For about five months, buying a pair of running shoes without leaving a chat window was a real option for millions of ChatGPT users. Then, on March 4, 2026, OpenAI quietly pulled the plug on Instant Checkout, the feature it had launched on September 29, 2025 to let people tap “Buy,” confirm shipping, and pay without opening a browser tab. That single retreat reframed the entire fight over agentic commerce — the push to let AI chatbots not just recommend products but actually complete the purchase. Now four companies are running four different playbooks. OpenAI is rebuilding around its Apps model. Microsoft is testing Copilot Checkout with retailers instead of a single in-chat button. Google is betting on a portable protocol called UCP that works across Search and Gemini. And Anthropic, arriving last, just handed retailers a blueprint for building their own shopping agents on Claude rather than building a consumer checkout button at all.
This comparison breaks down how ChatGPT, Microsoft Copilot, Google Gemini, and Claude actually handle agentic commerce as of September 2026 — not the marketing pitch, but the protocols, retailer counts, payment rails, and shutdown history behind each one. If you searched for an ai shopping assistant comparison or wanted to understand what agentic commerce actually means for your next purchase or your company’s e-commerce stack, this is the breakdown with the numbers attached.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Agentic Commerce Actually Means in 2026
Agentic commerce describes AI systems that do more than answer a product question — they carry a shopping instruction forward, connecting to product catalogs, shopping carts, checkout systems, and sometimes payment credentials to finish a purchase with minimal human clicking. That is a meaningfully different job than a chatbot that just links you to a retailer’s website. A true AI shopping assistant needs four things working together: a way to discover relevant products, a structured feed of pricing and inventory, a checkout path that an AI agent (not just a human browser) can call, and a fraud and payment layer that can tell the difference between an authorized agent purchase and a hijacked account.
Two competing protocols have emerged to standardize that plumbing. The Agentic Commerce Protocol (ACP), built by OpenAI and Stripe, ties merchant product feeds and Stripe-based payments directly into ChatGPT’s shopping surfaces. The Universal Commerce Protocol (UCP), backed by Google alongside retail partners including Shopify, Etsy, Walmart, and Target, aims to be protocol-agnostic — a standard that any AI agent, not just Gemini, could implement for discovery through checkout and even post-purchase support. Whether one of these becomes the de facto standard the way HTTP did for the web, or whether retailers end up supporting both, is the single biggest open question in agentic commerce right now, and it explains most of the differences in the comparison table below.
Why Retailers Are Racing Into Agentic Commerce Now
The timing of these launches isn’t random. Anthropic released Claude Commerce Agents on September 2, 2026, explicitly ahead of the holiday shopping season, and Microsoft’s Copilot Checkout expansion plans target the same fourth-quarter window. Retailers have watched search behavior shift toward conversational AI tools for product research over the past two years, and the fear driving every one of these announcements is the same: if a shopper starts their research inside ChatGPT, Gemini, or Copilot instead of a search engine results page, a retailer that isn’t represented in that AI agent’s product feed effectively doesn’t exist for that shopper. That is a much bigger threat to e-commerce businesses than losing a single sale — it’s losing the discovery step entirely.
That explains why Shopify shows up as a partner across three of the four platforms in this comparison — Copilot’s Brand Agents, UCP’s launch backers, and (indirectly) as one of the payment options in Copilot Checkout’s processor stack. Shopify doesn’t want its merchant base locked out of any single AI shopping surface, so it’s hedging by supporting multiple protocols simultaneously rather than picking a side in the ACP-versus-UCP fight. Individual merchants without that kind of platform-level flexibility face a harder decision, which is exactly why the migration guide later in this piece focuses on protocol-agnostic groundwork — structured feeds and API-based checkout — that pays off regardless of which chatbot ends up winning the largest share of AI-driven shopping traffic.
ChatGPT and Instant Checkout: The Rise and Fall of In-Chat Buying
OpenAI moved first. Instant Checkout launched inside ChatGPT on September 29, 2025, letting shoppers tap “Buy,” confirm order details, shipping, and payment, and complete a purchase without leaving the conversation. It ran on the Agentic Commerce Protocol, with Stripe handling payment processing and fraud review for agent-initiated transactions. For a moment, ChatGPT looked like it might become a genuine shopping destination in its own right, backed by a user base OpenAI has previously cited in the hundreds of millions of weekly active users.
Instant Checkout was discontinued on March 4, 2026 — 156 days after launch. OpenAI told trade outlet Modern Retail the feature was moving to its Apps model, shifting the shopping experience toward ChatGPT’s broader third-party apps ecosystem rather than a standalone buy button. Commerce analysts have since framed this as part of a wider industry pattern dubbed “discover in AI, buy on site”: chatbots are proving very good at product discovery and comparison, but actual payment and order completion keep drifting back to merchant-owned checkout pages, according to a mid-2026 analysis from Digital Applied’s review of the stalled AI checkout trend. The Agentic Commerce Protocol itself has not been shut down — Stripe and OpenAI continue to describe it as a live standard — but its flagship consumer-facing implementation, Instant Checkout, is gone for now.
Microsoft Copilot Checkout: Betting on Retailers Over a Single Buy Button
Microsoft took a different route. Copilot Checkout began testing on Copilot.com in the US, letting shoppers complete purchases without leaving the conversation, with products surfaced from partner retailers directly inside Copilot’s results. Payment processing draws on a mixed stack of PayPal, Shopify, and Stripe rather than a single processor, and Microsoft has said it plans to extend Copilot Checkout across Bing, MSN, and Edge as the rollout matures, according to Thurrott’s reporting on the Copilot shopping rollout.
The more interesting piece of Microsoft’s strategy is Brand Agents — AI assistants that Shopify merchants can deploy directly on their own storefronts, trained on that merchant’s product catalog and speaking in the merchant’s brand voice. Instead of centralizing shopping inside one Microsoft-owned interface, Copilot becomes an embedded layer that helps shoppers find products, complete purchases, and get upsell recommendations on the retailer’s own site. Industry commentary has summarized the split bluntly: ChatGPT has the users, Copilot has the retailers. If you use Microsoft’s coding-focused Copilot day to day, it’s worth noting this is a completely separate product track from Copilot’s developer tooling comparisons — Copilot Checkout lives inside the consumer assistant, not the IDE.
Google Gemini and the Universal Commerce Protocol
Google’s approach avoids building a single “buy in Gemini” button and instead pushes a protocol that spans its entire AI surface. The Universal Commerce Protocol is designed to cover discovery, checkout, and post-purchase support, and it already counts Shopify, Etsy, Walmart, and Target among its retail backers. According to Google’s own commerce and ads blog, UCP will power a new checkout feature on eligible product listings inside AI Mode in Search and the Gemini app, letting US shoppers complete a purchase from participating retailers as they research — without ever opening a dedicated shopping app.
The strategic logic is different from OpenAI’s. Rather than concentrating agentic commerce inside the Gemini chat interface alone, Google is tying checkout to Search itself, where the bulk of product-research traffic already lives. That means Gemini’s shopping experience inherits Google’s existing advertising and product-listing infrastructure rather than starting from scratch, and it explains why UCP is described as portable — in principle, any AI agent that implements the protocol could use it, not just Gemini. For a broader look at how Gemini stacks up against rival assistants outside of shopping, see our GPT-6 Astra vs Claude Opus 5 vs Gemini 3.8 Flash comparison.
Claude Commerce Agents: Anthropic’s Merchant-First Blueprint
Anthropic arrived last, and it arrived with a fundamentally different pitch. On September 2, 2026, the company released Claude Commerce Agents, an open blueprint for building shopping agents and merchant-operations agents on top of Claude, timed deliberately ahead of the 2026 holiday shopping season as retailer demand for conversational search tools grows, per The Register’s coverage of the Claude Commerce Agents launch. Crucially, Anthropic is not building a consumer-facing “buy it in Claude” button at all. Claude Commerce Agents does not place orders or charge a customer’s card. It is explicitly a set of reference implementations — shopper agents that make preference-based product suggestions and add items to a cart, and merchant agents that help retailers manage inventory, pricing, and marketing campaigns.
The blueprint ships with runnable vertical examples for retail, travel, and ticketing, a safety harness, evaluation patterns, and deployment paths across the Claude API, Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI. Catalogs, checkout, payment, and fulfillment stay on the merchant’s own infrastructure — Anthropic positions Claude purely as the intelligence and runtime layer inside a retailer’s existing experience, not as a new shopping destination competing with the retailer. That is the opposite bet from OpenAI’s original Instant Checkout, and arguably the more conservative one after watching that feature get shut down. If you’re evaluating Claude’s broader model lineup for building this kind of agent, our breakdown of Claude Fable 5.1 and Mythos 5.1 covers the current model generation Anthropic expects developers to build commerce agents on.
Specs Comparison: How the Four Shopping Agents Stack Up
The table below lines up the four platforms on the specifics that actually determine whether a shopper or a retailer can use them today, rather than the future-tense roadmap language each company uses in its own announcements.
| Spec | ChatGPT (OpenAI) | Copilot Checkout (Microsoft) | Gemini / UCP (Google) | Claude Commerce Agents (Anthropic) |
|---|---|---|---|---|
| Protocol | Agentic Commerce Protocol (ACP) | Mixed: ACP-compatible feeds + proprietary integrations | Universal Commerce Protocol (UCP) | No proprietary checkout protocol — API-first blueprint |
| Consumer buy button | Discontinued (Instant Checkout ended March 4, 2026) | Live in testing (US, Copilot.com) | Rolling out via AI Mode in Search + Gemini app (US) | None — not a consumer feature |
| Launch date | September 29, 2025 (Instant Checkout) | 2026 (public testing phase) | 2026 (UCP checkout rollout announced) | September 2, 2026 |
| Current status | Shutdown; migrating to Apps model | Active US testing | Active rollout to eligible listings | Active; open blueprint/reference code |
| Payment processor(s) | Stripe | PayPal, Shopify, Stripe | Google Pay-linked retailer rails | None — merchant’s own payment stack |
| Confirmed retail partners | Merchants onboarded via Stripe/ACP feeds | Shopify merchants (Brand Agents) | Shopify, Etsy, Walmart, Target | Reference implementations only; no named retail partners yet |
| Who owns checkout | OpenAI (when live) / merchant now | Merchant site + Copilot surface | Retailer, surfaced via Google | Merchant, entirely |
| Places orders autonomously | Yes (when Instant Checkout was live) | Yes, with user confirmation | Yes, with user confirmation | No — explicitly out of scope |
| Deployment surfaces | ChatGPT web + mobile | Copilot.com, expanding to Bing, MSN, Edge | Google Search (AI Mode), Gemini app | Claude API, Amazon Bedrock, Microsoft Foundry, Google Cloud Vertex AI |
| Merchant integration effort | Structured product feed + ACP checkout API | Shopify app install or PayPal/Stripe integration | UCP feed implementation | Build your own agent using Anthropic’s reference code |
| Fraud/agent-transaction handling | Stripe-managed fraud review for agent purchases | Processor-dependent (PayPal/Shopify/Stripe) | Retailer-managed, Google-assisted | Entirely merchant-managed |
| Underlying assistant reach | Hundreds of millions of weekly ChatGPT users | Growing Copilot consumer + enterprise base | Tied to Google Search’s global reach | Enterprise/developer audience, not consumer |
Timeline: How the Agentic Commerce Race Unfolded
Laid out chronologically, the sequence of launches and reversals makes the volatility of this category obvious — a full product cycle, from launch to shutdown, has already played out in less than a year.
| Date | Event |
|---|---|
| September 29, 2025 | OpenAI launches Instant Checkout in ChatGPT, built on the Agentic Commerce Protocol with Stripe as payment processor |
| Late 2025 – early 2026 | Microsoft begins testing Copilot Checkout on Copilot.com in the US |
| Early 2026 | Google formalizes the Universal Commerce Protocol with retail partners Shopify, Etsy, Walmart, and Target |
| March 4, 2026 | OpenAI discontinues Instant Checkout, 156 days after launch, shifting the feature toward its Apps model |
| Mid-2026 | Industry analysts document the “discover in AI, buy on site” pattern as the dominant real-world implementation across the sector |
| July 2026 | Independent AI Checkout Tracker counts roughly 30 retailers/brands with some form of live AI checkout integration |
| September 2, 2026 | Anthropic releases Claude Commerce Agents, an open blueprint for merchant-built shopping and operations agents |
| September 2026 | Google’s UCP-powered checkout begins rolling out to eligible product listings in AI Mode in Search and the Gemini app |
ACP vs UCP: The Protocol War Behind the Chatbots
Underneath the branding, this comparison is really a fight between two plumbing standards. ACP, built by OpenAI and Stripe, ties a merchant’s product feed and checkout flow specifically to Stripe’s payment rails and to ChatGPT’s interface. It was the first mover, live since September 2025, and it remains the technical foundation Copilot Checkout partially borrows from since Microsoft’s stack also accepts ACP-compatible product feeds alongside its own PayPal and Shopify integrations.
UCP takes the opposite design philosophy: build a protocol that isn’t tied to one payment processor or one chatbot. Google’s pitch to retailers is that a single UCP integration could, in theory, serve any AI agent that adopts the standard — not just Gemini — covering the full lifecycle from a shopper’s first product search through post-purchase questions like returns and order tracking. That is a materially bigger scope than ACP’s checkout-focused design, but it also means UCP is newer and less battle-tested. Anthropic, notably, has not thrown its weight behind either protocol. Claude Commerce Agents ships as merchant-owned code with no Anthropic-branded checkout protocol at all, which sidesteps the ACP-vs-UCP fight entirely by refusing to compete for that layer in the first place.
Benchmarks: Adoption Numbers From Three Independent Trackers
Unlike model benchmark comparisons with standardized test suites, agentic commerce doesn’t have a shared scoreboard yet. But three independent sources give a real read on how far actual adoption has gotten as of September 2026, and the picture is more modest than the announcement headlines suggest.
| Source | Metric | Data point |
|---|---|---|
| Taylor Sicard’s AI Checkout Tracker (2026) | Retailers/brands with any form of live AI checkout integration | Roughly 30 rows tracked as of July 2026, spanning full ACP/UCP checkout to lighter “discover-in-chat, buy-on-site” integrations |
| Digital Applied commerce analysis | Direction of the trend since Instant Checkout’s shutdown | Industry pattern described as “discover in AI, buy on site” — discovery inside chat, final purchase still redirected to the merchant’s own site or app in most live implementations |
| Instant Checkout lifecycle (OpenAI/Stripe) | Time between launch and shutdown of ChatGPT’s flagship consumer checkout feature | 156 days — September 29, 2025 to March 4, 2026 |
Taken together, these numbers point to a category that is real but early. Roughly 30 tracked retail integrations is a rounding error next to the number of merchants selling through Google Shopping or Amazon, and the fact that the most publicized consumer feature — Instant Checkout — lasted barely five months before being pulled back is the strongest evidence yet that full autonomous checkout inside a chat window is harder to sustain than demos suggest. Analysts increasingly frame 2026 as the year AI shopping assistants got good at discovery and comparison, while actual payment execution mostly stayed where it already was: on the merchant’s own checkout page.
Security and Fraud: The Problem Nobody Has Fully Solved
Every one of these four platforms runs into the same unsolved problem: standard fraud-detection systems were built to distinguish real human shoppers from bots trying to abuse a checkout flow, not to authorize a legitimate AI agent acting on a real customer’s behalf while still blocking a compromised or hijacked agent session. Stripe’s fraud review for ACP-based transactions and the processor-dependent fraud handling in Copilot Checkout both have to make that same judgment call in real time, and none of the four companies has published detailed fraud-loss data for agent-initiated purchases specifically.
This is also the strongest argument in favor of Anthropic’s more conservative design. Because Claude Commerce Agents never touches payment credentials or completes a transaction itself, Anthropic has effectively pushed the entire fraud-liability question back onto the merchant, who already has fraud tooling built around their own checkout. That’s a smaller attack surface for Anthropic, but it also means Claude offers no shared, cross-merchant fraud intelligence the way Stripe’s ACP integration can in principle pool signal across every retailer using the protocol. Retailers evaluating any of these four options should treat “how does this platform handle a compromised agent session” as a non-negotiable question before integration, not an afterthought — agent-initiated transaction patterns look different enough from human browsing that off-the-shelf fraud models can misfire in both directions, either blocking legitimate agent purchases or missing fraudulent ones.
Consumer Trust: Why Shoppers Are Still Hesitant to Let AI Buy For Them
Adoption numbers for agentic checkout look modest partly because consumer trust in autonomous AI purchasing hasn’t caught up to the technology. A September 2026 poll from Elon University and The Washington Post found that 27% of Americans turn to AI chatbots for personal, emotional, or social queries — a figure that shows genuine day-to-day reliance on tools like ChatGPT and Gemini, but for a very different kind of task than authorizing a financial transaction. Letting an AI agent recommend a product is a low-stakes ask; letting it enter a payment method and complete a purchase autonomously is a much higher bar, and the data on Instant Checkout’s five-month run suggests that gap hasn’t closed yet.
That trust gap shows up directly in how each platform is designed. ChatGPT, Copilot, and Gemini all still require explicit user confirmation before a purchase completes — none of the four platforms currently allows a fully autonomous purchase with pre-encoded approval and no in-the-moment confirmation, despite that kind of full autonomy being technically possible. Anthropic’s decision to avoid consumer-facing checkout entirely with Claude Commerce Agents can be read as a direct response to that trust gap: rather than asking shoppers to trust an AI company with their payment details, Anthropic is asking merchants to build the trust relationship themselves, using Claude only as the reasoning layer behind an experience the shopper already recognizes as the retailer’s own.
Pricing: What It Costs Retailers and Shoppers
None of the four platforms charge consumers directly to use their shopping features — the cost sits on the merchant and developer side, and it varies a lot depending on which protocol and payment stack a retailer chooses to support.
| Platform | Consumer cost | Merchant integration cost | Payment processing fees |
|---|---|---|---|
| ChatGPT / ACP | Free (feature currently discontinued for consumers) | Engineering cost of building a structured product feed + ACP checkout API | Standard Stripe processing rates |
| Copilot Checkout | Free to shoppers on Copilot.com | Shopify app install (low) or direct PayPal/Stripe integration (moderate engineering cost) | Standard PayPal/Shopify/Stripe processing rates |
| Gemini / UCP | Free within Search’s AI Mode and the Gemini app | UCP feed implementation; higher effort for the broader discovery-through-post-purchase scope | Retailer’s existing Google Pay-linked processing setup |
| Claude Commerce Agents | Not applicable — merchant-facing, not a consumer app | Highest upfront engineering cost: build the agent yourself from Anthropic’s reference code | Retailer’s existing payment stack; Claude API usage billed separately per Anthropic’s standard token pricing |
The practical takeaway for a merchant deciding where to spend integration time: Copilot Checkout via Shopify is the lowest-effort entry point right now if you already run a Shopify storefront. UCP asks for more upfront feed work but plugs into Google’s existing Search traffic, which is where most shopping research already starts. Claude Commerce Agents demands the most engineering investment of the four because there’s no packaged consumer surface to plug into at all — you’re building the whole shopper-facing or merchant-facing agent yourself, which is exactly the trade-off Anthropic is betting large retailers will accept in exchange for full control over the checkout experience.
The AI Models Actually Running Behind Each Shopping Agent
It’s easy to treat “ChatGPT,” “Copilot,” “Gemini,” and “Claude” as interchangeable brand names for a shopping feature, but the underlying models doing the reasoning differ in ways that matter for how well each agent actually understands a shopping request. ChatGPT’s shopping and discovery features run on OpenAI’s current model lineup, the same family of models covered in our flagship model comparison, which handles the natural-language product comparisons shoppers actually type into ChatGPT before ever reaching a checkout step. Copilot draws on a mix of OpenAI’s models and Microsoft’s own in-house model efforts, giving Microsoft more flexibility to route shopping queries to whichever model is cheapest or fastest for a given task rather than depending entirely on a single external vendor.
Gemini’s shopping and UCP-powered checkout features are built on Google’s own model family, which also powers Google Search’s AI Mode more broadly — meaning the same model doing your general search summaries is the one evaluating product listings for the checkout feature. Claude Commerce Agents, meanwhile, is designed to run on whichever current Claude model a retailer chooses to deploy, from smaller, cheaper models for high-volume product-matching tasks up to the larger models in Anthropic’s lineup for more complex merchant-operations reasoning like pricing strategy or inventory forecasting. Because Claude Commerce Agents is a blueprint rather than a fixed product, retailers get to make that model-selection trade-off themselves, balancing cost against reasoning quality in a way that ChatGPT, Copilot, and Gemini shoppers can’t, since those three platforms make the model choice for you behind the scenes.
Real-World Examples: Who’s Actually Building on Each Platform
- Shopify merchants running Copilot Brand Agents. Shopify-based retailers can deploy a Microsoft-powered assistant directly on their own storefront, trained on their catalog, to help shoppers find products, finish checkout, and get add-on recommendations without leaving the merchant’s site.
- Walmart, Target, and Etsy backing UCP. These three retailers are named backers of Google’s Universal Commerce Protocol, positioning their catalogs to surface — and eventually checkout — directly inside Google’s AI Mode in Search.
- Travel and ticketing companies piloting Claude Commerce Agents. Anthropic’s blueprint ships with runnable vertical examples specifically for travel and ticketing merchants, not just generic retail, reflecting demand from booking-heavy businesses for conversational search tools ahead of the holiday season.
- Merchants who briefly integrated with ChatGPT’s ACP feeds. Retailers who built structured product feeds for Instant Checkout during its five-month run now have to decide whether to maintain that ACP integration for a possible relaunch inside OpenAI’s Apps model, or redirect that engineering effort toward UCP or Copilot instead.
- Enterprises deploying Claude Commerce Agents via cloud marketplaces. Because the blueprint deploys through Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI in addition to the direct Claude API, companies already standardized on one of those clouds can stand up a merchant-operations agent without a separate Anthropic contract.
Best Use Cases: Which Shopping Agent Fits Which Need
Because none of these four platforms is a complete, mature product yet, the right choice depends heavily on whether you’re a shopper, a small merchant, or an enterprise retail team.
- Casual shoppers comparing products before buying elsewhere: ChatGPT and Gemini both remain strong for discovery and comparison even without a live in-chat buy button — use them to narrow options, then complete the purchase on the retailer’s site.
- Small Shopify merchants wanting the fastest integration: Copilot Checkout’s Brand Agents are the lowest-friction option since they build directly on top of an existing Shopify app install rather than a custom protocol implementation.
- Large retailers with existing Google Ads/Shopping spend: UCP is the natural extension of infrastructure these retailers likely already have, since it plugs into Search traffic they’re already paying to capture.
- Enterprises needing full control over checkout, compliance, and branding: Claude Commerce Agents’ merchant-owned model is the better fit, since Anthropic never touches payment or customer data — everything stays on the retailer’s own systems.
- Travel, ticketing, and appointment-based businesses: Claude Commerce Agents’ purpose-built travel and ticketing examples give these merchants a head start that generic e-commerce-first tools like ACP or UCP don’t specifically address.
- Developers who want to avoid picking a protocol at all: Building on Claude’s API-first blueprint sidesteps the ACP-vs-UCP question entirely, at the cost of more custom engineering.
Migration Guide: Preparing a Storefront for Agentic Checkout
If you run an e-commerce operation and want to be ready for whichever protocol wins out — or want to hedge by supporting more than one — the technical prep work looks similar across ACP, UCP, and Copilot’s integrations. Here’s the sequence retailers are actually following in mid-to-late 2026.
- Audit your current product catalog and identify gaps between what’s in your database and what a structured, machine-readable feed requires (pricing, inventory, variant data, shipping rules).
- Build or export a daily machine-readable product feed rather than relying on HTML page scraping — both ACP and UCP require this as a baseline.
- Expose checkout as a callable API endpoint, not just a web form, so an agent can submit an order programmatically.
- Decide which protocol(s) to support first based on where your traffic actually comes from: Google-heavy traffic favors UCP, ChatGPT referral traffic favors ACP, existing Shopify infrastructure favors Copilot’s Brand Agents.
- Tune your fraud detection system specifically for agent-initiated transaction patterns, which differ from human browsing behavior and can trigger false positives on standard fraud models.
- If you’re Shopify-based, install the relevant Brand Agent or Shopify commerce app rather than building a custom integration from scratch.
- If you’re evaluating Claude Commerce Agents, start from Anthropic’s reference implementation for your vertical (retail, travel, or ticketing) rather than building an agent from a blank slate.
- Set up analytics that can specifically attribute sales originating from AI agents versus organic search or direct traffic, since standard analytics tools often can’t distinguish agent-driven purchases yet.
- Run a limited pilot with one protocol before expanding, given how quickly this category has already seen one major feature (Instant Checkout) launch and shut down within a single year.
- Maintain a fallback to your standard web checkout for every agent-initiated flow, since none of the four platforms guarantee 100% uptime or long-term feature stability yet.
Pros and Cons of Each Platform
ChatGPT / Agentic Commerce Protocol
- Pro: First-mover protocol with the biggest existing chatbot user base to draw discovery traffic from.
- Pro: Stripe’s mature payment infrastructure and fraud tooling underpin the protocol.
- Con: The flagship consumer feature, Instant Checkout, has already been discontinued once.
- Con: Merchants who integrated for Instant Checkout now face uncertainty about whether that investment carries over to OpenAI’s Apps model.
Microsoft Copilot Checkout
- Pro: Lowest integration friction for the large base of existing Shopify merchants.
- Pro: Multi-processor flexibility (PayPal, Shopify, Stripe) avoids single-vendor lock-in.
- Con: Still in limited US testing on Copilot.com, not yet broadly available across Bing, MSN, and Edge.
- Con: Smaller consumer chatbot user base than ChatGPT to drive organic shopping traffic.
Google Gemini / UCP
- Pro: Ties into Google Search’s existing product-research traffic rather than requiring a separate destination.
- Pro: Backed by major retailers (Shopify, Etsy, Walmart, Target) at launch, giving it early scale.
- Con: Broader scope (discovery through post-purchase) means a more complex integration than a checkout-only protocol.
- Con: US-only rollout for the checkout feature as of September 2026.
Claude Commerce Agents
- Pro: Merchant retains full ownership of checkout, payment, and customer data — Anthropic never touches the transaction.
- Pro: Deployable across Claude API, Bedrock, Microsoft Foundry, and Vertex AI, fitting into whatever cloud a retailer already uses.
- Con: No packaged consumer-facing shopping app — retailers must build the experience themselves from a blueprint.
- Con: No named retail partners confirmed yet at launch, unlike UCP’s Shopify/Etsy/Walmart/Target backing.
The Verdict: Which AI Shopping Agent Actually Wins in 2026
There isn’t a single winner yet, and the data explains why. ChatGPT proved agentic commerce could work technically, then walked it back within 156 days — a clear signal that consumer trust and merchant economics for full in-chat checkout weren’t there yet. Microsoft’s retailer-first approach through Copilot Checkout and Shopify Brand Agents looks like the more durable near-term strategy precisely because it doesn’t require Microsoft to own the entire shopping experience. Google’s UCP has the broadest retail backing at launch and the strongest structural advantage — it rides on Search traffic that already exists — but it’s the newest of the four and still unproven at scale. Anthropic’s Claude Commerce Agents is the most conservative bet of the group: no consumer checkout button to shut down, no protocol war to lose, just tooling for merchants who want to build their own agents on their own terms.
If you’re a shopper, the honest answer in September 2026 is that no chatbot yet reliably replaces a normal checkout page — ChatGPT, Copilot, and Gemini are best used for product discovery and comparison, with the actual purchase often finishing on the merchant’s own site regardless of which assistant you started with. If you’re a retailer choosing where to invest engineering time, the roughly 30-integration adoption tracker and Instant Checkout’s five-month lifespan both argue for starting with the lowest-commitment option that fits your existing stack — Shopify merchants toward Copilot, Google-Ads-heavy retailers toward UCP — rather than betting your entire commerce roadmap on any one chatbot’s checkout button surviving the next twelve months unchanged.
How This Compares to Other AI Assistant Categories
Agentic commerce is just one front in a much broader push to turn chatbots into agents that complete tasks rather than just answer questions. The same four companies are running comparable experiments in AI-powered web browsing — see our comparison of Comet vs Gemini Agent vs ChatGPT Atlas for how that plays out in general-purpose browsing agents rather than shopping specifically. And if pricing and free-tier limits factor into which assistant you rely on day to day for shopping research, our breakdown of ChatGPT vs Claude vs Gemini free tier message caps is a useful companion piece, since heavier product-comparison sessions can burn through free-tier limits faster than typical chat use.
It’s also worth separating consumer chatbots from Microsoft’s other Copilot products entirely — our Copilot vs Gemini vs Perplexity comparison covers the assistant landscape more broadly, including search-first tools like Perplexity that are adjacent to but distinct from the checkout-focused agents covered here.
What Retailers Should Watch Next
Three signals will determine whether agentic commerce becomes a durable retail channel or a 2026 footnote. First, whether OpenAI’s Apps model actually reintroduces a checkout-capable feature, since that would validate ACP’s second attempt after Instant Checkout’s shutdown. Second, whether UCP’s named retail partners — Shopify, Etsy, Walmart, and Target — report meaningful transaction volume once the AI Mode checkout feature is broadly live, rather than just discovery traffic. Third, whether any large retailer publicly adopts Claude Commerce Agents’ blueprint at scale, since Anthropic’s approach has zero confirmed retail partners as of its September 2026 launch and needs a visible flagship deployment to prove the merchant-owned model works in practice.
Regulatory attention is also building around agentic transactions generally, given how much scrutiny AI companies are already under for unrelated compliance issues. Retailers building agent-facing checkout flows should expect fraud review and payment compliance requirements to tighten further as transaction volume grows, since agent-initiated purchases don’t fit neatly into fraud models built around human browsing behavior. For a snapshot of how Shopify itself frames the broader shift, see Shopify’s own overview of agentic commerce.
Frequently Asked Questions
Is ChatGPT’s Instant Checkout still available?
No. OpenAI discontinued Instant Checkout on March 4, 2026, 156 days after its September 29, 2025 launch, and has said the functionality is shifting toward its Apps model instead of a standalone in-chat buy button.
What is the Agentic Commerce Protocol (ACP)?
ACP is a standard built jointly by OpenAI and Stripe that connects merchant product feeds and Stripe-based payment processing to AI shopping agents, originally designed to power ChatGPT’s Instant Checkout feature.
What is the Universal Commerce Protocol (UCP)?
UCP is Google’s competing protocol, designed to be portable across any AI agent rather than tied to one chatbot or one payment processor. It covers discovery, checkout, and post-purchase support, and launch backers include Shopify, Etsy, Walmart, and Target.
Does Claude let you buy products directly in the chat?
No. Claude Commerce Agents, released by Anthropic on September 2, 2026, is a blueprint for retailers to build their own shopping and merchant-operations agents on Claude. It explicitly does not place orders or process payments itself — that stays entirely on the merchant’s own systems.
Is Microsoft Copilot Checkout available outside the US?
As of September 2026, Copilot Checkout is in testing on Copilot.com in the US only, with Microsoft stating plans to expand into Bing, MSN, and Edge as the rollout progresses.
Which AI shopping assistant should a small e-commerce store integrate with first?
For merchants already running Shopify, Microsoft’s Copilot Brand Agents currently offer the lowest-friction integration path. Retailers with significant Google Ads or Shopping traffic may see a faster return by prioritizing UCP instead.
Why did OpenAI shut down Instant Checkout after only five months?
OpenAI has not published a detailed technical postmortem, but industry analysts point to the broader “discover in AI, buy on site” pattern — most live agentic checkout implementations across the industry still redirect final payment to the merchant’s own site rather than completing it end-to-end inside the chat, suggesting standalone in-chat checkout wasn’t converting the way OpenAI needed it to.
How many retailers currently support some form of AI checkout?
Independent tracking from the 2026 AI Checkout Tracker counted roughly 30 retailers or brands with some form of live AI checkout integration as of July 2026, ranging from full ACP/UCP-based checkout to lighter discover-in-chat, buy-on-site flows.
Can an AI shopping agent complete a purchase without any human confirmation?
Not on any of the four platforms covered here as of September 2026. ChatGPT, Copilot, and Gemini all require explicit user confirmation before completing a purchase, and Claude Commerce Agents doesn’t process payments at all — full autonomous purchasing with pre-encoded approval remains technically possible but isn’t how any mainstream implementation currently operates.
Should a retailer support both ACP and UCP?
Many commerce teams are hedging by preparing infrastructure — structured product feeds and API-based checkout — that works for either protocol, then deciding which one to prioritize based on where their traffic actually originates. Shopify has taken this hedging approach at the platform level, supporting integrations tied to both Microsoft’s and Google’s ecosystems rather than picking a single side.
Related Coverage
- GPT-6 Astra vs Claude Opus 5 vs Gemini 3.8 Flash: 107x Gap [2026]
- GPT-5.6 Sol vs Fable 5.1 vs Gemini 3.8 Flash: 30-Point Gap [2026]
- ChatGPT, Claude and Gemini Down: 3 AI Giants Crash [2026]
- AI Image Generator API Costs: 12 Steps, $150 Gap [2026]
- ChatGPT Down Again: 74,000 Reports, Codex Hit Too [2026]
