OpenAI charges 4 percent. But the real cost of agentic commerce runs much deeper than a single line item.

In January 2026, OpenAI flipped a switch that changed the economics of online retail. Shopify merchants selling through ChatGPT's Instant Checkout now pay a 4 percent fee on every completed purchase, automatically deducted before payouts reach their accounts. The fee kicked in after a 30-day free trial, and it lands on top of everything merchants already pay Shopify.

Four percent. That was the number that made the headlines. That was the number that launched a thousand LinkedIn posts comparing OpenAI to Amazon. And that was the number that entirely missed the point.

Because 4 percent is not the cost of agentic commerce. It is the visible fraction of a much larger economic shift, one that touches payment processing, catalogue readiness, customer data ownership, and brand equity. When you stack every layer, the true cost of selling through an AI agent looks nothing like the figure on the invoice.

The mechanics are straightforward. A shopper asks ChatGPT a purchase-intent question. The AI recommends a product, and if the merchant has opted into Instant Checkout, the buyer can complete the transaction without leaving the chat. OpenAI takes 4 percent. Shopify takes its standard processing fees. The merchant gets what remains.

OpenAI built this on the Agentic Commerce Protocol, an open standard co-developed with Stripe that handles discovery, purchasing, and fulfilment. The protocol is open source. The fee is not.

The initial reaction from the industry split predictably. Max Sinclair, CEO of Azoma.ai, called it "a land grab for the future of commerce," but also "a bargain, considering Amazon's 8 to 15 percent referral fees." Others were less charitable. SmartCompany framed the story as merchants facing "double fees," paying OpenAI on top of their existing Shopify processing costs.

Both reactions focused on the wrong thing. The 4 percent is the most visible cost. It is not the most significant one.

To understand what agentic commerce actually costs a merchant, we need to compare three channels side by side. Take a $100 sale and trace where the money goes.

Direct website sale: approximately $4.20 total cost. The merchant pays standard card processing, typically 2.9 percent plus $0.30 through Shopify Payments. On a $100 transaction, that is $3.20. Add hosting, platform subscription costs amortised across transactions, and you land around $4.20. The merchant keeps roughly $95.80 and owns the entire customer relationship, every data point, every email address, every browsing pattern.

Amazon marketplace sale: approximately $19.50 total cost. Amazon's referral fee runs 8 to 15 percent depending on category, with most consumer goods sitting around 15 percent. Add the FBA fulfilment fee (roughly $3.50 on an average item) and the advertising spend now effectively required to maintain visibility, and the all-in cost on a $100 sale easily reaches $19.50 or more. The merchant keeps roughly $80.50 but surrenders significant customer data to Amazon and competes directly with Amazon's own private-label products.

AI agent checkout sale: approximately $9.20 total cost (and climbing). This is where it gets interesting. Start with OpenAI's 4 percent fee: $4.00. Add Shopify's standard processing at roughly 3.2 percent: $3.20. That gets you to $7.20 in visible fees alone. But Forrester's analysis puts the true per-transaction cost at 7 percent or more before accounting for additional tech and catalogue readiness costs. Factor in the structured data preparation, API maintenance, and inventory feed management that AI checkout requires, and the realistic all-in cost on a $100 sale sits around $9.20.

The positioning matters here. As Sinclair pointed out, OpenAI's $400 million monthly burn rate is forcing rapid monetisation. Google can afford to figure out its model later. OpenAI needs revenue now. That urgency shapes the fee structure merchants are being asked to accept.

The fee stack above accounts for the transactional costs. But the most consequential expenses in agentic commerce never appear on a payment statement.

Catalogue readiness is not free. To participate in AI checkout, merchants need structured product data that agents can parse: clean descriptions, standardised attributes, real-time inventory feeds, and pricing that updates dynamically. For large retailers with established product information management systems, this is incremental work. For mid-market merchants, it is a project. Dwayne Gefferie, writing in Payments Strategy Breakdown, mapped the full economic architecture of agentic commerce and found that merchants face new infrastructure costs at every layer, from data feed management to protocol compliance. On a standard $100 online credit card purchase, merchants already pay $3.10 split six ways across interchange, network assessments, processor markup, and fraud detection. Agentic commerce adds entirely new cost layers on top of that existing stack.

Customer data evaporates. This is the cost that should keep merchants awake at night. When a customer buys through your website, you capture their email, browsing history, purchase patterns, and preferences. When that same customer buys through ChatGPT, you get an order and a shipping address.

McKinsey's research on agentic commerce describes this as a fundamental shift in who owns the customer relationship. The AI agent consumes detailed product and pricing data while owning the customer interaction, reducing merchants' direct visibility into consumer needs. According to a Deloitte 2026 Retail Industry Outlook, 81 percent of surveyed retail executives believe generative AI will weaken brand loyalty by 2027.

That data loss has a compounding cost. Every customer acquired through an AI agent is a customer you cannot retarget, cannot upsell through email, and cannot build a loyalty profile around. Over a 12-month customer lifetime, the inability to run personalised retention campaigns could cost more than the 4 percent transaction fee ever will.

Brand becomes invisible. AI agents do not browse. They do not notice your packaging, your brand story, or your homepage redesign. They evaluate structured data: price, specifications, ratings, delivery speed, return policy. Chris Jones, Managing Director at PSE Consulting, told The Paypers that the fundamental difference is where intent gets resolved. In traditional ecommerce, the consumer arrives at your site and begins discovery. In an agentic journey, preferences are gathered, constraints applied, and intent resolved inside the AI interface, before the buyer ever touches the merchant's ecosystem.

Here is the competitive wrinkle that makes the cost calculation even more unstable. Not every AI platform charges the same way.

Google's AI Mode checkout, powered by its new Universal Commerce Protocol, currently charges no additional transaction fee beyond standard processing. Google co-developed UCP with Shopify, Etsy, Wayfair, and Target, and the protocol has been endorsed by more than 20 companies including Visa, Mastercard, American Express, and Adyen. Checkout is rolling out now in the US, with purchases secured through Google Pay.

Microsoft Copilot similarly charges no additional checkout fee.

The absence of fees from Google and Microsoft throws OpenAI's 4 percent into sharper relief. But the "free" model comes with its own cost structure. Google is testing Direct Offers, which let retailers promote exclusive discounts to shoppers expressing purchase intent in AI Mode. Brands like Petco, e.l.f. Cosmetics, and Samsonite are already participating. The fee is not on the transaction. It is on the visibility.

This is the pattern we have seen before. The platform launches free, establishes distribution, captures merchant dependency, then monetises. Google's advertising machine already generates $238 billion from shopping and search ads. The question is not whether Google will monetise AI Mode commerce, but how.

Gefferie's analysis in Payments Strategy Breakdown quantifies the stakes. Agentic commerce threatens $459 billion in annual commerce monetisation that depends on human discovery: $140 billion in retail media, $238 billion in Google search and shopping ads, $165 billion in Meta advertising, and $18.5 billion in affiliate marketing. When AI agents bypass all of those channels, the money has to come from somewhere. It will come from merchants, one way or another.

So what should merchants actually do with this information?

The conversion numbers provide one compelling counterargument to the fee anxiety. ChatGPT traffic reportedly converts at 15.9 percent compared to 1.8 percent for Google organic search, a nearly nine-fold improvement. If those numbers hold at scale, a merchant paying 9.2 percent all-in on an AI agent sale might still achieve better unit economics than paying for Google Ads clicks that convert at under 2 percent.

Jones at PSE Consulting framed it precisely: the real question is not "Is 4 percent high?" but "What is a high-intent, agent-sourced buyer worth to my business?" When the agent does the work of qualification, comparison, and intent resolution, this is not a fee. It is AI monetising commercial outcomes rather than engagement metrics.

But the conversion advantage does not neutralise the data and brand costs. McKinsey projects that agentic commerce could influence $3 trillion to $5 trillion in global retail by 2030. Forrester's Q3 2025 CMO Pulse Survey found that 92 percent of US B2C marketing executives are already developing agentic commerce strategies. This is not a niche channel. It is becoming the channel.

The merchants who navigate this well will treat agentic commerce the way sophisticated sellers treat Amazon: as a distribution channel with known costs that requires deliberate margin management. Forrester advises merchants to monitor profit and loss per channel, factor agent fees into product pricing, and avoid selling low-margin products through answer engines entirely. The parallel to dropshipping economics is instructive: you do not put your entire catalogue in every channel. You put the right products in the right channels at the right margins.

The companies building for this future are already re-architecting. McKinsey urges merchants to create agent-ready digital infrastructure, what some are calling "agentic landing pads," optimised for AI interaction rather than human browsing. The 2025 to 2026 window, McKinsey argues, is when early movers establish infrastructure advantages that late entrants cannot replicate.

Four percent is what OpenAI charges. The agent tax is everything else: the processing fees, the catalogue investment, the customer data you will never own, and the brand equity that dissolves when your product becomes a row in an AI's comparison table. The merchants who thrive will be the ones who calculated the full cost before they opted in.

Related reading from Major Matters:

What is your full cost of selling through an AI agent, and have you actually done the maths?