For most of ecommerce history, the cost of selling something online sorted into two buckets. There was the cost of getting a shopper to your store, advertising, SEO, email, affiliates. And there was the cost of taking their money once they arrived, the payment processing fee. Two budgets, two teams who argued about them.
A third bucket has quietly opened. In January 2026, OpenAI began charging Shopify merchants a fee on every sale completed through ChatGPT's checkout. Not a payment fee. Not ad spend. A cut of the transaction itself, paid to the assistant that closed it. The number is 4 percent, and most merchants will first meet it as a line item shaving their payout that they did not budget for and cannot easily see.
The instinct is to file this next to Stripe's processing fee and move on. That instinct is wrong, and acting on it will cost you. The agent fee behaves far more like a marketplace commission than like the cost of moving money. Understanding the difference is the difference between pricing your products correctly for the next few years and slowly bleeding margin you never saw leave.
Where the agent fee came from
To see why the agent fee is a new species rather than a new instance, remember what the existing fees actually pay for.
A payment processing fee, the familiar 2.9 percent plus 30 cents, pays for the machinery of moving money. Interchange to the card-issuing bank, a slice to the card network, a margin to the processor. In the United States the all-in cost for a typical small or mid-sized merchant runs roughly 2.2 percent to 3.5 percent per card transaction, with interchange the largest single component. The processor does not care where the sale came from. It is a utility charge on the act of payment.
A marketplace take rate is a different animal. When you sell on Amazon, eBay, or Etsy, the platform charges a referral or commission fee because it brought you the customer. Etsy charges a 6.5 percent transaction fee on the sale price. eBay's final value fee sits around 13 percent in most categories. Amazon's referral fees vary by category and, once fulfilment and advertising are stacked on, the all-in cost of selling there can climb much higher. The take rate is not paying for plumbing. It is rent on access to demand. The marketplace owns the audience, the search box, the moment of discovery, and it charges you for standing in front of it.
For two decades these two cost types stayed cleanly separated because of one structural fact: your own store and a marketplace were different places. You paid processing on your store, commission on Amazon. The shopper physically went to one or the other.
Agentic commerce collapses that separation. When a shopper asks ChatGPT to find a pair of running shoes and buys without leaving the chat, where did that sale happen? It happened on your store, in the sense that you are still the merchant of record, you ship the goods, you own the customer data and the return. But it also happened inside someone else's interface, mediated by someone else's audience and ranking. The assistant did the job a marketplace does, so it charges the fee a marketplace charges, while everything else about the transaction still looks like your own store. That is the new bucket: a take rate on a sale that, on paper, is still yours.
OpenAI's path to that fee shows how new this is. When OpenAI announced Instant Checkout in September 2025, launching with US Etsy sellers and more than a million Shopify merchants to follow, it said merchants would pay a small fee on completed purchases and declined to put a number on it, calling the terms confidential contracts with its partners. It stressed two things: shoppers pay nothing extra, and product results are organic and unsponsored, with checkout-enabled items given no preference. The fee was real from day one. It just had no public number.
The number arrived in January 2026. As Shopify merchant onboarding opened up, the figure was confirmed at 4 percent on completed ChatGPT checkout sales, with a 30-day free trial that begins after a merchant's first qualifying order. That is when the agent fee stopped being an abstraction and became a line on a payout statement.
What the 4 percent fee is, and what it is not
Precision matters here, because this fee is easy to overstate and easy to understate, and both errors lead somewhere costly.
What it is: a 4 percent fee charged by OpenAI to the merchant on the value of a sale completed through ChatGPT's Instant Checkout. It is calculated on the full order value, item price plus shipping plus tax, and deducted before the merchant's payout reaches their bank. If a customer returns the item, the fee is reversed in proportion to the refund. The merchant remains the merchant of record throughout, keeping the customer relationship, the contact details, and control of fulfilment.
What it is not: it is not a universal rate for AI-assisted commerce. This is the most important qualifier in the whole subject, and the one most coverage gets wrong. The 4 percent figure is specific to OpenAI's ChatGPT checkout and to the commercial arrangement struck with Shopify. It is not a standard. As of early 2026, sales completed through the in-chat checkout surfaces of Google's AI Mode and Gemini, and Microsoft's Copilot, carried no additional fee of this kind. Reporting consistently framed that as the situation for now, which is the right way to hold it: not a permanent gap, but not a settled rule either. Anyone telling you the agent tax is 4 percent across the board is selling a tidy story the facts do not support.
It is also not a payment processing fee, and it does not replace one. The 4 percent sits on top of the merchant's existing Shopify Payments charge and Shopify's own platform fees. A merchant does not switch processors to use ChatGPT checkout. The card is still charged the normal way, by the normal processor, at the normal cost. The agent fee is an additional, separate layer, and that stacking is the part that hurts. It deserves real math.
The margin math, stacked
Consider a Shopify merchant on the Basic plan selling a 100 dollar item through ChatGPT Instant Checkout. The fees, layer by layer, look roughly like this.
OpenAI's agent fee at 4 percent takes 4 dollars. Shopify Payments processing at 2.9 percent plus 30 cents takes about 3.20 dollars. A Basic-plan merchant who keeps Shopify Payments avoids Shopify's extra transaction fee, but the picture changes fast if any non-Shopify gateway is in the mix, where a Basic-plan surcharge of around 2 dollars on a 100 dollar sale would apply. Take the cleaner case of Shopify Payments only. Stacked together, the merchant is parting with somewhere around 7 dollars of a 100 dollar order before a single cost of goods, pick-and-pack, or shipping subsidy is counted. The merchant keeps roughly 93 dollars where, on a sale through their own storefront with no agent involved, they would have kept about 97.
Four dollars on a hundred does not sound frightening in isolation. The danger is what it does to the number that actually runs an ecommerce business, which is not revenue but contribution margin, the slice left after the cost of the product and the cost of selling it.
Take a merchant whose product carries a 40 percent gross margin, typical for a mid-market consumer brand. On a 100 dollar order, 60 dollars is cost of goods. That leaves 40 dollars of gross margin to cover everything else: payment fees, fulfilment, marketing, overhead, profit. The 4 dollar agent fee is 10 percent of the merchant's entire contribution margin on that sale, gone to the assistant. For a merchant running a 25 percent gross margin, common in commoditised categories, the same 4 dollar fee consumes 16 percent of contribution margin. The headline rate is 4 percent of revenue. The rate that matters is the share of margin, several times larger and varying sharply by how thin the merchant already runs.
This is the arithmetic that makes a marketplace take rate feel heavier than a processing fee even when the percentages look similar. The processing fee is a fixed cost every channel carries. The agent fee is an extra, channel-specific levy that lands only on agent-mediated sales and stacks on everything else. It is rent, and rent comes straight out of profit.
Who can absorb it and who cannot
Whether 4 percent is a rounding error or an existential threat depends entirely on who you are.
The merchants who can absorb it comfortably share a few traits. High gross margins, so the fee is a small fraction of contribution. Strong brand pricing power, so they are not competing purely on price. And a real customer acquisition cost problem that the agent channel might relieve. If a brand is paying steep prices for paid search and social clicks that convert poorly, a sale that arrives pre-qualified through an assistant, with the shopper already decided, can be cheaper all-in than the channel it replaces, even after the 4 percent. For that brand the agent fee is a customer acquisition cost, and possibly a favourable one.
The merchants who cannot absorb it are the mirror image. Thin-margin operators in commoditised categories, resellers, anyone whose contribution margin is already a sliver. For them, a fee worth 15 percent or more of contribution margin is the difference between a profitable order and a break-even one. These are also, awkwardly, the merchants most exposed to agent-mediated buying, because when an assistant optimises a purchase it weighs price, availability, and return policy heavily, precisely the terrain where commodity sellers compete. The merchants with the least margin to give up are the ones whose sales are most likely to migrate into the channel that takes the cut.
There is a tempting escape hatch: raise prices inside the agent channel to recover the fee. It mostly does not work. An assistant comparing options will notice, and a higher price is a direct hit to ranking in a channel where ranking is the whole game. And OpenAI's framing has been that Instant Checkout does not change the price a shopper sees. The fee is structurally a merchant cost, not a shopper surcharge, by design. The merchant absorbs it or declines the channel.
Where the agent fee goes from here
The 4 percent figure is a snapshot of an immature market, and three forces will pull on it.
The first is competition between assistants. Google and Microsoft entered agentic checkout without charging a comparable fee, which puts obvious pressure on a 4 percent rate that a rival is undercutting at zero. That pressure cuts both ways. It could push the going rate down. It could also mean the rivals are buying merchant supply now and will introduce their own take rates once shoppers are habituated, the well-worn platform pattern of subsidise first, monetise later. Either way, a single durable industry rate is unlikely soon.
The second force is whether assistants stay neutral referees or become paid marketplaces. Today OpenAI insists ranking is organic and unsponsored. The commercial gravity of a high-intent shopping surface is enormous, and it is reasonable to expect pressure over time for sponsored placement, a higher fee tier for better visibility, or other mechanisms that make the agent channel look more like a retail media network than a neutral guide. If that happens, the 4 percent transaction fee becomes the floor of a more complex pricing stack, not the whole of it.
The third force is the most underrated, and it is performance. The agent fee only matters if agent-mediated checkout actually becomes a major channel, and the early evidence is mixed. Walmart, which tested around 200,000 items through OpenAI's Instant Checkout, reported that in-chat purchases converted at about a third of the rate of shoppers who clicked out to Walmart's own site, with its head of product and design, Daniel Danker, calling the experience unsatisfying. In March 2026 OpenAI changed course, stepping back from running the checkout itself and moving toward merchant-controlled checkout while focusing its own effort on product discovery and comparison. Fewer than 30 of Shopify's millions of merchants had gone live with Instant Checkout by then. The take rate model did not vanish in that pivot, but the episode is a clear signal that the channel, the fee, and who collects it are all still being negotiated in public.
What is not in doubt is the direction. McKinsey's October 2025 analysis projects that agentic commerce could account for as much as 1 trillion dollars of orchestrated US retail revenue by 2030, and 3 trillion to 5 trillion dollars globally. When a channel reaches that scale, the entity that controls discovery and checkout inside it will charge for the privilege, because every intermediary in retail history eventually has. The 4 percent may move. The agent transaction layer, sitting on top of payment processing and every existing channel cost, is now a permanent feature of the commerce stack. Merchants who model it as a real, separate cost line, channel by channel, will price and plan correctly. Merchants who treat it as a quirk of one platform's checkout will keep being surprised by their own payout statements.
The takeaway
The 4 percent ChatGPT checkout fee is not a payment fee and not a universal AI commerce rate. It is a marketplace-style take rate, charged by an assistant for closing a sale, stacked on processing and platform fees, and it lands hardest as a share of contribution margin rather than of revenue. Whether it is survivable depends on a merchant's gross margin and pricing power, not on the headline percentage. The rate will move as assistants compete and as some inch toward paid placement, but the layer itself, an agent toll on agent-mediated sales, is here to stay. Treat it as a named cost in the channel mix, not a footnote, and the math stays honest.
Council summary
This post argues that OpenAI's 4 percent ChatGPT checkout fee is a marketplace-style take rate, not a payment fee, and that its real bite is measured against contribution margin rather than revenue. The council verified the central facts: the rate applies specifically to Shopify merchants on OpenAI's Instant Checkout, it began in late January 2026 with a 30-day trial after a first qualifying order, and Google and Microsoft charged no comparable fee at the time of writing, so the post correctly avoids framing 4 percent as a universal AI-commerce rate. We tightened the McKinsey projection to its published range, fixed the stacked-fee math so the Shopify Payments case no longer double-counts a transaction surcharge, and corrected the title of Walmart's product chief. The takeaway is to treat the agent fee as a named, channel-specific cost line and model it against margin, because the headline percentage understates the damage on thin-margin products.
Comments