ChatGPT Payment Pause Doesn't Signal an AI Shopping Retreat: How Brands Can Win AI Recommendations

OpenAI launched direct payment within ChatGPT in 2025, which was once seen as a milestone for "agentic commerce." But according to a March 2026 report by Digital Commerce 360, OpenAI has moved Instant Checkout from the chat interface into individual stores' apps, with the official rationale being to "make search and product discovery work better in ChatGPT." This is not the end of agentic commerce but a key divergence: AI handles recommendations and filtering, while the final transaction still happens on the brand's own website.

From "in-chat payment" to "in-chat discovery": OpenAI's strategic pivot

In September 2025, OpenAI launched Instant Checkout, allowing US users to complete purchases without leaving ChatGPT. The first partners included Etsy, and it claimed that over 1 million Shopify stores would be integrated. Alongside it was the Agentic Commerce Protocol, an open protocol co-developed with Stripe.

Six months later, however, plans changed. OpenAI explained to industry media: "Instant Checkout will move into the apps, where purchasing can happen more smoothly. We're prioritizing making search and product discovery work well in ChatGPT." The protocol was not abolished; it remains as infrastructure connecting users and stores. Large chains like Target already have their own apps within ChatGPT. What was shelved was only the idea that "all purchases begin and end in chat." In the same month, Shopify by default made its stores' products discoverable in ChatGPT—discovery in AI, transaction in the store.

This adjustment reveals a reality: the AI conversational interface is not a universal checkout counter. Consumers trust AI for recommendations, but the payment step involves security, after-sales service, habits, and multiple other factors, and migration has been much slower than expected. OpenAI chose to play to its strengths and concentrate resources on what it does best—helping users find products.

Agentic Commerce Protocol is an open protocol co-developed by OpenAI and Stripe, aimed at establishing a standardized way to exchange transactions and product information between AI agents and merchant systems. Its core idea is to enable AI assistants to complete purchases on behalf of users in a secure and auditable manner, including product queries, order placement, and payment authorization. The protocol is not tied to the ChatGPT interface but serves as underlying infrastructure that can be integrated by different AI agents and merchant platforms. After OpenAI moved Instant Checkout out of the chat interface, the protocol continues to exist precisely because it addresses "how AI talks to commerce systems," not "where payment happens." Understanding this makes it clear that OpenAI's adjustment is a channel strategy change, not an abandonment of the technical foundation of agentic commerce.

Data speaks: AI traffic is small but converts better

Although in-chat payment has been paused, data on AI-driven traffic to e-commerce sites continues to grow. Adobe Analytics, based on over 1 trillion visits to US e-commerce sites during the 2025 holiday shopping season, shows:

• Traffic from AI assistants increased 693% year over year;
• This traffic converted 31% better than overall store traffic;
• Bounce rate was 33% lower.

It should be noted that this is a study of a single market in a single quarter, and the direction matters more than the specific numbers. The signal is clear: visitors who arrive at a store from an AI recommendation have already done their homework and have stronger purchase intent. If brands focus only on payment protocols and ignore the AI recommendation stage, they are handing high-intent customers to competitors.

What consumers want is "advice," not "buying on their behalf"

Sopra Steria surveyed 8,400 consumers across eight European countries (including Spain) in February 2026, asking in which shopping scenarios they would be willing to let an AI agent make decisions on their behalf. The results show that even for the most accepted categories, the proportion did not exceed half:

• Electronics: 45%
• Energy bills: 39%
• Insurance: 36%
• Financial services: 22%
• Health or food: 16%

In the Spanish portion of the survey, the vast majority of respondents wanted AI agents to only provide options, with the final decision remaining in their own hands. Today's consumers want an assistant that helps narrow down choices, not a robot that pays for them. This explains why Instant Checkout had limited appeal and shows that a brand's visibility in AI recommendations has a more direct impact on conversion than payment interface compatibility.

AI Agent refers to an AI system that can autonomously perceive its environment, make plans, and execute multi-step tasks, unlike a chatbot that only answers questions. In shopping scenarios, AI agents can search for products, compare prices, fill out forms, and even complete payment on a user's behalf. Consumers' limited acceptance of "buying on their behalf" stems from authorization boundaries, liability attribution, and correction costs: once an agent buys the wrong item or a dispute arises, the user bears the consequences. Therefore, most current AI shopping products limit agent capabilities to "information aggregation and filtering," returning the final decision to click buy to the user. This also explains why in the Sopra Steria survey, even for the most accepted categories, the proportion willing to let AI agents make decisions entirely on their behalf did not exceed half.

Agents are stepping out of the chat box

OpenAI moving payment out of chat does not mean no one uses AI to shop. In September 2026, OpenAI launched the personal agent Dots, Meta launched Muse (using one-time virtual card payments via Stripe's Link), and agents such as Grok Bot, Instinct, and Tab adopted similar models. They do not buy directly in conversation but open a browser and complete the transaction on the store's website.

This precisely confirms the core judgment of this article: AI determines the shortlist, and the sale is completed on the brand's official website—even if the final visitor browsing the site is an agent. For brands, this means the readability of the official website and the degree of structuring of product data directly affect whether AI can correctly understand and recommend your products.

What it means for brands: three things determine success

If AI handles discovery and the official website handles conversion, then the battle for the store is spread across two fronts, and brands can fully control only one of them.

First, in the conversation: get AI to mention you when users ask. If you cannot enter the AI's shortlist, subsequent traffic will never happen. This requires brands to manage visibility in AI engines, rather than relying solely on traditional SEO.

Second, on the official website: let visitors from AI find what was promised. Price, inventory, and delivery time must be clear at a glance. What AI says in its recommendation and what users see after arriving must match; otherwise, even high-intent traffic will be lost.

Third, in the middle layer: let AI understand your website and product data. Whether robots.txt allows AI crawlers access and whether pages contain structured data—these technical details determine whether AI's description of you is accurate.

A quick self-check: use your web analytics tool to see how much traffic comes from chatgpt.com, gemini.google.com, or perplexity.ai, and compare the conversion rate of that traffic with overall traffic. According to GenScore, its tool can detect the proportion of times a brand appears in answers from ChatGPT, Gemini, Claude, and Perplexity, audit basic configurations of robots.txt and structured data, and also offers a free agentic commerce checker.

Structured Data refers to a machine-readable format that annotates webpage content (such as product name, price, inventory, and ratings) using standard vocabularies like Schema.org, usually embedded in pages via JSON-LD. AI crawlers and search engines rely on structured data to accurately understand product attributes rather than guessing from human-readable HTML. If a brand's website lacks structured markup, AI may miss key information or produce incorrect descriptions when generating recommendations, causing a mismatch between the recommendation and what users see upon arrival. Alongside structured data is the access permission setting for AI crawlers (such as GPTBot and Google-Extended) in robots.txt; together they determine whether AI can "read" and "cite" brand products.

The battle for the shortlist happens before the visit

In-chat payment may return in another form in the future, perhaps through e-commerce platforms or under a different name. But the part already operating in markets like Spain is precisely the part most dependent on brands' own efforts: getting AI to recommend you. A brand that waits until payment protocols become widespread before acting will have already lost all the customers who asked AI and did not see it.

The competition for AI recommendation slots is essentially a fight for the entry point of user decisions. As consumers become more accustomed to asking AI before buying, a brand's presence in AI answers is the new shelf position. This shelf is not owned by the platform, but brands must actively compete for it.

Generative Engine Optimization (GEO) is an extension of traditional SEO in the era of generative AI. Its goal is not to rank webpages higher in search results but to have brand information adopted and cited by large language models when generating answers. Its methods include ensuring content is crawlable by AI crawlers, organizing product information in a Q&A structure, building brand mentions in authoritative sources, and monitoring the frequency of brand appearances in answers from ChatGPT, Gemini, Perplexity, and others. GEO's competitive logic differs from SEO—AI answers usually list only a few candidates, and brands that do not make the shortlist are eliminated at the user's decision entry point, and this process happens before users visit the website, making it difficult for brands to recover through on-site optimization.

One-line Summary

OpenAI moves in-chat payment out of ChatGPT, splitting agentic commerce into AI recommendation and official-site conversion.

Paragraph Summary

OpenAI launched Instant Checkout payment within ChatGPT in 2025, but six months later moved it to store apps and prioritized product discovery. This adjustment is not the end of agentic commerce but a split into a two-stage model: "AI handles recommendation and filtering, brand official websites handle conversion." Adobe data shows AI-driven traffic grew 693% year over year and converted 31% better; a Sopra Steria survey shows limited consumer acceptance of AI agents buying on their behalf, with the highest category at only 45%. Agents such as OpenAI Dots and Meta Muse also complete purchases through browsers on brand websites. Brands need to align on AI recommendation visibility, official website information consistency, and technical readability to capture the traffic dividend.