AI agent orders reached 5.3×: what Forter’s commerce study shows

Asking AI to find a product and then buying it yourself is one shopping pattern. Another is emerging: an AI agent advances the purchase process and places the order.

Forter, which provides services including transaction fraud prevention, reports that AI-agent orders observed on its network reached 5.3 times their early-August level. It also reports that agents operating merchants’ normal checkout pages accounted for 65% of detected agent orders. The Forter article was published on October 1, 2026 and updated on October 8.

This article explains the order routes behind those numbers and what a merchant should distinguish in its own records. The 5.3× figure is not revenue growth, and 65% is not AI’s share of all online purchases.

AI-commerce orders reached 5.3× their early-August level

Conceptual index comparison based on Forter’s report, with early-August order levels set to 1. The 5.3× change is not an absolute order count or revenue figure.

The figure describes a change in AI-agent orders detected by Forter. “AI commerce” can include AI involvement in discovery and shopping decisions, but this dataset concerns orders actually placed through agents.

Forter counted orders placed through AI agents

Someone asking AI for recommendations is not necessarily making an agentic order. They may stop after researching or buy the product themselves. The observed outcome here is an order placed through an AI agent.

To understand 5.3×, treat the early-August level as 1 and compare it with 5.3. Do not invent an initial count of 100 or 10,000 orders. The article does not disclose the absolute baseline behind that multiplier.

Order counts also differ from sales value. Basket size and returns can change revenue even when the number of orders rises. This is not a case study establishing that a merchant’s AI optimization increased revenue fivefold.

Distinguish the publication date, update date and study period

We reviewed the October 8 version, covering August 1 through October 7, 2026. The multiplier uses early August as its comparison point.

Date or period	Meaning
October 1, 2026	Original publication
October 8, 2026	Update date of the version reviewed
August 1–October 7, 2026	Data period
Early August	Baseline for the 5.3× comparison

Keeping these dates separate prevents mixing an earlier article’s figures with a revised dataset. This is not presented as an October 11 announcement. If the source changes again, check the accompanying period as well as the headline number.

Browser agents represented 65% of detected AI-agent orders

Top: a dedicated integration. Bottom: an agent operating the normal checkout. Availability depends on the merchant and service.

Forter distinguishes two broad routes: agents using a dedicated commerce integration, and agents operating a merchant’s ordinary checkout interface.

The latter group grew from 1% of detected agent orders in early August to 65% in the period the source calls “last week.” The denominator is detected AI-agent orders, not every ecommerce order placed by people and agents combined.

Orders through a dedicated AI checkout connection

With a dedicated connection, the merchant’s supported order system exchanges information with the AI service. Forter groups OpenAI’s ACP-based checkout and Google’s UCP-based checkout in this category.

As a conceptual example, a shopper could ask an assistant to order one item, with a supported integration passing the product and order details to the store. The available actions and the moment at which approval is required depend on the service and its integration conditions.

For a merchant, the useful question is whether its own contract, region and payment methods are supported. Knowing a protocol’s name is different from being able to accept an order through it.

Orders through the merchant’s normal checkout screens

A browser agent opens the store’s pages and operates the shopping interface, including product selection or checkout fields. It follows a path closer to a person using the site.

A conceptual request might be to find an item meeting certain conditions in a specified store and advance its checkout. Even without a dedicated AI order connection, the store’s existing purchase screens may therefore be relevant.

That does not mean every agent can complete a purchase at every store. Sign-in, identity checks, delivery conditions and merchant policies may prevent completion. Success at a particular checkout requires separate verification; this is not an argument for disabling protection across the site.

The observed order mix changed with Muse

The date an order occurs is distinct from the date a measurement system can reliably identify its type.

Forter says it began detecting Muse orders on September 21. Its updated observations show a changed mix between browser-operated orders and orders using dedicated commerce connections.

A detection date is not a service’s launch date

September 21 marks when Forter could reliably identify Muse orders. It does not establish the first date on which Muse ever placed an order.

The distinction also matters in a merchant’s own reporting. If a new detection feature is enabled in October and starts finding AI orders, that does not prove there were none beforehand.

Record measurement changes alongside the order trend. Otherwise an increase in actual ordering can be confused with improved ability to recognize orders that were already occurring.

Distinguish ChatGPT’s protocol checkout from its browser agent

Orders associated with the same provider can still follow different routes. Forter reports OpenAI’s dedicated checkout route separately from the browser-operating ChatGPT Agent.

A label such as “from ChatGPT” could obscure several situations: a person following an AI recommendation and buying manually, an order arriving through a dedicated integration, or an agent operating the store’s checkout.

Alongside the service name, record who advanced the checkout and what evidence supports the classification. Improving product information and fixing an interrupted checkout are different tasks that require looking in different places.

What Japanese merchants should check in their own order records

Left to right: a human buys after an AI recommendation, an agent uses an integration, and an agent operates checkout. Check which routes your setup identifies.

For a Japanese merchant, the immediate lesson is to understand what its current records can identify. Forter’s observations are concentrated in North America and do not provide a Japanese-market growth rate.

Separate visits from AI answers and orders placed by AI

A person may follow a link in an AI answer, visit the store and purchase manually. Measuring that visit does not establish that AI performed the purchase.

Conversely, an agentic order may not leave a clearly recognizable AI service name in ordinary traffic analytics. Referral data alone should not be assumed to classify every order.

Question	Example evidence to inspect
Did a person arrive from an AI answer?	Referral data or supported AI-traffic measurement
Did an order arrive through a dedicated connection?	Order information recorded by that integration
Did an agent operate the normal checkout?	Supported agent-detection records matched to an order

This is a classification framework, not a promise that every cart or analytics product provides all three. If an item cannot be measured, record “unknown with our current method” instead of zero.

Ask your current cart provider a focused set of questions

Start with the cart provider or fraud-prevention service you already use. The following is an editorial example of a support inquiry. Replace the two descriptive fields with your own details.

Please explain which AI-agent orders we can identify with our current setup.

Store: our shop URL

Services: our contracted cart and fraud-prevention services

1. Can we identify orders arriving through a dedicated AI commerce connection?

2. Does detection also cover agents operating our normal checkout pages?

3. If supported, which admin field should we inspect, and how do we match it to an order?

4. Which regions or payment methods are unsupported, and how are unclassified orders shown?

As one concrete implementation example, Forter’s checkout documentation describes orders flowing from its integration into existing order management, with information identifying their AI origin. That applies to the described integration; it is not a screen every merchant can already access.

Use the provider’s response to agree on the exact screen and field your team will inspect. If unsupported, clarify the order types and coverage before assessing the cost of a new service. There is no need to turn off fraud protection merely to identify orders.

Frequently asked questions about the AI-commerce order study

The 65% denominator is the inner group of AI-agent orders. The outer/inner areas and person counts do not represent measured shares of all ecommerce.

When using these numbers in an internal presentation, label the population as AI-agent orders observed on Forter’s network. Preserve that scope when placing them beside your own results or Japanese-market data.

Does 65% mean AI places most online orders?

No. It is the share of detected AI-agent orders attributable to agents operating normal checkout interfaces. It does not reveal AI’s share of all online orders, including those placed by people.

Have AI orders also grown 5.3× in Japan?

This source does not establish that. The multiplier compares Forter’s observed agent orders with their early-August level. A Japanese merchant needs to check what can be identified for its region and services, and on what evidence.

Preparing to be recommended by AI and preparing to receive agentic purchases are related, but recommendation counts cannot show whether checkout succeeded. This study gives merchants a reason to inspect the order route as well.