Google Universal Cart vs ChatGPT Checkout, Scored Six Ways
Google's Universal Cart and ChatGPT's Instant Checkout compared on reach, feed lift, take rates, and customer data, with a decision tree by merchant type.
On this page
Two checkout buttons are competing for the same engineering quarter. Google introduced Universal Cart at I/O 2026 on May 19: one intelligent cart that follows shoppers across Search, the Gemini app, YouTube, and Gmail, with agentic checkout underneath. OpenAI's Instant Checkout, live since late 2025 and extended to Shopify merchants in early 2026, completes purchases without the shopper ever leaving ChatGPT.
Most merchants cannot staff both integrations at once. So the real question is sequencing. We scored both surfaces on six criteria. Same rubric, same order, no favorites.
Why this decision exists at all
The traffic shift is measurable. Adobe tracked AI-source traffic to U.S. retail sites up 393% year over year in the first quarter of 2026, and found 39% of consumers have already used AI for online shopping. Discovery is moving into conversations. Checkout is following it. The merchants who sequence integration well will buy distribution cheaply; the ones who sequence badly will pay take rates on orders they would have gotten anyway.
Criterion 1: Reach
Google. The ceiling is enormous. Google says AI Overviews reach two billion people monthly, and it reported more than 750 million monthly active users for the Gemini app in February 2026. Universal Cart spans four surfaces: Search, Gemini, YouTube, Gmail. The catch is timing. The U.S. rollout starts in summer 2026, so live transactional reach today rounds to zero.
OpenAI. ChatGPT passed 800 million weekly active users, per TechCrunch coverage of DevDay in October 2025. Instant Checkout is U.S. only and covers a subset of that base, but it has processed real orders since launch, starting with Etsy sellers.
Edge: split. Google wins on ceiling. OpenAI wins on shipped, transacting reach.
Criterion 2: Feed requirements
Google. The Universal Commerce Protocol, detailed on the Google Developers Blog in January 2026, builds on Merchant Center and launched with Shopify, Etsy, Wayfair, Target, and Walmart as collaborators. If you already run Shopping campaigns, most of the data work is done. What remains: feed hygiene, UCP checkout capability endpoints, order webhooks. Payments ride on the AP2 protocol Google Cloud announced with Mastercard, PayPal, American Express, and dozens of other partners.
OpenAI. The Agentic Commerce Protocol, co-developed with Stripe, needs a product feed in ACP's spec, checkout API endpoints, and a Shared Payment Token flow. On Shopify or Etsy, the platform does this for you. On a custom stack, it is a net-new build with its own maintenance burden.
Edge: Google, for anyone already in Merchant Center. That describes most merchants who advertise.
Criterion 3: Take-rate economics
OpenAI. The launch language was "a small fee" on completed purchases, refunded on returns. Reporting settled the number: PYMNTS, citing The Information, put it at 4% for Shopify merchants, stacked on top of Shopify's own fees. Run the math. At a 10% net margin, four points of take rate is 40% of the profit on that order. At 55% margins, it is a rounding error. Category economics decide this criterion, merchant by merchant.
Google. No published per-transaction fee for Universal Cart checkout as of this writing. Google monetizes the surface the way it always has: ads and Merchant Center participation. That could change once the cart earns habit. Today, the marginal take is zero.
Edge: Google, decisively, until pricing appears.
Criterion 4: Who owns the customer record
OpenAI. You stay merchant of record. ACP passes the order, shipping details, and a scoped payment token to your backend; you fulfill, you handle returns and support. But the conversation belongs to ChatGPT. Product questions, preferences, repeat intent: all of it accrues to OpenAI's memory of the shopper. Your record starts at the order and ends at delivery.
Google. Two modes. Native checkout through Google Pay keeps you merchant of record under UCP. Or the handoff: shoppers can transfer cart items to the merchant's site and complete the purchase there. The handoff preserves everything: session, consent flow, upsell, email capture. It is the quiet win in Google's design, and merchants should default to it wherever conversion cost allows.
Edge: Google, narrowly, because the handoff mode exists. Either way, feed both order streams into your first-party data program so the second purchase happens on owned channels.
Criterion 5: Integration lift today
Google. Partial. The spec is public and the partner list is long, but Universal Cart itself ships to U.S. users this summer. You can prepare; you cannot finish.
OpenAI. Documented and live. The ACP spec is on GitHub, Stripe tooling exists, and Shopify merchants toggle it on. A prepared team can transact within weeks.
Edge: OpenAI. It is real today.
Criterion 6: Execution risk
OpenAI. CNBC reported in March 2026 that OpenAI's first crack at shopping stumbled and the company was retooling for a second wave. Early merchant adoption ran thinner than the launch coverage implied. The 4% fee is also young enough to move in either direction.
Google. Universal Cart is weeks old, unproven with real buyers, and Google has retired more commerce products than most merchants can name. Betting a roadmap on a Google shopping launch has burned people before.
Edge: split. Both are moving targets. Budget integration hours accordingly.
The verdict matrix
| Criterion | Universal Cart (Google) | Instant Checkout (OpenAI) | Edge |
|---|---|---|---|
| Reach | ~2B monthly AI Overviews users; four surfaces; US launch summer 2026 | 800M weekly ChatGPT users; live and transacting since 2025 | Split |
| Feed requirements | Merchant Center plus UCP endpoints | ACP feed spec, checkout API, Stripe token | |
| Take rate | None published | 4% reported for Shopify merchants | |
| Customer record | MoR retained; site-handoff mode available | MoR retained; discovery relationship stays in ChatGPT | |
| Integration lift today | Preparable, unfinishable until rollout | Documented, shippable in weeks | OpenAI |
| Execution risk | Brand new, Google commerce history mixed | March 2026 scale-back, fee uncertainty | Split |
Score it flat and Google leads three to one with two splits. Score it by what you can ship this month and OpenAI leads. Both readings are correct. That is why the decision tree matters more than the tally.
The decision tree, by merchant type
You sell on Shopify or Etsy. Enable both. Your platform absorbs the protocol work, so the marginal cost is configuration and QA. Gate low-margin SKUs out of ChatGPT exposure and let everything else run.
DTC brand, net margin under 15%, AOV under $100. Google first. The 4% take rate eats too much of a thin order, and your Merchant Center feed already does most of the UCP prep. Spend the waiting period tightening feed quality and your ecommerce ad targeting, because the same product data powers both.
High-AOV, considered purchases: furniture, jewelry, equipment. Google first, handoff mode preferred. A $2,000 order should end on your site, with your financing options, your configurator, and your email capture. Four percent of $2,000 is also $80 of pure take. Model the difference in a ROAS calculator before exposing these SKUs anywhere.
High-margin niche products with strong ChatGPT discovery. OpenAI first. If shoppers already ask ChatGPT questions your product answers, 4% is cheap customer acquisition compared to paid CAC in most categories. Pair it with AI audience segmentation on your ad accounts so the demand you see in chat informs the audiences you buy elsewhere.
Subscription and replenishment brands. Sequence by record ownership. Take Google's handoff to own the relationship from order one; treat any ChatGPT purchase as acquisition and move the refill onto owned channels fast. Enriching those first orders into full customer profiles is where the LTV math gets won.
B2B, custom pricing, configured products. Neither, yet. Publish clean structured feeds, keep your product data machine-readable, and revisit in two quarters. Agents cannot quote what they cannot parse.
The referee's summary: this is a sequencing call, and sequencing calls expire. Google's zero take rate is an introductory condition. OpenAI's 4% is a first draft. Re-run this scorecard the day either number moves.