Who Really Owns Your Customers in Agentic Commerce?

Over the past week, everyone on LinkedIn has been celebrating Agentic Commerce. Open standards. Small fees. Merchant of record. Safe, right?

Not exactly. We have seen this before. Marketplaces, app stores, and ad platforms all promised the same. In the end, they took the margins and controlled visibility.

What matters now is ownership, whether you will still control your customer, your data, and your independence once the buying interface belongs to Shopify, OpenAI, or another gatekeeper.

First, here is how the rails are being laid, so you can see who will own what.

What just happened in AI-led shopping

At the end of last month, OpenAI announced Instant Checkout inside ChatGPT, which is co-developed with Stripe. The feature is currently live in the United States with Etsy and is slated to be available with Shopify next.

OpenAI calls this an open standard: the Agentic Commerce Protocol or ACP. They state that merchants remain the merchant of record, data sharing is minimal, and merchants pay a small fee for each completed purchase.

On the surface, it sounds straightforward.

Independent reporting backs this up. Reuters confirmed the fee model and the near-term scope, including multi-item carts and regional rollout. Axios reported the same timing and highlighted Stripe’s role beneath the “protocol” label.

Google is pursuing a parallel track with its Agent Payments Protocol or AP2. Instead of shopping results, AP2 focuses on normalizing agent-initiated payments across cards, wallets, and processors.

Different angle, same direction: agents transact on behalf of users, and the stack is being standardized.

Now that the rollout is clear, let us define what agentic commerce actually means.

What is Agentic Commerce?

Agentic Commerce (ACP) flips the script on online shopping. Instead of you clicking through a site or adding items to a cart, an AI agent takes the request, finds options, compares them, and completes the purchase. The agent not only recommends, but also decides and executes.

Think of it as moving from “show me options” to “just handle it for me.” A shopper can say, “find me a durable gym bag under $100 that ships this week,” and the agent will scan catalogs, filter based on policies and reviews, and check out on their behalf.

In B2B, an AI agent could reorder inventory when stock runs low, evaluate suppliers, or manage recurring contracts with little to no human touch.

The core traits are autonomy, speed, and context. AI agents rely on structured product data, customer history, and real-time signals to act more like digital proxies than search assistants.

The focus shifts away from store design or brand storytelling to the completeness and credibility of your product data.

Understanding the model is the easy part. Spotting the traps before you hand over control is the hard part.

Nine traps that could cost you control in Agentic Commerce

1: Ranking neutrality versus commercial incentives

Neutral rankings don’t exist.

Every platform claims it’s fair, but incentives always seem to creep in. If enabling “Instant Checkout” becomes a ranking factor, that’s not user experience. It is instead self-preference disguised as optimization.

According to OpenAI, product results in ChatGPT are organic and unsponsored. However, buried in their own explanation is the key: availability, price, quality, seller status, and whether Instant Checkout is enabled. That last one tilts the field. In the same way, Amazon favors Prime sellers and Apple nudges developers toward its billing system, AI agents will reward compliance with their own stack.

Google has been testing its Agent Payments Protocol (AP2) with partners, and early pilots already funneled shoppers into single marketplaces during trials. Even the suggestion of preferential routing changes how merchants behave and their long-term strategy. If the rules say “play ball or sink in rank,” merchants will comply, and competition will get distorted.

In the meantime, regulators are watching. The European Union’s Digital Markets Act prohibits self-preferencing by designated gatekeepers and requires that business users maintain the ability to communicate directly with end-users. If agent platforms operate as gatekeepers in Europe, those obligations will apply to ranking, data access, and communication paths. The UK competition authority has also warned about concentration risks in foundation models that span search, cloud, and apps.

The unresolved issues are:

• Will an agent demand a native checkout or protocol to qualify for top placement?
  
• Will that count as a preference?
  
• How will neutrality across sellers and payment providers actually be proven?

Once a platform controls the rails, it sets the tolls.

2: Fees that start small and grow with dependence

Call it what it is: a new tax on eCommerce.

OpenAI describes “a small fee” for completed purchases. That’s how it always starts. Every merchant who has lived through Amazon, eBay, or app store commission creep knows the playbook. Fees begin as a nudge to drive adoption, then climb once sellers depend on the channel.

Instant Checkout is not a gift. It is a revenue stream for OpenAI. If agent-led sales grow into a meaningful share of your revenue, your margin structure rests entirely on a platform’s willingness not to squeeze. That is not a strategy; it is wishful thinking.

The reality check comes down to three things:

• Will there ever be a published cap, or will rates float in the dark?
  
• Will some payment methods incur higher fees than others?
  
• And when the next fee increase comes with little notice, what option do you actually have?
  

Costs are painful, but losing the customer relationship is worse.

3: Who really owns the customer relationship

If the agent controls the interface, you don’t own the customer; you just fulfill the order.

OpenAI says merchants remain the “merchant of record.” Orders flow through your system, and only the data needed to complete the transaction is shared back. On paper, that sounds good for privacy. In practice, it means the agent becomes the primary interface for discovery, decision, and even support.

This is the Amazon problem all over again. When the platform holds the identity and mediates every message, loyalty doesn’t belong to you. It belongs to the layer that decides what the buyer sees and who they hear from.

Europe’s Digital Markets Act has already flagged this risk, requiring gatekeepers to let business users communicate directly with end users. But regulation lags reality. If agents withhold verified emails, consent records, or service identifiers, your ability to build retention crumbles even as your sales volume grows.

The crux is this:

• When the order is placed, will you receive the customer’s verified email, consent flags, and a direct line to contact them?
• Or will every interaction, like a shipping, warranty, or even a service update, be mediated by the agent?

Industry Perspective:
Liam Quinn, Director of Innovation at Visualsoft, adds:

“There’s some really interesting and important points made here about the customer relationship and data ownership. From what we know as of now, a Shopify Merchant would continue to collect the same customer data when an order is placed – and I don’t expect anything to disrupt that short term.

Longer term, there is the ability for that middle layer to have more influence on data ownership especially if ordering this way gains traction. Similarly (although less likely) there would be the chance that Agentic commerce brings with it the long discussed decentralisation of data ownership, allowing consumers to have more say on where their data is shared.

Regardless, if either or neither of these happen – what we do know is the transactional aspect becomes more commoditized in conversational commerce. Meaning building brand recognition and customer relationships elsewhere becomes even more significant.“

When AI holds customer identity, your data becomes your only voice.

4: Data quality and feed governance as a competitive moat

With Agentic Commerce, your feed is now your storefront. If your product data is sloppy, you are invisible.

Agents do not care about your homepage design or brand colors. They care about structured attributes like size, material, warranty, return policy, sustainability claims, and shipping speed. If you cannot provide them in clean, machine-readable fields, you will not appear when an agent applies filters.

This is not merchandising. It is data governance. Feedonomics said it best: data becomes the storefront. And in an agent-driven world, incomplete or outdated fields do not just cost you polish, they cost you distribution.

The fix is simple, but not easy. Treat your feed like a product. Assign an owner, add monitoring, version your schema, and publish policy data as fields that machines can parse instead of prose that crawlers have to guess at.

The uncomfortable truth: your ability to compete no longer depends on how persuasive your design is, but on whether your feed passes a machine’s checklist. Are you ready for that shift?

You can perfect your feeds, but if fewer shoppers ever click through, you are fighting to win visibility in a funnel that is shrinking by the day.

5: Search displacement and the shrinking top of the funnel

AI is eating clicks. If answers live in Google’s AI-overviews box, fewer people ever reach your site.

Publishers and retailers are already seeing what happens when AI-generated answers appear before links. Pew Research found that users who see an AI summary click less often. The Wall Street Journal reported that AI search captured a growing share of desktop queries.

While this is not new, it mirrors the rise of zero-click search, where users get what they need right on Google without ever leaving the results page.

As fewer clicks reach your pages, more decisions happen in the answer box or inside an agent. If you are preparing for agent channels, you are hedging that shift. If you do not, you are exposed to a slow leak in demand that you cannot control.

What you need to model now:

• How dependent are you on classic search? How much uplift would you need from shopping agents to offset those losses?
• How much margin disappears once you factor in agent fees?

If fewer clicks reach your website, trust signals will decide who even gets considered.

6: Review integrity and trust signals in an agentic world

If your reviews are not verified, you are not even in the game.

Agents will lean heavily on review data to decide who ranks. And the rules are tightening fast. In the United States, the Federal Trade Commission now bans the buying and selling of fake reviews and can issue civil penalties. In the European Union, the Omnibus Directive requires proof that reviews are authentic and prohibits the use of undisclosed ranking signals.

This means your legal risk and your ranking are tied together. If you cannot prove that your reviews come from real buyers, you risk both fines and invisibility inside the shopping agent.

The solution is not optional. Verified purchase badges must be baked into your feed. Review verification methods must be exposed in AI-readable form. Trust fields, such as warranty, authenticity, and return windows, must be available as structured data that an agent can parse instantly.

The litmus test is straightforward. If OpenAI were to demand proof of review integrity tomorrow, could you provide it without scrambling?

After you pass the trust test, price fairness becomes the next filter.

7: Personalized pricing, fairness, and the coming patchwork of rules

Dynamic pricing is about to collide with regulation.

Agentic channels open the door to negotiation and personalized offers. That sounds like innovation, but it also triggers fairness questions that regulators are already circling. In the US, the FTC has begun reviewing how personal data is used to set individualized prices. States are moving even faster. For example, New York, where I live, has moved to require disclosure when personal data informs pricing, and that rule has already triggered litigation.

This is not just an American issue. Consumer groups are pushing dozens of state-level bills targeting algorithmic pricing. International regulators are also warning that dynamic ranking and pricing systems can quietly steer shoppers toward higher costs without a clear value rationale. None of these bans personalization. It does mean that opaque personalization will be a target for enforcement.

The test for you is whether your pricing logic can withstand scrutiny. Could you explain it to a regulator? Could you defend it to a buyer? If not, build the guardrails now: disclosure language that agents can surface, logic you can justify, and hard limits on which data can influence price. Because once agents start negotiating on your behalf, regulators will not be far behind.

Regulation is only one side of the risk. The bigger question is whether your business can survive if a single platform changes the rules.

8: Protocols today, lock in tomorrow

What happens when your main sales channel belongs to someone else?

OpenAI and Stripe present the Agentic Commerce Protocol as an open standard, backed by Stripe’s Shared Payment Token, so that agents can transact without forcing a processor switch. Google’s AP2 carries the same “open” promise, supported by a long list of card, wallet, and processor partners. On the surface, that looks safe.

But the core risk is not about code. It is continuity. The platform that runs the agent can raise fees, change eligibility rules, or shut you out altogether. And there is nothing you can do about it. Merchants who built their growth on Amazon’s marketplace already know how this movie ends: sudden commission hikes, forced ad spend to stay visible, and overnight suspensions that cut revenue off at the knees.

The stakes are even higher now. Agents sit closer to the buyer than any marketplace ever did. If an agent becomes the default path to purchase in your category, you are no longer just a tenant in a marketplace. You are giving control of discovery, checkout, and follow-up to a single intermediary. That is dependence at a level deeper than most brands have faced before.

Ask yourself now: if Shopify or OpenAI killed agent checkout tomorrow, how much of your revenue would vanish? And how long could you survive without it?

Even with the right protocol, your business still depends on a platform that can change the rules overnight.

9: Platform risk and business continuity

What happens when the rules change overnight?

If Shopify or OpenAI adjusts a policy, raises fees, or cuts access, your revenue can take an immediate hit. Publishers have already felt this with Google’s AI overviews reducing clicks. Merchants have lived through sudden marketplace commission hikes or search algorithm updates. Agent channels will be no different.

The core truth is this: you cannot outsource your entire buyer relationship to an agent and still expect resilience. If an agent pauses your feed, you should still be able to reach your customers through your email list, loyalty programs, and direct communication channels. This is why the EU’s Digital Markets Act requirement for direct merchant–buyer communication is not an abstract legal point. It is a survival mechanism.

Before you sign onto an agent protocol, scrutinize the terms for three things:

1. Notice periods for fee changes – so sudden cost spikes do not blindside you.
   
2. Minimum data access guarantees – so you always know who your customer is and can serve them directly.
   
3. Communication rights – so you can contact buyers for service and safety updates even if the discovery started inside an agent.
   

Your business continuity plan should assume that agent platforms can and will change the rules.

Platforms have always moved the goalposts, and agentic channels will be no different. The only real hedge is what you own: your data, your customer list, your direct communication rights.

Without that, continuity is at the mercy of corporate policy updates.

Where does this all lead?

What the next two years might look like

Here is my personal, grounded view of where the trajectory is heading:

Discovery will move upstream into agents.

AI answers will keep eating search clicks. Your top of funnel will shrink, while assistants drive higher-intent buyers straight to checkout. The paradox: less traffic but more qualified orders.

Fees will stay modest until they don’t.

Platforms need adoption first, so expect token fees in the short run. But the long game looks like every other digital channel: fees creep up, premium features move behind higher tiers, and “choice” becomes a tax.

Protocols will multiply, but interoperability will be expensive.

OpenAI will push ACP, Google will rally AP2, and others will emerge. They will all promise openness. In reality, the best ranking and smoothest flows will favor each platform’s native stack. Multi-agent participation will be possible, but only if you can afford the engineering overhead.

Regulators will stop watching and start acting

Europe’s self-preference rules will get tested on agent ranking. US regulators will zero in on disclosure in personalized pricing and review authenticity. State-level rules will add more compliance burden, and cross-border sales will force merchants into a legal patchwork.

Merchants will split into two camps

Some will go all-in, accept the fees, and optimize their entire strategy around agent feeds. Others will hedge, keeping agents as a channel but doubling down on owned assets, such as public structured data, loyalty programs, and direct customer communication. Both strategies can work, but drifting in the middle is not a strategy.

The bottom line: the agent layer is here, and it will shape discovery and checkout. The question is not whether you participate, but whether you participate on your terms or theirs.

Industry Perspective:
Andy Cresodina, Co-Founder and CMO at Orbit Media, adds:

“Brands that market through a Big Tech platform (Google, Amazon, OpenAI) do so at a cost. It’s a trade-off.

They gain exposure but lose control. So rather than rush to be visible in a new channel, think carefully about the long-term cost to the brand.

Is the greater revenue worth weaker customer relationships?

 

If that is the trajectory, the obvious next step is to ask: what should merchants actually do right now?

A checklist for merchant independence

If you decide to play in the agentic channel, do it on your terms. Here is the checklist I would hold myself to before handing over any control.

1. Demand a portability test.
   Your checkout should run on at least two agent ecosystems without a single line of code change to your core systems. Push for proof that your ranking will not collapse if you use an alternative processor or protocol flavor. If the platform cannot show you this, you are looking at lock-in, not openness.
   
2. Treat your catalog like an API product.
   Your product feed is no longer a side file; it is the storefront. Publish structured data for attributes, certifications, warranty, returns, delivery windows, and sustainability claims. Add machine-readable trust signals for reviews and authenticity. Track errors, freshness, and schema changes the same way you monitor inventory.
   
3. Model your fee exposure.
   Start by assuming a small percentage takes on every agent-driven order. Then run scenarios at higher rates, because fees rarely move down. If your margin collapses when fees rise, you do not have a sustainable channel.
   
4. Keep a direct line to your buyer.
   Negotiate for identifiers and consent flags. In Europe, insist on your legal right to communicate directly for service and safety. If the platform insists on mediating every message, assign a real cost to that loss of loyalty and remarketing.
   
5. Document your ethics bar.
   Write down your pricing logic, the data you use, and the data you refuse to use, and your review verification process. Assume you will need to explain it to a regulator or a journalist. If you cannot defend it, do not ship it.
   
6. Use Shopify’s built-in AI to raise your baseline.
   Use Shopify’s built-in AI to get your house in order. Sidekick and Shopify Magic let you automate description generation, image cleanup, and store management. Use them to raise the quality of the data agents will consume, while you remain in control of your stack.
   

The checklist is your shield; the broader fight is how the industry will be shaped.

Where this leaves the industry

Right now, the agentic layer is being defined by a handful of platforms. OpenAI and Stripe are pitching ACP as merchant-friendly. Google is framing AP2 as an open payment standard. Both wave the flag of “open” and “trust.” But words are cheap. The real test will be what happens when ranking, defaults, and token formats collide with commercial incentives.

If history is any guide, control will not remain neutral. Amazon did not build the marketplace to be neutral. Apple did not run the App Store to be neutral. Google did not run Shopping to be neutral. Each promised openness, then tightened its grip as the channel matured. Agent commerce will be no different unless merchants demand otherwise.

If innovation moves only at the speed of two or three gatekeepers, then there is risk. Merchants compete on feed quality, structured data, and delivery promises, while the real leverage stays upstream with the platforms that own discovery and checkout. That is not a healthy balance.

The counterweight has to come from merchants themselves. Data quality must be treated as a strategic initiative, not merely a housekeeping task. Portability must be demanded, not assumed. Customer relationships must be defended wherever the law allows.

If you do not fight for these things now, you will not get them later.

Questions to keep in mind

If you are considering agent channels, stop and ask yourself these questions before you give up control.

1. Should the act of enabling a native agent checkout count as a ranking factor at all, even if it improves conversion, or should protocols forbid it to protect neutrality across payment stacks?
   
2. If agent channels become a double-digit share of a merchant’s revenue, what is the right oversight model for fees and for sudden changes to ranking or eligibility?
   
3. What identifiers and consent artifacts should be transferred with an agent order by default so that the merchant relationship with the human buyer is not reduced to a black box?
   
4. How should regulators apply self-preference rules to agent context, and what would a fair audit of agent ranking look like in practice?
   
5. Where should the line be drawn on personalized pricing for inside agents, and which disclosures will maintain trust while still allowing for a fair value exchange?

Answering these questions honestly will tell you whether you are shaping the future of your business or letting someone else do it for you.

Closing thought

You have two paths.

Either treat Agentic channels as another rented storefront, hoping the rules never change, or use them as leverage while building independence everywhere you can. Clean data, portability, direct communication, and trust signals are not nice-to-haves, they are your insurance policy.

I am not saying avoid Agentic Commerce. I am saying do not confuse access with control. Use the rails, but never hand over the keys. The day you cannot reach your customer without permission from an AI platform is the day you stopped running your own business.

The choice is here now. Build for agents, protect your independence, and keep the customer relationship where it belongs.

Further Reading

Make your Shopify store AI‑friendly → How to Get ChatGPT and Perplexity To Index Your Shopify Store
Understand how AI is reshaping search visibility → AI SEO for Ecommerce
Avoid risks in dynamic, AI‑driven pricing → The Role Of AI In Predictive Dynamic Pricing
Optimize your product feeds for discoverability → The Shopify Catalog API: A Deep Dive Into AI‑Powered Commerce
Need help with data, feeds, or AI strategy? → Shopify SEO & Data Migration Audit