Zero-Click Commerce: What Happens When the Customer Never Walks Into Your Store

I've seen the merchant's problem from three different chairs, and until recently I didn't realize they were all the same chair.

First, as a merchant myself — my years at DSM, running e-commerce operations hands-on, living and dying by the same numbers every merchant lives by: what a visitor costs, what a visitor converts at, what a customer is worth over time.

Second, from the builder's side — at Beez, working directly with merchants on loyalty programs and customer experience, trying to help them increase customer LTV. That's where I learned something that sounds obvious but isn't: merchants don't lack tools. They drown in tools. What they lack is control — over the relationship, over the data, over the economics of their own repeat customers.

And third, from the financial infrastructure side — the last years I spent building banking-grade products, and from that seat I watched the same merchant issues appear again, just wearing different clothes: disintermediation, dependency on rails they don't own, economics decided by someone upstream.

Same problem, three angles. The merchant builds the business; someone else owns the door the customer walks through.

And then this year I had the opportunity to be in Paris at the Visa Payment Forum, where I managed to get into an all-day, intense workshop on agentic e-commerce — listening to how Visa and the big players actually see this space from the inside. Not the conference-stage version. The working version: protocols, authentication, mandates, who holds liability when an agent buys the wrong thing. I left Paris with one conviction and one big open question.

The conviction: there is no doubt in my mind that users will do product discovery and purchasing through AI agents — whether that's ChatGPT, Claude, Perplexity, or whatever comes next. That part of the future is settled. The curve is too steep, the convenience too real.

The open question — the one that keeps me up at 5 AM — is what happens to the merchant in that future. Because I've sat in their chair, and I know what it feels like when a platform shift arrives and you're holding a spreadsheet instead of an engineering team.

What "zero-click commerce" actually means

You've probably heard of zero-click search: someone asks Google something, Google answers it directly on the results page, nobody clicks anything. The publisher who wrote the article that fed the answer gets nothing. That was the warm-up act.

Zero-click commerce is the same pattern, one floor down — and this time it's not about traffic, it's about the transaction itself.

The sequence looks like this. A customer tells an AI assistant: "I need a gift for my father, he hikes, budget €100." The agent researches, compares, narrows, recommends — all inside the conversation. The customer never visits your category page. Never sees your brand story. Never gets your exit-intent popup (a small mercy, I'll admit). Increasingly, the customer doesn't even check out on your site — the agent completes the purchase through a protocol, and you receive something that looks like an order fell from the sky.

Discovery, evaluation, and now transaction are all migrating into a layer the merchant doesn't control and mostly can't even see.

Figure 1 — Discovery, evaluation, and increasingly the transaction itself migrate into a layer the merchant doesn't own. What's left on the merchant's side is fulfilment. (AI-generated diagram.)

Here's where I stop being just a guy with an opinion

Because this could sound like another AI doomer take, and you know how I feel about beautiful bullshit with no data behind it. So let's look at what's actually measured:

58.5% of US searches and 59.7% of EU searches already end without any click to an external website. When Google's AI Overviews appear, the zero-click rate jumps to an average of 83%, and clicks to the pages below the overview drop by roughly a third (SparkToro; Ahrefs).

AI-driven traffic to US retail sites grew 393% year-over-year in Q1 2026 (Adobe Analytics, tracking over a trillion visits). During the 2025 holidays it was up 693%.

Here's the one that made me sit up: AI agents drove an estimated 20% of global orders during the 2025 holiday season — $262 billion in sales (Salesforce). Not 20% of some niche. Twenty percent of global orders.

And the conversion inversion: in March 2025, AI-referred traffic converted 38% worse than normal traffic. By March 2026 it converted 42% better — an 80-point swing in twelve months (Adobe). Revenue per visit from AI referrals now runs 37% above non-AI traffic.

Figure 2 — In one year, AI referrals went from the worst-converting traffic you could get to the best. (AI-generated diagram.)

Read that last point again, because it's the whole story in miniature. The agents are not sending you browsing traffic. They're sending you decided customers. The deciding is what moved off your site.

Meanwhile, the infrastructure for full machine-to-machine commerce got standardized while most merchants were busy with Q4. What I saw confirmed in Paris is now public: five competing protocols (OpenAI/Stripe's ACP, Google's UCP and AP2, Visa's Trusted Agent Protocol, Mastercard's Agent Pay) define how agents authenticate, negotiate, and pay. OpenAI already walked back its in-chat checkout in favor of merchant-controlled checkout via apps — the industry is converging on a model where the store never gets visited, only queried.

The merchant's dilemma: three doors, all of them bad

My brain works in systems — I can't look at this without asking who controls what, and where the risk accumulates. And when I map it, the merchant faces three options, and I don't like any of them.

Door one: block the agents. Amazon did exactly this — blocked OpenAI's crawlers, and in March 2026 won a preliminary injunction stopping Perplexity's browser agent from buying on its platform. The result? Amazon gets under 3% of ChatGPT's referral traffic while Walmart gets ~20%, Etsy over 20%, Target ~15%. Amazon can afford this; it has an advertising business worth tens of billions to protect and its own demand gravity. A small merchant blocking agents isn't defending a moat — it's turning off the lights in a store that's already hard to find.

Door two: become a feed. Expose your catalog, let the agents scrape it or query it, become a row in someone else's comparison table. You keep the order but lose everything around the order — the brand, the upsell, the relationship, the data. You become a warehouse with a logo. I ran businesses where the differentiation was the experience. A feed can't sell. It can only answer.

Door three: pay the toll. Accept that the agent platforms are the new gatekeepers and buy your visibility — the way merchants accepted Amazon's terms, and before that, Google's ad auction. This is probably what most will do, because merchants are pragmatic. But let's be honest about what it is: a new intermediation tax on top of the old ones.

Figure 3 — Every available door costs the merchant something structural. None of them puts the merchant back in charge of the sale. (AI-generated diagram.)

Notice what all three doors have in common: the merchant is reacting. The layer where the customer relationship used to live — the store — is quietly being removed from the merchant's control.

What do merchants actually want? (Honest answer: I don't know)

This is where I have questions, not conclusions, and I'd rather be straight with you about it.

My hypothesis, from years of building loyalty and LTV tools for merchants, is that they want four things: attribution (know the sale came through an agent — right now roughly 70% of AI referrals are invisible in standard analytics, which is insane), control (decide how agents represent their products and prices), economics (the agent channel shouldn't cost more than the margin it brings), and the relationship (the customer, or at least the data, should be theirs — this was the entire war we fought with loyalty programs, and it's about to be fought again on new terrain).

But here are the questions I keep circling:

Will merchants treat agents as a revenue channel or as a threat? My bet: most will initially see it the way they saw Google Ads and display — just another performance channel to plug into, another line in the marketing budget. And that framing might be the trap. Google Ads sat on top of your store and sent traffic to it. The agent channel doesn't send traffic — it replaces the visit entirely. Treating it as "just another channel" is how you wake up one day and realize the channel owns the customer and you're the fulfillment layer.

How does the performance revenue model evolve when there's no click to buy? The entire economics of digital marketing — CPC, CPM, affiliate last-click — assumes a measurable human journey through pages. In zero-click commerce, what's the unit you pay for? Placement inside a recommendation? A "sponsored answer"? A rev-share on agent-completed transactions? Whoever defines that unit defines the next decade of merchant economics — the way Google's auction defined the last one. I don't think this is settled, and I think merchants should be in that conversation now, not after the toll booth is built.

Do merchants even care about the relationship, or only about the order? Marketplace history suggests the uncomfortable answer. Merchants complained about Amazon's terms for twenty years — and kept selling on Amazon, because volume beats philosophy. Will they do the same with agents, and is "merchant control" something builders like me assume they want more than they actually do?

What does "selling" even mean when the buyer is a machine? Your agent can't be charmed. But it can be negotiated with, qualified, offered bundles, given reasons. Does the art of the store clerk translate into protocols — or does commerce flatten into pure price/spec matching?

I don't know. I have opinions, but opinions are cheap and I've been wrong before — I once burned millions on BNPL assumptions that looked just as solid on a whiteboard.

What happens to the small and medium merchant?

This is the part that actually worries me, because at Beez and at DSM I worked with exactly these businesses, and I know what a platform shift looks like from their side of the table.

Look at who's winning the early agentic traffic: Walmart, Etsy, Target, eBay — players with protocol partnerships, structured catalogs, and teams for this. Shopify merchants get Agentic Storefronts handed to them by the platform. The independent SMB on a custom stack gets... a robots.txt file and a prayer.

I see two possible futures, and I honestly can't tell you which one we're heading into:

The optimistic scenario: the great democratization. Amazon blocked the crawlers and locked itself out of the fastest-growing discovery channel in commerce. For the first time in fifteen years, there's a crack in the wall. AI agents, in theory, recommend the best-fit product, not the biggest ad budget. A small merchant with genuinely good products and clean, structured data could get recommended next to giants without paying Google Ads tolls. The agent doesn't care about your brand budget. It cares about whether your product fits the brief.

The pessimistic scenario: the new toll booth. The agent platforms become the new gatekeepers, and they're better at it than Google ever was, because they don't send you ten blue links to compete in — they make one recommendation. In a world of one recommendation, you're either the answer or you're invisible. No second page of search results to slowly climb. And once the platforms monetize that recommendation — and they will, they always do — small merchants pay a tax that makes the Google Ads auction look gentle.

My gut says both happen, sorted by category. Commodity products flatten into price/spec matching, and the small merchant there gets squeezed to the margin. Considered purchases — the hiking gift, the good furniture, the thing that needs advice — those still reward merchants who can sell, even to a machine. But that's a gut feeling, not a finding.

What can actually be done

I'm a builder, so I can't end on "the future is uncertain." Here's the ladder as I see it, in order of effort:

See the traffic. If ~70% of AI referrals are invisible in your analytics, you're making decisions on a number that's wrong by a factor of three. Segment it, tag it, look at it weekly. This costs an afternoon.
Become machine-readable. An agent can only recommend what it can parse. Clean, structured, honest product data — specs, constraints, availability, real trade-offs — is table stakes now, not an SEO nicety.
Choose your door on purpose. Block, feed, or toll — pick deliberately, per category, with the margin math in front of you. Defaulting into one of them is how you end up in the worst version of it.
Build the clerk.

That fourth one is where I want to go deeper, because it's the thing I'm personally obsessing over, and it starts from first principles.

Forget e-commerce for a second. Strip commerce down to what a good merchant actually did for the last two thousand years. He put his store where the foot traffic was, and he advertised. But then the important part happened — the part we digitized away. A potential customer walked in, and the merchant greeted him. Assessed him. Asked what he was looking for. Watched which products his eyes kept returning to. Engaged with more questions. Recommended. Presented options adapted to the customer's constraints — "do you need it today, or is something I can order for next week acceptable?" And when he sensed the customer was close — that specific hesitation every good salesman recognizes — he moved to close: a deal, a bundle, a discount, a reason to come back.

That entire sequence — greet, assess, ask, watch, recommend, adapt, close — is the actual art of commerce. Websites kept the shelf and amputated the clerk. Conversion optimization, recommendation engines, exit-intent popups — twenty years of e-commerce tooling is, if you squint, one long attempt to compensate for the missing clerk with statistics.

Now here is the flip that I can't stop thinking about. Everyone looks at agent-to-agent commerce and sees the final amputation: not even the shelf survives, just a feed that machines query. I see the opposite. For the first time in history, the clerk can scale.

Because an AI agent on the merchant's side can do every one of those moves, natively, in protocol: greet the incoming shopping agent and verify who it represents; assess and ask — budget, constraints, urgency, "does your customer need this now or can it ship later"; watch what the agent examines — and an agent's questions are richer browsing data than any clickstream ever was; recommend and present options with real trade-offs, not a sorted list; and close — propose the bundle, the discount, the loyalty enrollment, dynamically, per conversation, under the merchant's margin rules.

Figure 4 — The website kept the shelf and lost the clerk. A merchant-side agent gets the clerk back — and, for the first time, one per customer. (AI-generated diagram.)

A physical merchant could never afford a personal clerk for every customer. A website could never have a real conversation. But a merchant-side agent can hold a unique, individual selling conversation with every single customer's agent, simultaneously, at zero marginal cost. An infinite number of clerks, each one treating each customer as an individual — that's never been possible before. Not in the bazaar, not on Amazon.

The merchants who win the agentic era won't be the ones with the cleanest feed. They'll be the ones whose clerks sell best. That's the company I want to build.

The question I leave you with

Fifteen years of building taught me that platform shifts don't announce themselves to the people they hurt most. The merchants who got killed by Amazon weren't stupid — the shift just looked small until it didn't. Right now, AI referrals are roughly 1% of site traffic. Google still sends ~190 visitors for every one ChatGPT sends. It looks small.

But the growth curve is vertical, the conversion is better than any channel we've ever measured, and after Paris I've seen with my own eyes how seriously the biggest payment infrastructure on earth is taking this. So the question isn't whether zero-click commerce arrives. It's whether, when it does, the merchant is standing at the door of their own store — greeting the agents, setting the terms, doing the selling — or whether the store becomes a warehouse that agents query and merchants pay to be found in.

I know which one I'd rather build for. If you're a merchant seeing agent traffic you can't explain, or a builder working on this layer — reach out. I'm figuring this out in public, as usual, before I'm ready.

Building from Cluj-Napoca, Romania

Sources / further reading

SparkToro & Datos — Zero-Click Search Study (US and EU click-through behaviour)
Ahrefs — AI Overviews Reduce Clicks (zero-click rate under AI Overviews)
Adobe Analytics — AI Traffic to US Retail Sites (Q1 2026 growth, conversion and revenue-per-visit data)
Salesforce — 2025 Holiday Shopping Report (agent-influenced share of global orders)
SE Ranking — ChatGPT Referral Traffic Report
Conductor — 2026 AI Search Traffic Report
Protocol specs: OpenAI/Stripe Agentic Commerce Protocol, Google UCP and AP2, Visa Trusted Agent Protocol, Mastercard Agent Pay