---
title: "The Checkout War, Seven Months On"
date: 2026-08-13
description: "In January I called checkout the new battleground. The search data since says the hype crested that month and something more useful took its place."
tags: ["Artificial Intelligence"]
canonical: https://www.serp-secrets.com/blog/checkout-war-seven-months-on/
---
# The Checkout War, Seven Months On

## What did I claim in January?

That Google and OpenAI had formalized competing rails for agent-driven purchase, and that merchants would have to decide which to build for. Seven months ago I called [checkout the new battleground](https://www.serp-secrets.com/blog/checkout-new-battleground/), setting Google's Universal Commerce Protocol against OpenAI's Agentic Commerce Protocol, argued that product catalogs built for keyword search would not survive conversational reasoning, and closed by saying the protocol wars were not about which company won but which standard became the default rails.

Seven months of search data are now available on top of that. Some of it supports the argument. One part of it contradicts the framing, and that part is the interesting one.

## What does the demand curve actually say?

That I published in the peak month. Searches for **agentic commerce** in the US hit **8,100 in January 2026**, the highest point in the entire series. The article went out on 13 January.

Here is what happened next, month by month: 6,600 in February, 6,600 in March, 6,600 in April, 5,400 in May, **4,400 in June**. That is a **33% decline over the quarter**, against a twelve-month average of 5,400.

Now the other half of the same number. Year over year, **agentic commerce is up 132%**. In July and August 2025 it was doing 1,900 a month. The floor it has settled onto is more than double where it started.

Both readings are true and the combination is the finding. This was a hype cycle with a real base underneath it. The spike deflated. The base did not.

I did not predict the spike would deflate. I described a landscape as though it would keep expanding at the rate it was expanding when I looked at it, which is the standard error in writing about a category during its loudest month.

## Did the protocol war resolve?

The audience stopped watching it, which is not the same thing and is more useful to know.

**agentic commerce protocol** peaked at **2,900 searches in October 2025** and now runs at 390 a month, **down 46% in a quarter**. Ranking difficulty on it is 63, which tells you plenty of people wrote the explainer while the volume was there.

The vendor-specific pattern is sharper. **openai agentic commerce** is **down 57% over the quarter**. **google agentic commerce** is **up 267% year over year**. **paypal agentic commerce** is down 36%, **visa agentic commerce** down 19%.

The clearest signal is what happened to the announcement queries. Three of them appear in the data with volumes that look substantial until you read the monthly series:

* **google paypal agentic commerce partnership**: 5,400 searches in September 2025, then 10 a month, every month, ever since
* **mastercard paypal agentic commerce**: 4,400 in October 2025, then 10 a month
* **wix paypal agentic commerce**: 1,600 in October 2025, then 10 a month

Each partnership announcement produced one enormous month of attention that then went to essentially zero and stayed there. Nobody came back to check how the integration was going. The announcement was the product, as far as public interest was concerned.

That is the piece my January framing got wrong. I treated the standards contest as the main event that merchants needed to track. The public treated it as news, consumed it, and moved on within thirty days.

## Where did the attention actually go?

To the platforms merchants already run on, and this is where the growth is now.

* **shopify agentic commerce**: 320 a month, **+767% year over year**
* **stripe agentic commerce**: 480 a month, **+336%**
* **adyen agentic commerce**: 70 a month, **+133% over the quarter**
* **agentic commerce platform**: 90 a month, **+27% over the quarter**

The pattern is unmistakable once you see it beside the protocol decline. Interest moved from "which standard will win" to "what does my commerce stack support". That is the transition from a strategy question to an implementation question, and it happens when a technology stops being interesting and starts being scheduled.

A merchant searching **shopify agentic commerce** is not evaluating the future of retail. They are trying to find out whether the thing their platform announced is switched on, what it costs, and who has to configure it.

For most of them the answer is that it is already on.

**Shopify** switched Agentic Storefronts on for all US merchants on **24 March 2026**, having previewed them in the Winter '26 Edition the previous December. Eligible products syndicate to ChatGPT, Microsoft Copilot, Google AI Mode, Gemini and the Shop app with no merchant configuration at all. The **Spring '26 Edition on 17 June** shipped more than 150 updates framed almost entirely around agentic discovery. In January the open question was which protocol Shopify would build for. It built one: the **Universal Commerce Protocol**, co-developed with Google and backed by Amazon, American Express, Etsy, Mastercard, Meta, Microsoft, Salesforce, Stripe, Target, Visa and Walmart.

**Stripe** shipped the primitive rather than the storefront. **Shared Payment Tokens** let an agent pay on a buyer's behalf without ever handling card credentials, scoped to a single business, capped by time or amount, revocable, and observable through webhooks. The detail that matters to a merchant is that there is nothing to enable: any seller already processing with Stripe supports them automatically. Coverage has since extended to Visa and Mastercard network tokens and to Affirm and Klarna. Etsy and URBN were among the first at volume.

**Adyen** arrived last and drew the most useful conclusion. **Adyen Agentic**, announced **16 June 2026**, is three API layers, a feed, a cart and a payments layer, and its entire selling point is that it speaks Meta's AI checkout, UCP, Google's Agent Payments Protocol and OpenAI's ACP simultaneously. It is in limited US enterprise availability with no published pricing.

Read those three together and the protocol war has an ending nobody framed as a possible outcome in January, including me. It did not resolve by one standard winning. It resolved by the payments layer declining to choose. Stripe co-authored ACP with OpenAI and also backs UCP. Shopify built UCP with Google while its merchants sell into ChatGPT over ACP. Adyen productized the refusal and called it a universal translator.

Merchants were never going to have to pick, because the infrastructure absorbed the fragmentation. That is what infrastructure is for, and it is the part I got wrong. Adyen's Karan Katyal puts current maturity at **0.5 out of 5** and expects roughly 1.5 within a year, naming catalog normalization, fraud liability and protocol fragmentation as the constraints. A vendor talking down their own launch is usually the most reliable number in a press release.

## Who is asking now?

A later and less specialist audience, which is the healthiest number in the whole dataset.

**what is agentic commerce** does **1,000 searches a month, up 306% year over year**. **agentic commerce meaning** does 320, **up 420%**. Definitional demand is growing fast while the head term declines.

It is the same composition change visible in [GEO's own keyword curve](https://www.serp-secrets.com/blog/geo-grew-up/), and it means the same thing here. The early cohort learned the answer and stopped searching. A much larger mainstream cohort is arriving now and starting from zero.

For anyone publishing on this, that has a direct consequence. The audience for a deep protocol comparison peaked ten months ago and has left. The audience for a clear explanation of what agentic commerce means for a mid-sized retailer is growing at 300% a year and is currently served by content written for the previous audience.

## What should a merchant do differently?

Three things change, and one of my January recommendations should be retired.

**Stop treating protocol selection as a decision you make.** I framed this as a choice between UCP and ACP. In practice the choice is made by whatever your commerce platform and payment processor support, and the growth in platform-specific queries shows merchants have worked that out. The useful question is not which standard to back, it is which of them your existing stack has already implemented.

**Keep the catalog work.** This part of the January article holds up and is now more urgent, not less. Product data built for [keyword matching](https://www.serp-secrets.com/blog/keywords-in-seo-why-they-still-matter/) still does not survive a query that carries occasion, constraint, budget and history at once. Granular attributes, current inventory, enriched reviews and content that explains why this product for this person: all of that is retrieval material, and it is the same discipline as [chunk-level optimization](https://www.serp-secrets.com/blog/pages-retrieved-chunks-cited/) applied to a feed instead of an article.

**Measure presence in agent answers, not just protocol readiness.** Being technically capable of completing a transaction inside an agent is worth nothing if the agent never surfaces your product. That is a visibility problem before it is a checkout problem, and it needs the kind of repeated, variance-reported measurement that [most AI visibility scores skip](https://www.serp-secrets.com/blog/measuring-ai-visibility/).

**Do not budget against the announcement cycle.** The three partnership queries above are a warning about how this category communicates. Announcements are loud, frequent, and followed by silence. If your roadmap moves every time a payment network issues a press release, you will spend the year re-planning.

## Where does this go?

Toward something duller than the January framing implied, and duller is the correct destination.

The pre-Sputnik line I quoted in January was right about the moment and wrong about what would follow. There was no launch event. There was a slow shift from people asking what this is to people asking whether their platform supports it, visible in the data as a declining head term and a set of exploding vendor queries.

That is what adoption looks like from the outside. It does not arrive as a winner being declared. It arrives as a configuration screen.

The standards contest I described is still unresolved and it matters less than I said it did. The catalog work I described is still unfinished and it matters more.

Seven months on, the battleground moved. It is in your product feed.