ChatGPT Shopping and Instant Checkout: Exploring the future of commerce

Disclaimer: This information is accurate as of March 2026.

For a brief window, it even let shoppers complete purchases inside the chat. But OpenAI has since walked that back and on March 24, 2026, launched a revamped shopping experience focused on product discovery, with merchants handling checkout through their own apps, some built directly inside ChatGPT.

While direct checkout hasn’t stuck, the shift signals something bigger. Shopping-related queries increasingly flow through ChatGPT. This has moved well past novelty into real channel territory.

This matters for incentives marketing teams. If loyalty programs and promotional campaigns aren't machine-readable, an agent can't reliably explain or compare the value.

This article covers:

ChatGPT Shopping is OpenAI's product discovery feature. It turns ChatGPT into a personalized shopping research assistant. OpenAI rolled it out in 2 phases in late 2025: Instant Checkout came first, followed by Shopping Research, focused on product discovery and buyer guides.

Shopping Research is what survived. On March 24, 2026, OpenAI significantly upgraded it as part of a broader shopping revamp (covered in detail below). At its core, the feature lets shoppers describe what they want in plain language. ChatGPT asks clarifying questions, then researches across the web. It returns a curated buyer's guide with ranked recommendations, prices, reviews, and links to buy.

This is a different game from Google Shopping or Amazon search. You can't buy your way to the top. Visibility is earned through data quality, product relevance, and how well your catalog matches what a shopper is actually asking.

The experience feels less like browsing a store and more like talking to a knowledgeable friend with the entire internet at their fingertips. The flow moves through five stages.

The feature kicks in through a few entry points. A suggested action card may appear in conversation (for example, "Research the best running shoes"). Shoppers can also select "Shopping research" from the tools menu. It can also activate automatically when a query shows clear shopping intent.

Once triggered, ChatGPT opens a conversation. It could ask clarifying questions about budget, recipient, key features, and preferences. Additionally, it can also use expert roundups and product listings and recommend options on that basis.

Behind the scenes, ChatGPT researches across the web. It typically prioritizes quality sources over spammy pages. The output is a buyer's guide that explains tradeoffs, not a ranked list of links.

ChatGPT sometimes displays results as a product carousel. Each card shows an image, name, price, star rating, and a short descriptor like "compact and stylish." Clicking a card reveals more images and detail, plus direct links to the merchant's website. Since the March 2026 revamp, rich visual display (including image-based search and side-by-side comparison) has become a standard part of the shopping experience, not an occasional one.

This is where the current model lands. In most cases, shoppers click through to the retailer's site to complete the purchase. ChatGPT handles discovery, while the merchant handles the transaction. However, since March 2026, some merchants (including Walmart and Sephora) have launched dedicated apps inside ChatGPT that let shoppers link accounts, use loyalty benefits, and complete transactions without leaving the chat entirely.

ChatGPT Shopping started with a more ambitious goal than product research. OpenAI originally wanted to own the entire purchase flow inside the chat window.

When OpenAI launched Instant Checkout in late September 2025, the ambition was clear: let shoppers buy without leaving the chat. A shopper could tap "Buy," confirm shipping and payment, and complete the order inside ChatGPT.

Under the hood, OpenAI and its payments partner, Stripe, initially proposed an Agentic Commerce Protocol (ACP). It aimed to standardize how AI agents and merchants exchange checkout sessions and order information.

OpenAI framed the model as merchant-led commerce. The merchant still runs payment, tax, fulfillment, and post-purchase operations. ChatGPT acts as the discovery and interface layer.

In March 2026, OpenAI reversed course. It first pulled back from in-chat checkout, then on March 24, officially revamped its entire shopping experience to focus on product discovery.

OpenAI said the initial version of Instant Checkout "did not offer the level of flexibility that we aspire to provide." Coverage pointed to a few pressures that made in-chat checkout hard to sustain. Product selection remained limited 6 months after launch. As of early March, only around 30 Shopify merchants were available through Instant Checkout. Onboarding merchants proved arduous, and the feature was prone to inaccurate or out-of-date product information.

Operational burden was a factor. Analysts at Gartner and Forrester noted that OpenAI underestimated how difficult the enablement of transactions would be. Product focus also played a role. OpenAI appeared to prioritize being the discovery layer over owning checkout.

The common thread is risk. Checkout pulls a platform into edge cases that are expensive to own. The retreat looks less like a single failed feature and more like a strategic move to reduce risk while defending distribution.

On March 24, 2026, OpenAI launched a revamped shopping experience focused entirely on product discovery. The new model is simpler: ChatGPT handles discovery, the merchant handles the transaction.

The updated experience introduces richer visual shopping tools. Shoppers can now browse products visually, upload images to find similar items, compare options side by side with key details like price, reviews, and features, and refine results through conversation. OpenAI says it has improved speed, relevance, and product coverage under the hood.

To power the new experience, OpenAI expanded its Agentic Commerce Protocol (ACP) to support product discovery. Merchants can now share product feeds and promotions directly with ChatGPT through ACP, so their catalogs are "fully represented" in the platform. Leading retailers including Target, Sephora, Nordstrom, Lowe's, Best Buy, The Home Depot, and Wayfair have already integrated into ACP for discovery. And for Shopify merchants, product data is already integrated through Shopify Catalog. As OpenAI put it, "millions of Shopify merchants are open for business in ChatGPT" without additional work required from individual merchants.

Alongside the broader discovery experience, merchants can also build dedicated apps inside ChatGPT for deeper integrations. These apps let retailers maintain control over the customer experience and transaction process. Walmart, for example, launched an in-app ChatGPT service that supports account linking, loyalty programs, and Walmart payments. Sephora also debuted its own ChatGPT app with Beauty Insider profile integration and loyalty rewards.

This positions ChatGPT as a referral and discovery layer, closer to a comparison shopping engine than a commerce platform. But even with the operational retreat from checkout, the underlying protocol conversation has grown. OpenAI stepped back from operating checkout as a shopper-facing product, not from the underlying infrastructure ambition.

ChatGPT's journey from "buy inside the chat" to "we'll point you to the merchant" previews a broader pattern taking shape across platforms: AI discovers, the merchant transacts.

That pattern splits the buying journey into two distinct layers. Discovery is where agents research, compare, and recommend. Transactions are where merchants run payment, fulfillment, and customer service.

Google has also publicized an agent payments protocol for agent-to-merchant commerce. This signals the architecture is becoming an industry default, not an OpenAI-specific experiment.

In practice, agentic commerce forces three layers to line up:

When the system that discovers your product and the system that sells it are no longer the same, many assumptions about ecommerce break down. The next sections cover what that means in practice.

The shift from human-browsed storefronts to agent-mediated discovery changes what good marketing looks like at a technical level. Four areas demand immediate attention.

ChatGPT builds its recommendations primarily from third-party sources: Reddit threads, expert blogs, review sites, independent roundups. That's mostly outside your control. You can't buy your way onto those pages either.

What you can control is your structured data. And in a channel where you can't outspend competitors on ads, that becomes the primary direct lever for visibility.

"Agent-readable" product data means your catalog answers common comparison questions without guesswork. Clear titles and attributes matter: consistent dimensions, materials, compatibility, and use cases.

So does accurate variant-level pricing with current inventory and shipping constraints. Add machine-readable markup (Product, Offer, and Rating signals), and trust signals like credible reviews and third-party coverage.

Keep your catalog data complete and fresh. If a shopper clicks through and finds a different price or an out-of-stock item, you'll lose them.

When ChatGPT recommends your product and sends a shopper to your site, there's a trust handoff. If the landing experience doesn't match what the AI promised, you lose the sale.

If a shopper clicks through and sees unclear pricing, the wrong variant, or missing availability, confidence drops fast.

Your product details now serve two audiences. Human shoppers need fast clarity and confidence. AI agents benefit from structured, unambiguous product facts they can reference during the discovery phase.

This is one of the most underappreciated implications. Many loyalty programs and promotional campaigns were designed around point of sale and checkout. They weren't designed to expose benefits to third-party agents through real-time APIs.

If an AI agent compares two similar products at similar prices, structured incentives can decide the winner. Without them, agents default to price. And a race to the bottom on price is a race nobody wins.

The fix is turning incentives marketing into queryable data, not UI interactions locked behind login screens. What needs to be exposed:

If an incentive only appears after login or only at checkout, an agent usually can't factor it into the recommendation. The promotion that lives behind an email-only code, a pop-up, or an authenticated portal effectively doesn't exist to an agent.

This is not hypothetical. When OpenAI revamped its shopping experience in March 2026, two of the first deep integrations involved loyalty. Sephora launched a dedicated app inside ChatGPT that connects to its Beauty Insider loyalty program, enabling personalized recommendations based on customer profiles, and making rewards, free shipping promotions, and samples accessible through the agent. Walmart launched its own in-app ChatGPT service with account linking, loyalty integration, and Walmart payments. Both are early signals that the brands investing in machine-readable incentives are the ones getting featured inside the new discovery layer.

The broader principle holds regardless of which platform you use: if your incentive logic lives only in session-based UI flows, agents can't see it. The brands that expose their incentives through real-time, API-accessible data structures will have an edge as agent-mediated commerce grows.

The challenge this article describes comes down to a structural gap. Agents need to query your incentive logic in real time. Most loyalty and promotions systems weren't built for that. They were built for checkout screens and campaign calendars, not for third-party agents evaluating your best offer on the fly.

When agents can't see your incentives, they default to the one variable they can always compare: price. That's how margin erodes at scale without anyone making a conscious decision.

This is why Talon.One is building the Unified Incentives Protocol (UIP), a platform-agnostic standard for exposing loyalty and promotions in a machine-readable way across agent-led shopping journeys.

In an agent-mediated world, loyalty shifts from a marketing campaign to core commerce infrastructure. It provides identity, eligibility, and long-term value signals that let agents optimize beyond price. The brands that treat incentives as structured, queryable data will be the ones agents can actually recommend.