OpenAI’s current shopping pitch is simple: describe the product you want, refine the constraints, compare candidates and use an image when words are not enough. The more consequential change sits underneath that convenience. After experimenting with native checkout, OpenAI has refocused ChatGPT on product discovery, creating a new recommendation layer that retailers increasingly want to reach without surrendering the customer relationship.
Table of Contents
On March 24, 2026, OpenAI introduced richer shopping in ChatGPT with visual product browsing, image-based similarity search, side-by-side comparisons and more detailed product information. In the same announcement, it said the first version of Instant Checkout had not provided the flexibility it wanted for merchants and that it would focus on product discovery while allowing merchants to use their own checkout experiences.
That combination matters more than the interface polish. ChatGPT is being positioned less as a universal cash register and more as a decision layer between consumer intent and the retailer. The user describes a need in natural language; the system narrows the field, interprets trade-offs and presents a short list. The retailer still has strong reasons to own payment, loyalty, merchandising and post-purchase data.
The shift is already visible in retailer behavior. Reuters reported on August 7 that companies including Walmart, Ulta Beauty and Wayfair were adapting their sites for chatbot discovery while trying to keep transactions on their own properties; it also reported that OpenAI had ended its original Instant Checkout in March and moved its emphasis toward discovery. The strategic prize is therefore not simply “AI shopping.” It is control over the moment when a shopper’s vague desire becomes a named product shortlist.
The shopping pitch now starts with intent, not keywords
OpenAI’s March update turns a familiar e-commerce workflow inside out. Traditional shopping search asks the customer to translate a need into keywords, filters and category choices. ChatGPT instead invites the customer to describe the desired outcome, budget, style, constraints or use case, then refine the request conversationally. OpenAI’s release notes say users can browse products in chat, upload images to find similar items and compare options side by side on attributes such as price, reviews and features.
The practical distinction is important. A phrase such as “quiet cordless vacuum for a small apartment” contains multiple criteria that a keyword search page may split across filters, review pages and product specifications. Shopping research is designed for exactly these multi-constraint decisions. OpenAI says the feature can ask follow-up questions about brand, size, performance, comfort, style or price, then conduct a multi-step discovery process and return a buyer’s guide with top picks, trade-offs and merchant links. The interface is not merely retrieving products; it is interpreting the shopper’s hierarchy of preferences.
That design has been building for more than a year. In April 2025, OpenAI added shopping-oriented product cards with images, prices, ratings and merchant links to ChatGPT search. The Verge reported at the time that OpenAI described those recommendations as organic rather than sponsored. In November 2025, OpenAI introduced a more deliberate “shopping research” mode for harder product decisions, saying it could search across the web, incorporate feedback and produce a personalized guide.
The March 2026 experience therefore represents a progression from product cards to conversational merchandising. A product search result is a destination. A shopping conversation is an iterative process in which the system can keep changing the candidate set as the user changes priorities. That makes the assistant more useful, but it also gives the assistant more influence over which options remain visible long enough to be considered.
OpenAI’s checkout retreat clarifies where the real power sits
OpenAI initially aimed further down the funnel. On September 29, 2025, it launched Instant Checkout in the United States with Etsy sellers and announced plans to bring Shopify merchants into the same flow. The Agentic Commerce Protocol, co-developed with Stripe, was presented as an open standard that could connect AI agents, merchants and payment systems. Stripe described a model in which the merchant remained merchant of record while a secure token allowed the transaction to be initiated from the AI surface.
Six months later, the emphasis changed. OpenAI’s March 24 product-discovery announcement says the initial version of Instant Checkout “did not offer the level of flexibility” it wanted to provide merchants, so the company was allowing merchants to use their own checkout experiences while concentrating on discovery. Shopify’s announcement the same day describes ChatGPT users discovering products through Shopify Catalog and completing purchases in an in-app browser on mobile, while desktop users can be sent to the merchant’s site. Shopify also stresses that merchants remain merchant of record and retain the customer relationship and data.
This is a strategic retreat from owning the transaction, not from owning the decision point. Reuters reported in August that consumers still appeared more comfortable completing purchases on retailer sites and that Etsy users discovering products through ChatGPT typically returned to Etsy to buy. Retailers, meanwhile, have an obvious incentive to defend checkout because that is where loyalty enrollment, cross-selling, payment choices, returns, identity and first-party customer data become operational.
The lesson from the failed first version is not that conversational commerce lacks value. It is that the highest-value position may be one step earlier. If ChatGPT can become the place where a shopper decides “which one,” it can influence demand without assuming every tax, fraud, fulfillment, loyalty and brand-experience problem that comes with “buy now.” That division of labor is less dramatic than a fully autonomous shopping agent, but it may be easier to scale.
A product feed now competes with the product page
The visible chat interface hides a data problem. To recommend products accurately, ChatGPT needs structured information about price, variants, availability, attributes, promotions and merchant identity. OpenAI says it has expanded the Agentic Commerce Protocol beyond checkout so merchants can share product feeds and promotions for discovery. It also says retailers including Target, Sephora, Nordstrom, Lowe’s, Best Buy, The Home Depot and Wayfair have integrated with ACP for discovery.
Shopify has pursued the same objective through its catalog infrastructure. In March, it said millions of merchants could become discoverable across AI channels through Agentic Storefronts, with product data synchronized across surfaces. It also said products from Shopify merchants can appear in ChatGPT through Shopify Catalog without a separate integration from each individual merchant. For merchants, structured catalog quality is becoming a distribution asset, not just back-office hygiene.
OpenAI’s Help Center makes the ranking inputs more concrete. It says shopping results can draw on structured metadata from first- and third-party providers, as well as model responses and other contextual information. When the user supplies a hard constraint, such as a budget, the system can weight that criterion more heavily. Shopping research may also use merchant product data supplied through ACP alongside publicly available product information and other retail sources.
That creates a subtle change in what “optimization” means. In conventional search, a merchant wants a page that can rank and persuade a human after the click. In conversational discovery, the merchant also needs data that an intermediary can parse, normalize and compare before the customer ever sees the product page. A richly written brand story may still matter for persuasion, but it cannot compensate for missing sizes, stale inventory, ambiguous materials or inconsistent product identifiers if the assistant is deciding whether the item fits the request.
The product page is not disappearing. It is acquiring a machine-facing counterpart whose job is to make the product legible before the visit happens.
The economics favor whoever owns the recommendation layer
The commercial attraction of this position is already measurable, although the numbers should not be mistaken for proof that AI will replace search or marketplaces. Adobe reported that during the June 23–26, 2026 Prime Day event, visitors arriving at U.S. retail sites from generative-AI sources converted 40% better than visitors from non-AI channels such as paid search, email and social media. Adobe also stressed that AI traffic remained modest relative to larger channels.
Reuters reported a similar reason retailers are paying attention: Adobe data cited in its August 7 report found that AI-referred visitors generated 41% higher revenue per visit than shoppers arriving through traditional channels, while Ulta Beauty said shoppers coming through Gemini and ChatGPT showed “double the conversion and intent.” Those are company and analytics-provider observations rather than a universal market law, but they point to a plausible mechanism: users who have already spent time describing needs and comparing products may arrive closer to a decision.
That makes recommendation access economically valuable even when checkout stays elsewhere. The assistant can sit at the top of a high-intent funnel, while the merchant monetizes the conversion and keeps the transaction record. It is a different bargaining arrangement from a marketplace that controls both discovery and sale.
OpenAI is also building an advertising business around decision moments. On August 18, it announced the expansion of ChatGPT Ads to 31 European markets and said ads are intended to reach people while they are exploring, comparing and deciding. OpenAI states that those ads are clearly labeled, run separately from answers and do not influence ChatGPT’s responses; its shopping Help Center likewise says product results are selected independently and are not ads.
The distinction will be commercially and reputationally important. A user may see an organic product recommendation and a paid placement in the same broader session. If that separation remains credible, OpenAI can monetize attention without explicitly selling ranking inside shopping answers. If users come to suspect that commercial relationships shape the shortlist despite those rules, the value of the recommendation layer would weaken quickly.
Shoppers gain speed but inherit a new verification problem
The strongest case for ChatGPT shopping is reduction of research friction. A shopper can describe a problem once, add or remove constraints, compare candidates and ask follow-up questions without rebuilding the search from scratch. Shopping research can surface products as it works and allow the user to reject options or request similar alternatives in real time. That continuity can turn a scattered browsing session into a single decision process.
But convenience changes where the verification burden appears; it does not eliminate it. OpenAI warns that prices, stock and discounts can change and that shopping research can make mistakes. Its Help Center tells users to confirm final price, taxes, fees and shipping on the retailer’s site before buying. It also says product prices shown initially may come from the first listed merchant and may not be the lowest available price.
Reviews require similar caution. OpenAI says review summaries may be generated from public websites and that displayed reviews and ratings are not verified by OpenAI. Labels such as “Budget-friendly” or “Most popular” are generated from information available to the model and are not guarantees or comprehensive market statements. A confident-looking comparison can therefore combine current structured data, inferred labels and unverified third-party review signals in one visual frame.
An early example showed why this matters. When The Verge tested a pre-release version of ChatGPT’s shopping improvements in April 2025 with the then-unreleased Nintendo Switch 2, the system surfaced questionable information about availability and third-party sellers. The reporter treated that as a reminder to double-check shopping results. OpenAI later built a shopping-specific research model and has continued to improve product-data freshness, but its own current documentation still acknowledges errors.
For low-stakes purchases, the time saving may easily outweigh the risk. For expensive electronics, safety-sensitive products or purchases where compatibility matters, the best use of the assistant is to narrow the field, then verify the decisive facts at the merchant or manufacturer.
Retailers face a visibility problem before a checkout problem
Retailers worried about AI commerce often focus on who owns the payment button. The more immediate problem is whether the assistant can understand and retrieve the product at all. Reuters reported that retailers are updating their websites so products rank more effectively in chatbot searches, while still preferring to complete transactions on their own sites. That is a distribution challenge before it is a payments challenge.
The pressure is reinforced by the way AI systems ingest product information. OpenAI’s discovery stack uses merchant feeds, structured metadata, public product information and third-party sources. Shopify says AI surfaces favor structured data with clean attributes and real-time accuracy. A merchant with a beautiful storefront but inconsistent catalog data can become less visible at the exact moment the customer asks an assistant to choose.
Specialist shopping companies have argued that general assistants also face data-quality limits. When OpenAI and Perplexity launched deeper shopping assistants in November 2025, TechCrunch reported executives from vertical shopping startups arguing that fashion, furniture and other categories benefit from domain-specific product data and merchandising logic that generic systems may not capture as well. That criticism does not negate ChatGPT’s reach; it identifies where specialist datasets can still outperform a broad interface.
Retailers therefore have two defensive assets. The first is authoritative product data that assistants can interpret. The second is proprietary customer knowledge that does not necessarily travel with the referral. Reuters quoted retail executives emphasizing that their own sites provide richer information about browsing, baskets, past purchases and loyalty behavior.
This explains why OpenAI’s March pivot may suit both sides better than its original checkout ambition. ChatGPT can become a powerful discovery channel without forcing every retailer to hand over the entire customer journey. Merchants still have to compete for inclusion in the shortlist, but once the click arrives they can use loyalty, merchandising and service to deepen the relationship.
Google’s rival stack shows the market is still unsettled
OpenAI is not defining conversational commerce alone. Google introduced the Universal Commerce Protocol in January 2026 with Shopify, Etsy, Wayfair, Target and Walmart, presenting it as an open standard spanning discovery, buying and post-purchase support. Google said the protocol would support checkout on eligible AI Mode and Gemini listings while keeping retailers as seller of record.
By May, Google had pushed further into transaction orchestration. Its Universal Cart was designed to work across retailers and Google services, with some purchases completed through Google Pay and others transferred to merchant sites. Google also announced Merchant Center tools intended to help brands understand their visibility on AI surfaces. The competitive direction is clear even if the winning architecture is not: AI assistants want to own more of product discovery, while merchants want interoperability without losing identity or data.
That tension helps explain why protocols matter. Stripe’s ACP work framed the problem as avoiding bespoke integrations for every agent while keeping merchants in control of catalog presentation, payment processing and fulfillment. Google’s UCP makes a similar interoperability argument but extends across a broader commerce workflow. The industry is therefore experimenting with multiple standards and multiple places to complete checkout rather than converging on one universal agent.
TechCrunch argued in October 2025 that apps, product discovery and Instant Checkout gave OpenAI many of the pieces needed to become a major commerce gateway, potentially competing not only with other AI companies but with established shopping platforms. The March 2026 pullback from native checkout narrows that ambition in the short term, but it does not remove the competitive threat to search engines and marketplaces.
For shoppers, this competition may improve choice and convenience. For retailers, it means there is no safe assumption that one feed, one ranking model or one conversational surface will dominate. The practical requirement is portability: clean product data, reliable inventory, consistent identifiers and a checkout experience that can receive high-intent traffic from several AI channels.
The winning merchant playbook is data discipline, not chatbot tricks
The first decision for a merchant is whether its product catalog can answer the questions an assistant needs to answer. That means complete titles, unambiguous variants, accurate dimensions, materials, compatibility, inventory, current prices, shipping terms and durable product identifiers. OpenAI’s own documentation says shopping results use structured metadata and merchant data; Shopify says synchronized, accurate catalog information is what lets products appear across AI surfaces. Catalog operations are now part of customer acquisition.
The second decision is measurement. Retailers should distinguish AI-referred visits from other acquisition channels, then compare conversion, average order value, return rates and new-versus-existing customer mix. Adobe’s Prime Day data suggests AI-referred users can arrive with unusually high purchase intent, but Adobe also says the channel is still small compared with established sources. That combination argues for instrumentation before large speculative spending.
The third decision is to protect what the assistant cannot easily replicate. Retailers cited by Reuters want discovery traffic without losing their direct relationship because loyalty programs, customer service, personalization and first-party behavior data can compound after the first transaction. A merchant should therefore treat the AI referral as the beginning of a relationship rather than a one-off anonymous conversion.
The fourth is not to confuse generative-engine visibility with conventional promotional copy. An assistant deciding among products needs evidence that maps to user constraints. A concise specification, authoritative compatibility statement, clear return policy or verified material description can be more useful than a page filled with vague superlatives. This is partly an editorial inference from the retrieval mechanisms OpenAI and Shopify describe, but it follows directly from the system’s need to compare products against explicit requirements.
Brands should still invest in human persuasion. The emerging division of labor is machine-readable truth before the click and differentiated brand experience after it.
Discovery is the durable layer until shoppers prove they want delegation
The evidence available in August 2026 supports a narrower judgment than the most ambitious “agentic commerce” forecasts. Consumers are clearly using AI systems to research products, and AI-referred traffic can be commercially valuable. Adobe’s 2026 Prime Day data and retailer comments collected by Reuters both point to high-intent behavior among at least some AI-referred shoppers.
At the same time, OpenAI’s own product history shows that moving from recommendation to transaction is not a trivial extension. The company launched Instant Checkout in September 2025, then said in March 2026 that the initial implementation did not provide the merchant flexibility it wanted and shifted its focus toward discovery. That is strong evidence against assuming that the winning AI shopping experience must own every step.
The durable advantage for ChatGPT is likely to be the compression of choice. Natural-language intent, persistent context, image input and side-by-side comparison can remove substantial friction before the user is ready to buy. The assistant becomes most valuable at the point where a market containing hundreds of plausible products is reduced to three or four candidates that fit a specific person’s constraints.
The thesis would weaken if three things happen: shoppers become comfortable delegating purchases at scale, merchant-controlled AI experiences match general assistants on discovery, or product-data errors make users distrust generated shortlists. It would strengthen if retailers continue optimizing for AI referrals, if structured product feeds become more comprehensive, and if high-intent referral economics persist across categories and markets.
For now, the important change is not that ChatGPT can sell a product. It is that more shopping decisions can begin before the shopper visits a store, marketplace or search-results page. Whoever earns trust at that earlier stage gains influence over demand; whoever owns the transaction still has a chance to own the customer.
Questions shoppers and retailers are asking now
Yes. OpenAI says ChatGPT shopping can use budget, preferences and other constraints to surface products, and shopping research can ask follow-up questions to refine the result.
Yes. OpenAI’s March 24, 2026 shopping update added image-based product discovery so users can upload an image as inspiration and ask for visually similar items.
Yes. Current shopping research documentation says products can be compared side by side on attributes such as price, features, reviews and other relevant details.
OpenAI says product results are selected independently and are not ads. It also says ChatGPT Ads are clearly labeled, separate from answers and do not influence the answers ChatGPT provides.
No. OpenAI says the price shown in an initial product listing may come from the first listed merchant and may not be the lowest price available. Price and shipping data can also lag retailer updates.
No. OpenAI says review summaries may be model-generated from public reviews, ratings are not verified by OpenAI, and labels such as “Budget-friendly” are not guarantees.
OpenAI’s current Help Center says some eligible products and merchants may still show an Instant Checkout option, but its March 2026 product announcement says the company shifted its main effort toward product discovery and merchant-controlled checkout experiences.
Retailers should maintain accurate, structured product data and keep key attributes, prices, variants and availability current. OpenAI and Shopify both describe structured merchant data as a core input to AI product discovery.
No. Google has built conversational shopping, the Universal Commerce Protocol and Universal Cart features across Search and Gemini, while other AI companies and specialist shopping startups are also competing in the category
Author:
Jan Bielik
CEO & Founder of Webiano Digital & Marketing Agency
This article is an original analysis supported by the sources cited below
Powering Product Discovery in ChatGPT
OpenAI’s March 24, 2026 announcement established the current discovery-first strategy, image search, side-by-side comparison, ACP product feeds and the pullback from the initial Instant Checkout design.
The Help Center documented how products are selected, how prices and labels are presented, the separation of ads from product results and current limitations around price freshness and reviews.
Using shopping research in ChatGPT
This source documented the interactive research workflow, follow-up questions, ACP and public-data inputs, buyer’s guides, comparisons, privacy and verification guidance.
The release notes independently dated the March 24, 2026 shopping update and confirmed image-based similarity search and side-by-side comparison as product features.
Introducing shopping research in ChatGPT
OpenAI’s November 24, 2025 launch explained the shopping-specific research model, personalized buyer’s-guide concept and acknowledged limitations around price and availability accuracy.
Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol
This September 29, 2025 announcement established the original native-checkout strategy and OpenAI’s initial framing of ACP as infrastructure for AI-mediated transactions.
ChatGPT Ads expands across Europe
OpenAI’s August 18, 2026 announcement established the European ads expansion and its policy that ads remain clearly labeled, separate from answers and unable to influence responses.
Millions of merchants can sell in AI chats
Shopify documented the March 2026 Agentic Storefronts rollout, ChatGPT discovery through Shopify Catalog, merchant-controlled checkout and retention of merchant-of-record status and customer relationships.
Developing an open standard for agentic commerce
Stripe explained the original ACP architecture, merchant control, secure transaction flows and the interoperability rationale behind agentic-commerce standards.
New tech and tools for retailers to succeed in an agentic shopping era
Google’s January 2026 announcement documented UCP, its retail partners and Google’s model for AI-assisted checkout while keeping retailers as seller of record.
Google’s May 2026 update documented Universal Cart, merchant-site handoff options and tools for measuring brand visibility across AI shopping surfaces.
AI shoppers convert 40% better as Prime Day hits $26.4B
Adobe provided June 2026 U.S. retail analytics showing stronger conversion from generative-AI referral traffic during the Prime Day period while noting that AI traffic remained a modest channel.
Retailers tap AI shopping traffic but fight to keep customer data
Reuters’ August 7, 2026 reporting supplied current retailer behavior, conversion observations, the importance of customer data and confirmation of OpenAI’s March checkout retreat.
OpenAI partners with Etsy, Shopify on ChatGPT payment checkout
Reuters documented the September 2025 Instant Checkout launch, merchant fees and the commercial significance of OpenAI’s original move into transaction handling.
TechCrunch provided independent context on specialist shopping startups, domain-specific data advantages and competitive pressure around AI-assisted product research.
OpenAI and the race for AI-driven commerce
TechCrunch supplied independent analysis of OpenAI’s 2025 commerce stack and the competitive implications of ChatGPT becoming a gateway to product discovery and transactions.
ChatGPT is getting better for shopping
The Verge documented the April 2025 shopping-card rollout and an early test that exposed product-availability and seller-quality errors, supporting the article’s verification cautions.
| Citing this article? Brief excerpts are welcome. Please credit Webiano.digital, name the author where stated, and include a link to https://webiano.digital and to this original article. Full or substantial republication requires prior written permission. Read our Copyright and Content Use Policy. |
This article was prepared with the assistance of artificial intelligence tools. The content underwent expert human review, and Webiano Digital & Marketing Agency assumes editorial responsibility for its final version and publication.
