Obenan Briefing · Signal
Agentic checkout tolls are becoming negotiable. Merchant readiness is the durable control point.
AI commerce surfaces are not converging on one checkout fee or one custody model. They diverge on what the merchant pays, who stays merchant of record, and where payment and fulfillment sit. The part that travels across all of them is current, authorized merchant readiness: the catalog, pricing, availability, policy, location scope, and acceptance state an agent reads before it acts.
Published June 14, 2026
The one line
Pick the work that holds its value as surfaces and fees keep moving. A toll can be renegotiated and a custody model can change, but an agent on any surface still needs the merchant's facts to be current, authorized, and committable. That readiness is merchant-controlled, and it is the part worth owning.
- Published
- June 14, 2026
- Format
- Signal briefing
- Sources
- 5 public, primary
- Development
- AI commerce surfaces, 2025 to 2026
Public sources only. Observed facts and our interpretation are labeled separately.
The 60-second read
Three questions, three answers.
What is actually diverging across AI checkout surfaces, why that makes a single surface a weak thing to build around, and where the durable control point sits for a multi-location merchant.
AI checkout surfaces are diverging on tolls and custody, not converging.
Across 2025 and 2026, the surfaces that let an agent buy from a merchant settled into different models. Fees range from a 4 percent charge on one surface to no commission on another. Merchant-of-record handling and where payment sits differ by surface. The one constant is that the merchant still owns orders, fulfillment, and the facts behind the sale.
A surface and its fee are the negotiable part. Readiness is not.
If you rebuild around one surface and its current fee, you are exposed when the fee moves or a second surface launches at a different price. The merchant-controlled work, keeping catalog, pricing, availability, policy, and acceptance state current and authorized, holds its value no matter which surface a customer uses.
Whether today's fees and custody models last.
The fee a surface charges, whether it stays merchant of record, and where custody sits are still changing. Treat any current fee or zero-fee stance as a snapshot, not a settled rule, and keep the readiness that travels across all of them current.
The evidence, dated
Five public signals, one pattern: the surfaces diverge.
Public materials from the surfaces themselves and from independent reporting, read in one place. Together they show divergence on fees and custody, and they agree on one thing: the merchant still carries the catalog, the order, and the fulfillment.
Instant Checkout keeps the merchant responsible
OpenAI's Instant Checkout and the Agentic Commerce Protocol let an agent complete a purchase inside the chat, while the merchant remains responsible for orders, payment, fulfillment, and acceptance. The surface presents the offer. The merchant still owns the transaction behind it.
A 4 percent fee on ChatGPT checkout sales
PYMNTS reported that merchants pay OpenAI a 4 percent fee on sales made through ChatGPT checkout, on top of existing payment fees. This is a distribution charge for reaching the surface's audience, separate from ordinary processing, and it is the kind of term that can be renegotiated.
Copilot Checkout takes no commission today
Microsoft states that Copilot Checkout takes no commission or affiliate fee today, that merchants remain the merchant of record, and that payments run on existing rails. The wording is present tense, which makes the zero-fee stance a current position rather than a permanent guarantee.
Store Sync and Agent Ready build on existing merchant setup
PayPal positions Store Sync and Agent Ready around merchant catalogs and existing PayPal configurations, with merchants remaining the merchant of record and keeping control of brand and customer communications. The merchant's own catalog and configuration are the input the surface depends on.
OpenAI shifts checkout toward merchant apps
Digital Commerce 360 reported that OpenAI is moving checkout toward merchant apps, citing inventory, tax, pricing, and the constant updates these require as reasons a centralized checkout was hard to scale. The complexity that pushed the shift is merchant-side truth, not the payment step.
Five surfaces, five different answers on fees and custody, one shared dependency. Every model assumes the merchant's catalog, price, availability, and acceptance state are current and correct. That assumption is the durable layer.
What actually travels across surfaces
Merchant readiness is the part that moves with the customer, not the toll.
A checkout fee, a merchant-of-record rule, or a custody arrangement belongs to a surface and can change when the surface changes. Current, authorized merchant readiness belongs to the merchant and is the same input every surface needs before an agent can act with confidence.
Readiness is concrete, not abstract. It is the set of facts an agent reads to decide whether it can present and complete an action, and it has to be current and authorized at the moment of the request, not at the moment of the last feed.
Catalog and product truth
What you sell, described accurately and consistently enough that an agent can match a request to the right item across any surface.
Pricing and availability
The current price and whether the item or slot is actually available now, not a cached value from the last sync.
Policy and acceptance state
Whether you can accept this order under your own rules: deposits, cancellations, eligibility, and the conditions under which a commitment holds.
Location scope
Which location the request maps to, with hours, service area, and per-location facts an agent needs before it acts for that site.
The right payment or fulfillment path
The merchant-chosen way the order is paid for and fulfilled, so the agent routes to a path the merchant actually supports rather than assuming one.
Current and authorized
Every fact above has to be fresh and traceable to the merchant, because an agent acting on stale or unauthorized data fails at the moment that matters.
Where building around one surface costs you
Four ways optimizing for one checkout surface backfires.
The risk is not any single surface. The risk is treating one surface and its current terms as the thing you build around, when the terms move and the readiness underneath is what every surface actually needs.
A brand rebuilds its catalog and operations around one tolled checkout surface, then a second surface launches at a different fee and the brand is least ready exactly where new demand arrives.
A team reads a current no-commission stance as permanent and designs around it, when the surface itself describes the position in the present tense and can revise it.
An operator treats the chat surface as the control point and lets merchant-of-record details, acceptance rules, and location facts drift, so the agent reads stale truth at the moment of purchase and the order fails.
A portfolio spends its effort negotiating one surface's fee instead of investing in readiness that lowers friction and failure across every surface at once.
In each case the surface is fine and the bet is wrong. Tolls and custody models are negotiable and changing. The merchant readiness underneath is the constant, and it is the part worth building.
Three layers
Three layers in an agentic purchase. Only one is yours to own.
An agentic purchase crosses several layers. The surface decides discovery and its own fee and custody terms. The rails settle the money. Underneath both sits the merchant readiness that decides whether the agent can act on correct, authorized facts at all.
The surface and the rails are supplied by others and keep changing. The readiness layer is merchant-controlled, it is assumed by every layer above it, and it is where a multi-location operator should put durable effort.
Checkout surface and its toll
Each AI surface sets its own checkout fee, merchant-of-record handling, and where payment custody sits. These diverge today and can be renegotiated or revised tomorrow.
Surface-controlled
Payment and settlement
Existing payment rails move and settle the money, with the merchant or its provider choosing the path. The protocol presents the offer; the rails clear the transaction.
Rail and provider-controlled
Current, authorized merchant readiness
The catalog, pricing, availability, policy, location scope, and acceptance state an agent reads before it acts. It is assumed by every surface and every rail, and supplied by none of them.
Merchant-controlled
Surfaces and rails are supplied by others and keep shifting. The readiness layer is yours, and every layer above it depends on it.
The operator rail
What to do now, what to monitor, what not to assume.
A bounded action set for the people who decide where a multi-location merchant invests. Separate the merchant-controlled readiness work from the surface-controlled terms you only watch.
Do now
- Make catalog, pricing, availability, policy, location scope, and acceptance state current and authorized, so the same readiness serves any surface a customer uses.
- Map which AI checkout surfaces your customers actually reach, and record each one's fee, merchant-of-record handling, and custody model as it stands today.
- Decide the payment and fulfillment paths you support per location, so an agent routes to a path you can honor rather than one it assumes.
Monitor
- Changes to any surface's fee, commission stance, or merchant-of-record handling, since today's terms are a snapshot, not a settled rule.
- New surfaces and the movement of checkout toward merchant apps, and whether your readiness is in place where new demand lands.
- Whether the live answer an agent gives about each location stays correct as your hours, prices, and availability change.
Do not assume
- Do not assume a current zero-fee or no-commission stance is permanent. Read it as the surface's present position.
- Do not assume one surface has won. The surfaces diverge, and a customer may arrive through any of them.
- Do not assume the surface holds correct facts about your business. The merchant authors and authorizes that truth.
The Obenan point of view
Own the readiness, not the surface.
The useful read of a diverging market is not to pick the winning surface or chase the lowest fee. It is to notice what every surface needs and none supplies. Fees and custody models are the negotiable, moving part. The merchant's current, authorized readiness is the part that travels with the customer across every surface, and it is the merchant's to own.
Our position is consistent and merchant-first. The merchant stays the hero of the transaction: it owns the catalog, the order, the fulfillment, and the facts. Obenan's role is to help keep that merchant truth current, authorized, and committable, so an agent on any surface can act on it. We do not move the money, route the transaction, or run the checkout. We keep the readiness underneath them sound.
Observed versus inferred
What we saw, what we concluded, what we are watching.
The public materials are observed fact, dated and attributable. The conclusion that merchant readiness is the durable control point is our interpretation. We keep them apart on purpose.
Observed
Public materials checked June 14, 2026. PYMNTS reported on January 21, 2026 a 4 percent OpenAI fee on ChatGPT checkout sales. OpenAI's Instant Checkout and Agentic Commerce Protocol keep the merchant responsible for orders, payment, fulfillment, and acceptance. Microsoft states Copilot Checkout takes no commission or affiliate fee today, keeps merchants the merchant of record, and uses existing rails. PayPal positions Store Sync and Agent Ready around merchant catalogs and existing configurations. Digital Commerce 360 reported on March 6, 2026 that OpenAI is shifting checkout toward merchant apps, citing inventory, tax, pricing, and update complexity.
Inferred
That because fees and custody models vary across surfaces and keep moving, the durable value for a multi-location merchant is current, authorized readiness that projects across surfaces, rather than optimizing for one surface or its fee. This is our reading, not a claim by any company named.
Watching
Whether current fees and zero-commission stances hold, how merchant-of-record and custody handling evolve, and how far checkout moves toward merchant apps. If the terms settle or a single surface consolidates demand, this reading is revisited.
What we are not claiming
Six things this briefing does not say.
The evidence is public and bounded. To keep it honest, here is what we deliberately do not claim.
- 01
We do not claim any partnership, endorsement, integration, pilot, or commercial relationship with OpenAI, Microsoft, PayPal, Shopify, or any other company named. We read public materials as an outside observer.
- 02
We do not claim Obenan processes payments, settles transactions, or acts as a payment provider, acquirer, or merchant of record.
- 03
We do not claim Obenan routes transactions, operates a checkout, or owns or implements a checkout or payment protocol.
- 04
We do not claim any fee or zero-fee stance is permanent. The fees and commission positions described are current and can change.
- 05
We do not claim any surface has won or lost. The surfaces diverge, and this briefing names no winner.
- 06
We do not claim the merchant readiness described is certified or committable by any named surface, network, or protocol. It is merchant-authored truth, not a network attestation.
Keep reading
Where this fits.
Deciding where to invest as checkout surfaces multiply?
If you need to separate the surface terms you only watch from the merchant readiness you actually own and keep current, that boundary is the work we operate on.
Public sources only. Observed facts and our interpretation are labeled separately.
Sources
Primary, public sources only. The development date and the date we checked are kept separate so the chronology stays inspectable.
Primary sources
- 1.OpenAI: Buy it in ChatGPT (Instant Checkout and the Agentic Commerce Protocol)openai.com · Public product page · Checked June 14, 2026
Primary source for Instant Checkout and the Agentic Commerce Protocol keeping the merchant responsible for orders, payment, fulfillment, and acceptance.
- 2.PYMNTS: Shopify merchants to pay 4% fee on ChatGPT checkout salespymnts.com · Published January 21, 2026 · Checked June 14, 2026
Independent reporting that merchants pay OpenAI a 4 percent fee on sales made through ChatGPT checkout, on top of existing fees.
- 3.Microsoft Advertising: Agentic commerce and Copilot Checkoutabout.ads.microsoft.com · Public product page · Checked June 14, 2026
Primary source for Copilot Checkout taking no commission or affiliate fee today, merchants remaining merchant of record, and use of existing rails.
- 4.PayPal Newsroom: PayPal launches agentic commerce servicesnewsroom.paypal-corp.com · Published October 28, 2025 · Checked June 14, 2026
Primary source for Store Sync and Agent Ready building on merchant catalogs and existing PayPal configurations, with merchants remaining merchant of record.
- 5.Digital Commerce 360: OpenAI shifts checkout plans in its agentic commerce strategydigitalcommerce360.com · Published March 6, 2026 · Checked June 14, 2026
Independent reporting that OpenAI is shifting checkout toward merchant apps, citing inventory, tax, pricing, and update complexity.
Obenan is an independent observer of these public developments. Obenan holds no partnership, integration, certification, or endorsement with any company named, does not process payments or operate a checkout, and this briefing reflects only public information available on the checked dates.