the agent triggers:
  • createCart
  • addLineItem
  • configure variant
  • calculate totals
    
• Instant Checkout appears in ChatGPT
  Payment data is handled securely by Stripe’s agentic payment rails.
• Order completes in the retailer’s commerce system
  With full order details, confirmation and IDs.

And the shopper never leaves the conversation.

This eliminates:

• Redirects.
• Multi-page checkouts.
• Form fields.
• Drop-offs.
• Login friction.

If discovery happens inside AI, checkout must too. Otherwise, the buyer drops out.

The rise of conversational commerce, and why this time is different

The first wave of conversational commerce failed because chatbots were reactive. They waited for prompts. They couldn’t reason. They couldn’t act.

LLMs + ACP + Instant Checkout create an entirely new model.

Agents now can:

• Interpret goals.
• Identify trade-offs.
• Recommend the right products.
• Build carts.
• Complete purchases.

This is not replacing the website. It’s replacing the journey.

From channels to surfaces

For years, commerce teams organized everything around channels:Your website, your mobile app, your marketplace listings and your ads. Each channel had its own strategy, its own analytics, and its own funnel.

That mental model no longer holds.

AI has turned any place where a shopper expresses intent into a potential buying moment.

Shoppers now ask real questions in tools like:

• ChatGPT.
• Microsoft Copilot.
• Perplexity.
• WhatsApp chat assistants.
• Retailer-owned AI agents.
• And a long list of emerging AI platforms.

These aren’t “channels” in the traditional sense. They don’t look like storefronts. There’s no homepage or navigation menu. The entire experience is a conversation that starts wherever the shopper happens to be.

What this means is simple: Commerce now lives on surfaces, not channels.

A surface is any interface where a shopper expresses intent and expects a helpful, actionable response.

It might be a chat window, a voice prompt or an embedded assistant inside a productivity tool. If the shopper reveals their intent, that’s where discovery and buying begin.

From browsing to autonomous flows

The shopper can say, “Build a full skincare routine for dry skin under £200.”

The agent:

• Selects the right products.
• Checks compatibility.
• Manages variants.
• Creates the basket.
• Offers Instant Checkout.

This isn’t conversational commerce anymore. This is agentic commerce.

The implications: AIO becomes essential for AI-driven discovery

As shoppers move from keyword searches to conversational intent, a new requirement emerges: Brands must make their products understandable and actionable for AI agents. This is the foundation of AIO, or AI Interaction Optimization.

AIO is the discipline of ensuring that AI systems can:

• Interpret your products correctly.
• Speak about them confidently.
• Take safe, accurate actions on them.

To succeed in AI-driven discovery, three key capabilities are most important:

1. Model-ready product data
   Agents need structured, clean, attribute-rich catalog data to produce reliable recommendations.
   
   
2. Agent-friendly commerce actions
   Standards like ACP provide this action layer, giving AI agents a predictable way to execute tasks safely.
   
   
3. AI channel distribution
   Each AI platform has unique expectations:
   • ChatGPT: ACP.
   • Copilot: Plugins.
   • Perplexity: inventory-aware feeds.
   • Gemini/AP2: upcoming commerce actions.

This is why platforms like commercetools AI Hub matter; merchants need one unified system to distribute, enrich, and manage product data across all AI surfaces.

What brands must do to win this new era

Here’s the practical roadmap:

• Clean and enrich product data for AI-driven reasoning: Not for SEO purposes. For agents. Attributes, variants, descriptions, and compatibility notes.
• Adopt an AI distribution layer: Don’t build one-off integrations. Use purpose-built systems (AI Hub) to unify activation across AI platforms.
• Become ACP-ready: Expose the standardized function set. Ensure your checkout and catalog can be agent-driven.
• Optimize for zero-step checkout: If your checkout requires redirects, you’re already losing conversions in AI channels.
• Treat every AI conversation as a storefront: These are not assistants. They are high-intent buying moments.

We’re entering the pre-search era

We’re moving from a world where people search for products to a world where products meet people inside conversations.

AI agents:

• Anticipate needs.
• Surface solutions.
• Manage choices.
• Complete purchases.

The search box marked the end of an old era of commerce. ACP, AI Hub and agentic interactions define the next one. We’re entering the post-search world — and the brands that prepare now will own the conversation later.

White Paper

Navigating the Agentic Commerce Era On Your Terms

The practical guide to help you approach agentic commerce strategically — not out of FOMO.

Download Now