OpenAI has launched a dedicated “shopping research” interface powered by a specialized version of its unreleased GPT-5 mini model. Available immediately to all users, the tool leverages reinforcement learning to curate personalized buyer’s guides.
It aims to replace traditional search engines with interactive, conversational discovery. However, unlike aggressive new tools from Google and Perplexity, OpenAI’s offering currently lacks direct checkout capabilities. This forces users to leave the platform to complete purchases – a strategic pause in its ambition to control the entire transaction loop.
GPT-5 Mini for A ‘Research-First’ Architecture
Under the hood, the system relies on “GPT-5 mini,” a previously unreleased iteration of OpenAI’s flagship model family specifically fine-tuned for commerce.
Unlike general-purpose models that often hallucinate product specs or invent pricing, this version utilized a training regimen focused heavily on reinforcement learning tailored to shopping tasks. Engineers prioritized building an engine capable of reading trusted sites, citing reliable sources, and synthesizing multi-source data into a coherent narrative.
Introducing shopping research, a new experience in ChatGPT that does the research to help you find the right products.
It’s everything you like about deep research but with an interactive interface to help you make smarter purchasing decisions. pic.twitter.com/jksGVpCXGm
– OpenAI (@OpenAI) November 24, 2025
According to the company’s official announcement:
“Shopping research is powered by a version of GPT-5 mini trained with reinforcement learning specifically for shopping tasks. We trained it to read trusted sites, cite reliable sources, and synthesize information across many sources to produce high-quality product research.”
“We also designed it to be an interactive experience that could update and refine its research in real time – incorporating new constraints and adjusting to feedback on user product preferences – resulting in a response that feels both well-researched and personalized.”
Such specialized training has yielded measurable gains in reliability, says OpenAI. Internal benchmarks reveal a significant improvement in the model’s ability to parse complex requests.
OpenAI’s latest model accurately identifies items matching all user criteria 64% of the time, a significant improvement over the 37% success rate observed in previous ChatGPT product queries.
Rather than a simple list of links, the model generates a structured “buyer’s guide” that highlights trade-offs, specs, and comparisons.
OpenAI frames this shift as moving beyond simple queries, noting that “shopping research is built for that deeper kind of decision-making. It turns product discovery into a conversation: asking smart questions to understand what you care about, pulling accurate, up-to-date details from high-quality sources, and bringing options back to you to refine the results.”
Visually, the tool abandons the text-only chat paradigm for an interactive UI that displays product cards and images directly in the stream. A feedback loop allows users to refine results using a “Tinder-like” mechanism, clicking “Not interested” or “More like this” to retrain the session in real-time.
Leveraging ChatGPT’s “Memory” feature, the system recalls historical preferences – such as a user’s interest in gaming – to contextualize future searches, prioritizing items like high-refresh-rate monitors without needing explicit prompts.
The Checkout Gap: Why OpenAI Can’t ‘Buy’ Yet
Despite the “agent” branding, the tool currently cannot execute purchases. Users must click outbound links to retailer websites to complete transactions, a limitation that stems from the fragmented nature of online payments.
Without a standardized commerce protocol, the AI cannot reliably navigate diverse third-party checkout flows. An OpenAI spokesperson explained the risk, stating that “without an integration, the model would be guessing its way through a checkout flow.”
Strategic considerations also influenced this decision. OpenAI has chosen to prioritize global availability and broad product discovery over deep, vertical integration with a few select partners at launch.
According to the company, launching globally first ensures that “everyone can shop anywhere, while instant checkout expands as more merchants adopt the integration.”
Previously teased via high-profile partnerships with PayPal and Walmart , the “Instant Checkout” feature remains in development with no confirmed release date.
In a move to build trust, OpenAI confirmed that user chat data is not shared with retailers, a key differentiator from ad-driven platforms.
Relying on public web crawling, the system requires merchants to actively “allowlist” OpenAI’s bots to ensure their real-time inventory and pricing are visible.
The Agentic Battlefield: Google, Amazon, and the Fight for the Transaction
While competitors rush to close the transaction loop, OpenAI’s approach is notably less disruptive to the status quo. By sending traffic downstream to retailers, the company avoids the immediate legal and commercial backlash that has plagued rivals.
Google recently escalated the war by deploying agents capable of physically calling stores to check inventory and executing purchases via Google Pay. Google’s recent deployment of AI shopping agents marked a significant step toward fully automated commerce.
Rival engine Perplexity offers a native “Buy with Pro” checkout but has faced significant headwinds. Amazon hits Perplexity with a cease-and-desist, highlighting the legal risks of bypassing ad ecosystems, as major retailers increasingly block third-party AI crawlers to protect their advertising moats.
Fundamentally, the absence of a “walled garden” model distinguishes OpenAI’s current strategy. Major retailers like Amazon are aggressively defending their data, creating a fragmented landscape where only authorized agents can operate effectively.
Microsoft, a key investor in OpenAI, is pursuing a parallel B2B strategy. The Microsoft Personal Shopping Agent empowers retailers to build their own branded AI storefronts, contrasting with the consumer-facing aggregation models of Google and Perplexity.
Economically, the stakes are high. This shift represents a fundamental threat to the traditional search-ads business model, moving value capture from the “search bar” to the “agent’s wallet.”
For now, OpenAI’s “safe” strategy of acting as a research assistant rather than a buyer positions it as a neutral arbiter in a market defined by aggressive vertical integration.
