Unified checkout for AI agents: found is not bought

Adobe has AI-referred visits to US retail sites converting 60% higher than other traffic in July 2026. That premium exists only where the shopper can complete the purchase where they are. A brand that gets recommended by an AI agent but cannot close inside that surface hands the advantage back at the handoff.

Unified checkout for AI agents is one integration that lets ChatGPT, Gemini, Copilot or Muse complete a purchase against your store under whichever protocol the surface speaks, with you as merchant of record, instead of a separate build per surface. This post covers why it matters; the complete guide covers the build.

Last updated: 13 October 2026.

The gap between being found and being bought

Most of the agentic commerce conversation is about discovery. Necessary, not sufficient: if the shopper still has to leave the conversation, find your site, re-enter their details and complete a conventional checkout, you have reintroduced every point of friction agentic commerce is supposed to remove.

The size of the prize is now measured. Adobe Analytics reports that AI-referred visits to US retail sites converted 60% higher than non-AI visits in July 2026, with 53% more revenue per visit, the eleventh straight month AI traffic out-converted other traffic (Digital Commerce 360, 19 August 2026). That is a recent turn: the same traffic was 23% less likely to convert in July 2025 and 49% less likely in January 2025 (Adobe, August 2025), and 42% more likely by March 2026 (Adobe Digital Insights, April 2026).

Two anchors keep the scale honest. Across 94 ecommerce stores in 2025, ChatGPT referrals converted at 1.81% against 1.39% for non-branded organic search, a 31% lift on a 14.3% lower average order value (180 Marketing, 23 February 2026). Across 973 ecommerce sites, organic LLM traffic converted "above paid social but below all other traditional channels" in its first year (Kaiser and Schulze, Marketing Science, 2026). The premium is tens of percent, not multiples, and it is measured on shoppers who reached a retailer's site and bought. A checkout that breaks the conversational flow is exactly where it erodes.

Editor's note, 13 October 2026: an earlier version opened with a conversion pair we attributed to Adobe. It is Seer Interactive's single-client lead-gen measurement and has been removed; the correction is in AI commerce conversion rate: what the data actually says.

What "unified checkout" actually means

Rather than a separate checkout integration for every AI surface, one for ChatGPT via ACP, one for Google via UCP, one for whatever comes next, unified checkout is a single integration layer that speaks every relevant protocol, so the transaction completes wherever the shopper already is.

Practically, that requires:

• Protocol coverage across ACP (OpenAI and Stripe), UCP (Google) and whatever emerges next, without re-architecting for each new standard.
• Payment authorisation compatible with AP2, the Agent Payments Protocol Google announced in September 2025 for how agents authorise and execute payments (Google Cloud, 2025).
• Merchant-of-record continuity: the brand keeps the customer relationship, the transaction data and the margin, rather than the AI platform intermediating the sale as its own transaction. The credentials that reach your gateway on each rail are in how AI agent payments work.

Why building this per-surface is the wrong instinct

The natural first move for a technical team is to integrate ACP for ChatGPT, ship it and move on. The problem is how fast the wrappers change. Stripe and OpenAI released ACP with Instant Checkout in ChatGPT on 29 September 2025 (Stripe, 2025). Google announced UCP on 11 January 2026 with its own coalition of partners (Google, 2026). Within six months OpenAI had reworked ChatGPT shopping so merchants could use their own checkouts (CNBC, 24 March 2026). A brand that hard-codes its checkout logic around one protocol redoes meaningful work each time the surface mix shifts. Betting on a single protocol staying dominant is a real risk, not a hypothetical one.

The durable work sits underneath all three: a catalog with variant-level price and availability, a guest path, no captcha at the pay step, fraud rules that do not auto-decline a fresh single-use card. What each personal agent needs from a store is compared in AI shopping agents.

Where this sits in the stack

Checkout execution is Layer 05 in the agentic commerce stack; the 7-layer map has the full picture. It sits downstream of discovery (Layer 06) and protocol connectivity (Layer 02): a brand can be perfectly discoverable and still lose the sale if Layer 05 is not built to close the loop.

It also sits upstream of measurement. An agent-placed order often arrives with no referrer and a payment credential your analytics has never seen, so GA4 files it as Direct. For MyMuse, GA4 reported ₹11.5K/mo of AI-referred revenue against ₹81.2K/mo measured over a 12-month attribution, 7× hidden (Tru Commerce case study, 2026); the mechanics are in dark agentic commerce traffic.

Where Tru Commerce fits

Agentic checkout is the product we run this on: checkout completes inside the conversation rather than bouncing the shopper to a cart page, and the merchant stays merchant of record. We have run it on Gray Porter and a group of US DTC brands: we made the product pages and checkout readable and finishable by an agent, then placed a real Muse order on the store. On measurement, MyMuse is the reference: GA4 showed ₹11.5K/mo of AI revenue, the real figure was ₹81.2K/mo, 7× hidden, and the Year 1 net surfaced was +₹29.34L. If discovery is already working for you and checkout is the missing piece, book a demo.