Last updated: 23 December, 2025

ChatGPT’s shopping research is a new feature from the $500 billion AI giant OpenAI that allows AI agents inside ChatGPT to research products, compare options, and guide purchasing decisions end-to-end through a single conversational interface.

This capability represents the early form of agentic commerce, a model in which AI systems take on tasks traditionally performed by users across search, comparison, and checkout. Recent launches from OpenAI, Google, and Perplexity AI suggest that artificial intelligence is moving beyond assisting the shopping journey and will soon become the primary interface for product discovery and selection.

While positioned as a product enhancement, the move represents something far more consequential. ChatGPT’s shopping research marks the beginning of a new commercial paradigm. Instead of navigating multiple websites, filters, and ads, consumers can rely on AI agents to evaluate options and surface recommendations, collapsing the traditional e-commerce funnel into a single AI-mediated workflow and introducing new technical and trust challenges for existing commerce infrastructure.

TL;DR

• ChatGPT’s shopping research tools affirm the rise of agentic commerce, where AI agents research, recommend, and execute purchases autonomously

• OpenAI’s integration with Stripe enables fully AI-driven transactions, extending beyond product discovery to payment execution and fulfillment

• Agentic commerce collapses the traditional ecommerce funnel, bypassing traditional search, ads, and product pages, reshaping how brands are discovered and chosen

• Autonomous payments introduce new fraud, authorization, and trust risks that legacy, human-centric systems are not designed to handle

• Winning in agentic commerce requires machine-readable product data, agent-compatible checkout, and AI-native risk infrastructure like Oscilar

How ChatGPT’s AI shopping capabilities work

At its core, ChatGPT’s shopping capability functions as conversational buyer agent, engaging users in conversational product discovery while pulling real-time pricing, reviews, and availability data from across the web. 

When a user expresses purchase intent, the system can: 

• Interpret natural-language shopping constraints such as price, quality, and delivery speed

• Compare products using real-time pricing, reviews, and availability data

• Recommend options optimized for the user’s stated preferences

• Progress toward checkout without requiring traditional search, browsing, or manual comparison

This experience builds on the company's e Instant Checkout capability, developed in partnership with Stripe, which enables AI agents to move from product research to payment via programmable payment APIs.

Critically, OpenAI has stated that shopping results are organic and unsponsored, distinguishing ChatGPT’s approach from ad-driven marketplaces and search engines. This matters because it positions ChatGPT as a transactional interface, not just a discovery layer.

Why agentic commerce breaks the traditional ecommerce funnel

Traditional ecommerce depends on a linear, multi-step funnel: search, ads, product pages, comparison, and checkout. Agentic commerce collapses this entire sequence into a single AI-mediated decision flow.

When AI agents like ChatGPT handle product evaluation and selection, high-intent commerce queries never surface in search results or marketplaces. Instead, purchasing decisions are made inside conversational interfaces, where:

• SEO rankings matter less than machine-readable product data and structured attributes

• Advertising influence gives way to utility, relevance, and fulfillment certainty

• Click-based attribution disappears, even when brands meaningfully shape the outcome

The shift to agentic commerce fundamentally alters how products are discovered, evaluated, and chosen and the infrastructure enabling this transformation is already taking shape.

Payment networks and platforms including Visa, Mastercard, PayPal, and OpenAI in partnership with Stripe are building tokenization frameworks and programmable payment APIs designed specifically for AI agent–initiated transactions. Morgan Stanley Research projects that the economic impact of agentic commerce could reach $385 billion in the United States alone by 2030, representing 20% of ecommerce activity.  

This simultaneously threatens the economic foundations of search advertising and marketplace-based commerce, while creating new opportunities for brands that are AI-readable, agent-compatible, and trusted by autonomous systems.

The AI shopping arms race: Agentic commerce protocols from Google and Perplexity

OpenAI isn't operating in isolation. Google —now the third most valuable company in the world, largely thanks to its investments in AI — has rolled out its own AI-powered shopping experiences  that combine conversational search, autonomous checkout, and AI-initiated calls to local retailers for real-time inventory verification.  These capabilities are powered by Google’s Shopping Graph, which aggregates data from more than 50 billion product listings, giving its AI systems deep, real-time commercial context.

Meanwhile, Perplexity AI has partnered with PayPal to enable  in-chat purchases. This positions Perplexity less as a traditional research engine and more as a transaction-first AI interface that prioritizes speed, completion, and minimal user friction over deep comparative analysis.

Can risk systems designed for humans protect AI-driven commerce?

As capital floods into agentic commerce infrastructure, a critical security question remains largely unanswered: how do you secure a transaction when neither party is human?

The same automation that makes AI shopping fast and frictionless also creates brand new attack surfaces. When AI agents execute purchases autonomously, traditional fraud detection signals lose effectiveness. Behavioral biometrics, session analysis, and manual review assume human intent and interaction patterns that simply do not exist in transactions executed by machines. 

Bad actors are already experimenting with adversarial prompts and manipulation techniques designed to coerce AI agents into unauthorized purchases, privilege escalation, or payment redirection. 

In agentic commerce, the very nature of fraud changes — from impersonating people to exploiting machine decision logic. 

The $385 billion fraud exposure

If agentic payments reach $190 to $385 billion in the U.S. by 2030, as projected by Morgan Stanley Research, even a modest rate of fraud translates to billions of dollars in losses. At that scale, risk is no longer an edge case, but a structural property of the system.

When transaction volumes reach hundreds of billions of dollars, even low-frequency, low-friction fraud compounds rapidly and systematically. Autonomous shopping agents operate continuously, execute repeat purchases, and optimize for efficiency, not caution. This makes small exploit vectors disproportionately expensive at scale. But this is not theoretical as adoption data already suggests the transition is underway. 

A meaningful share of consumers now research, compare, and complete purchases through AI systems, with early concentration in groceries and consumer packaged goods. These categories amplify exposure because they combine high transaction velocity, recurring spend, and minimal human review, creating ideal conditions for automated abuse.

The challenge compounds in B2B environments, where autonomous agents may negotiate contracts and execute payments worth millions without human oversight. A single compromised decision loop can trigger cascading financial losses across multiple transactions or vendors.

This is precisely why risk infrastructure must evolve in parallel with payment rails, not after the deployment. That's why Oscilar is building an AI-native fraud detection and risk decisioning system designed specifically for machine-initiated commerce. Oscilar focuses on modeling agent behavior, identifying anomalous decision patterns in real time, and applying adaptive risk controls to autonomous transaction flows.

As shopping agents proliferate, the organizations that thrive will be those integrating intelligent risk management directly into their agentic infrastructure rather than bolting on legacy fraud tools as an afterthought.

At the scale of agentic commerce, fraud prevention becomes a prerequisite for growth, not a downstream operational concern.

The unresolved authentication and authorization problem in agentic commerce

Fraud detection is only part of the challenge. Authentication and authorization remain fundamentally unsolved problems at the protocol level.

When an AI agent claims to represent a consumer, how does the merchant verify that authorization is valid, current, and scoped correctly? When two agents negotiate a B2B contract, what constitutes binding consent and how is it enforced?

These aren't hypothetical concerns,- they're active architectural decisions being made right now that will determine whether agentic commerce becomes a trusted commercial channel or a systemic fraud vector.

Until authorization, identity, and accountability are resolved at the protocol level, trust will remain the limiting factor for autonomous commerce at scale.

The future of commerce in a ChatGPT-powered, agentic economy

ChatGPT and other AI-powered shopping tools represent more than a new product capability, but a structural redefinition of how commerce is discovered, decided, and executed.

And the implications for financial technology extend well beyond consumer shopping. The same systems enabling AI-driven product purchases will soon facilitate autonomous B2B payments, dynamic credit decisioning, and agent-to-agent commercial negotiations.

For organizations across retail, payments, and financial services, this creates a  dual strategic imperative: Develop machine-readable interfaces and agent-compatible infrastructure while simultaneously building risk, authorization, and fraud frameworks sophisticated enough to secure fully autonomous transactions at scale.

The organizations that succeed will be those who recognize that we're witnessing not merely the evolution of search or payments, but the foundational restructuring of how economic transactions occur and how trust is enforced in a machine-mediated economy. 

Organizations that win in this environment will:

• Build machine-readable product, pricing, and policy data

• Enable agent-compatible checkout and programmable payment flows

• Embed AI-native risk, authorization, and fraud controls directly into transaction infrastructure

Agentic commerce is the future. But only if we can trust it.