You're Optimising for Search While Commerce Rebuilds Around You

Two commerce protocols launched within weeks of each other in late 2025 and early 2026. They don't just change how products are discovered — they change how products are evaluated, negotiated, and purchased, without the consumer ever visiting a website. The trillion-dollar projections for this shift are both too high and too low: too high because today's autonomous purchasing is near zero, too low because they miss the enormous invisible influence AI already exerts on how every consumer discovers what to buy. The honest position: agentic commerce is pre-Sputnik — and Sputnik has a launch date.

If you work in SEO or AEO, you've spent the last year debating how to rank in AI search results. While you were doing that, two things happened that change the question entirely.

First, OpenAI and Stripe shipped a commerce protocol that lets ChatGPT's more than 800 million weekly users buy products without leaving the conversation. It's been live since September 2025.

Second, Google announced its own protocol — with Walmart, Target, Shopify, Visa, and Mastercard — that covers everything from discovery through checkout to returns. It started rolling out in January 2026.

These aren't product announcements. They're infrastructure. And they mean the question is no longer "how do I rank in AI search?" It's "what happens when AI search becomes AI commerce?"

How did we get from AI search fragmentation to commerce protocol wars?

In our previous analysis, There Is No 'AI Search', we showed that what gets called "AI search" is actually at least three independent systems. The data was unambiguous: when 11 sites lost Google organic rankings, they lost Google AI citations in lockstep — but Perplexity citations barely moved (a decline of just 2.9%), and across AI assistants, only 12% of cited URLs overlapped with Google's top 10 — with Perplexity and ChatGPT pulling from fundamentally different source pools.

That fragmentation was about discovery — which AI platform recommends which sources. What's happened since runs deeper. The platform divergence we identified is becoming a full commerce stack divergence: not just different discovery engines, but different checkout systems, different payment rails, and different rules for who captures the customer's intent.

What are the competing commerce protocols — and why do they matter?

Every protocol and every platform in this landscape is competing for the same thing: the moment a consumer expresses what they want. The protocol war is an intent signal war. The merchant is always downstream.

OpenAI's Agentic Commerce Protocol (ACP)

OpenAI and Stripe launched ACP in September 2025. It enables ChatGPT Instant Checkout — a consumer asks for a product, ChatGPT finds it, and the purchase completes inside the conversation. The consumer never visits the merchant's website.

The mechanics: merchants implement REST API endpoints, ChatGPT sends checkout requests, the merchant accepts or declines on their own systems. Payment is processed through Stripe's Shared Payment Token — the AI agent never sees the card number. The merchant fee is 4% on completed purchases (plus standard Stripe processing), refunded on returns.

Two details matter for practitioners. First, product rankings in ChatGPT are organic and unsponsored — relevance-based, not pay-to-play. Second, ACP is open source under Apache 2.0, available at agenticcommerce.dev. Any AI platform could adopt it, not just ChatGPT.

Current partners include Etsy (live), Shopify (onboarding over 1 million merchants), Instacart, Target, and DoorDash.

Google's Universal Commerce Protocol (UCP)

Google announced UCP on 11 January 2026 at the National Retail Federation conference. Where ACP focuses on checkout, UCP covers the full commerce lifecycle — discovery, checkout, and post-purchase (tracking, returns, customer support). Agents query business profiles to discover capabilities and payment options dynamically.

The partner list reads like a roster of global commerce infrastructure: co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by over 20 organisations including Adyen, American Express, Best Buy, Flipkart, Macy's, Mastercard, Stripe, The Home Depot, Visa, and Zalando. Integration runs through REST APIs, Google's Agent2Agent protocol (A2A), and Anthropic's Model Context Protocol (MCP).

Google also announced complementary tools: a Business Agent — a branded AI agent that shoppers can chat with directly on Google Search (live with Lowe's, Michaels, Poshmark, and Reebok) — and new Merchant Center data attributes that go beyond keywords to include answers to common questions, compatible accessories, and substitutes.

The rest of the landscape

Amazon has refused to join either protocol. It is building proprietary everything — Rufus AI for product search, Alexa+ for voice commerce, and "Buy for Me" — a feature that lets consumers shop other retailers' websites without leaving Amazon's app. The intent signal stays inside Amazon's wall.

Shopify launched Agentic Storefronts in December 2025, enabling checkout through ChatGPT, Perplexity, and Microsoft Copilot. Its new "Agentic Plan" opens the Shopify Catalogue to brands that don't yet have a Shopify store.

PayPal launched agent-ready payments in October 2025, including "store sync" for AI discoverability, partnered with Perplexity for AI-native checkout.

Klarna has its own Agentic Product Protocol. Coinbase launched Agentic Wallets in February 2026 with the x402 crypto-payments extension — 50 million transactions already processed.

The strategic picture: these protocols aren't competing features. They're competing to own the moment a consumer expresses what they want. A merchant integrated with UCP is visible in Google's agentic surfaces. A merchant integrated with ACP is visible in ChatGPT. Neither gives you the other. Shopify merchants get both because Shopify is building integrations with each. Independent businesses without a platform intermediary need to choose — or implement both.

Why are the agentic commerce numbers so confusing?

This is the section that matters most. Every article you've read about agentic commerce either calls it a trillion-dollar revolution or dismisses it as a collective hallucination. Both camps cite data. Both camps are accurate. Both camps are measuring different things with the same label.

The case that it's massive

The projections are staggering:

• McKinsey projects $3–5 trillion in global orchestrated revenue from agentic commerce by 2030 — with US B2C alone reaching up to $1 trillion.
• Salesforce reported that AI and agents influenced 20% of all global Cyber Week 2025 orders — driving $67 billion in sales across 1.5 billion shoppers.
• Adobe's data shows AI-driven traffic to US retail sites surged 670% year-over-year on Cyber Monday 2025 — and 693% across the full holiday season.
• HUMAN Security tracked agentic traffic growth of 1,300% in the first eight months of 2025, with year-over-year growth reaching 6,900% by late 2025.
• Morgan Stanley estimated in December 2025 that 23% of Americans had made an AI-assisted purchase in the past month, and projected $190–385 billion in US e-commerce from agentic shoppers by 2030.

The case that it's nothing

The sceptics' data is equally credible:

• ChatGPT shopping remains tiny in absolute terms. An OpenAI-affiliated study of 1.5 million ChatGPT conversations found only 2.1% were classified as "purchasable products" queries. Over Thanksgiving weekend, ChatGPT referrals to e-commerce apps accounted for just 0.82% of sessions.
• Flagship Advisory Partners found that AI agents accounted for less than 1% of total retail and travel traffic during the 2025 holiday season. Growth had tapered by late 2025.
• Channel Engine's Marketplace Shopping Behavior Report (January 2026) found that while 58% of shoppers now use AI tools for product research, only 17% feel comfortable completing a purchase through an AI agent.
• A study of 973 e-commerce sites (Kaiser & Schulze, SSRN) found that affiliate links convert 86% better than ChatGPT referrals — despite AI-generated product recommendations showing 4.4x higher conversion rates in controlled tests.
• E-commerce itself — the supposed foundation of agentic commerce — still accounts for only 16–20% of total retail after 25 years.

Resolving the paradox

These numbers aren't contradictory. They're measuring three different things and calling all of them "agentic commerce."

The $3–5 trillion projections primarily measure AI influence — the vast, growing pool of consumers who use AI to decide what to buy, even when the purchase itself happens on a conventional website. The sceptics' data (<1%) measures AI autonomy — agents acting independently to complete transactions. Both are accurate. Both use the same label. This is why the industry discourse is confused.

The gap between 58% research adoption and 17% purchase comfort — a 41-point trust deficit — is the single number that matters more than any projection or dismissal. It measures the distance between AI's proven influence on discovery and its unproven ability to close a sale. Every business decision about agentic commerce should start from this gap, not from the trillion-dollar headlines.

There's also an invisible layer that neither side's metrics capture. When a consumer asks ChatGPT for a recommendation, receives an answer, and then navigates directly to that business's website to make a purchase — that's an AI-influenced sale that appears in no agentic commerce metric. It shows up as direct traffic or a branded search. Adobe notes this explicitly: "While the base of users remains modest, the uptick shows the value AI can deliver as a shopping assistant." The influence iceberg has most of its mass below the measurement waterline.

MetaRouter's analysis names this precisely: when shopping happens through AI agents, the customer's discovery and consideration process occurs inside the conversational interface, invisible to the merchant's analytics. Attribution collapses. Personalisation breaks down. The consumer decided what to buy before they ever reached the merchant's site — and the merchant has no idea how.

Why should anyone take this seriously if adoption is near zero?

The standard response to early-stage metrics is: "Let me know when the numbers are real." It's a reasonable instinct. It's also wrong in this case — not because the numbers are real yet, but because the infrastructure is. And the infrastructure has specific dates.

This is not a forecast deck. These are shipping products with named partners and live transaction data. The infrastructure is being built at unprecedented speed by unprecedented players — Google, OpenAI, Stripe, Shopify, Walmart, Visa, Mastercard. In the language of one industry observer: "We're pre-Sputnik launch phase. Everyone is building the spaceship, but no one has really launched it yet" — Juan Pellerano-Rendón, CMO of Swap, speaking to Modern Retail in January 2026.

But infrastructure is necessary and not sufficient. Voice commerce had infrastructure too. Amazon launched Alexa in 2014. Google Home followed in 2016. The technology worked. Consumers didn't adopt it for commerce. The question is whether agentic commerce will repeat that pattern.

That question is best answered by the people making the strongest case that it will.

What's the strongest case against everything we've just presented?

The moderate objection: "The same skills transfer"

There's a reasonable argument that the professionals reading this analysis are already well-positioned for the agentic commerce shift — even without changing anything.

The logic: structured data, reviews, content quality, consistent business information — the signals that make a business visible to AI search are the same signals that make it discoverable by AI commerce agents. If you're already optimising for AI visibility, you're already doing the work.

Michael Komasinski, CEO of Criteo, articulated this position in December 2025: AI-powered shopping assistants should be understood as an incremental layer that complements existing channels — "a new lane to shopping, not a detour." He told Beet.TV at CES 2026 that the real competitive advantage lies in access to high-quality commerce data at scale, not in adopting a fundamentally new paradigm.

Scot Wingo, CEO of ReFiBuy, compared agentic commerce to autonomous vehicles: the industry will ultimately reach full autonomy, but needs to "walk up to it" incrementally — and the walking uses familiar skills.

This objection has genuine force. For businesses that already have high-quality structured data, strong reviews, consistent information across platforms, and content structured for natural language queries — the transition to agent-readiness may be less dramatic than the protocol announcements suggest.

Where the objection breaks down is at the edges. Agent-readiness requires capabilities that go beyond current SEO and AEO practice: machine-readable product feeds compatible with ACP and UCP specifications, checkout infrastructure that agents can interact with programmatically, and content structured around problems solved rather than keywords targeted. Google's new Merchant Center attributes — answers to common questions, compatible accessories, substitutes — point to a data model that's richer than what search optimisation currently demands. An InfoQ analysis of UCP noted that being "discoverable" means having data quality sufficient for an AI agent to surface you as the primary option, regardless of brand size. That's a higher bar than appearing on page one.

The fierce objection: "Collective hallucination"

The strongest argument against everything we've presented comes from two independent sources who should be read in full.

Eric Seufert, a 20-year e-commerce veteran, published "Agentic Commerce Is a Collective Hallucination" in September 2025 on his Media, Ads + Commerce Substack. His argument, paraphrased: every capability currently labeled "agentic commerce" already exists in non-agentic form. The fundamental flaw is that agentic commerce violates retail platforms' core motivations — controlling the customer relationship and monetizing first-party data through advertising. Amazon and Shopify actively resist third-party agent access because they want exclusive ownership of discovery and behavioural data. Seufert draws a direct parallel to the voice commerce predictions of the late 2010s — when Amazon Echos became ubiquitous, identical forecasts were made, and they proved empty. He argues the predictions are essentially interchangeable. If agentic commerce becomes a well-worn consumer behaviour, the largest retail platforms would capture it themselves rather than cede it to third parties. He followed up with a second piece on Marketecture maintaining the same position.

---

Roi Iglesias, a partner at invidis impact, published a detailed analysis in February 2026 — making it among the most current sceptic assessments available. His central concept is "the verification tax": for every minute AI saves a consumer in research, the consumer spends another minute verifying they're not being hallucinated into a bad purchase. Per the Channel Engine data he cites, even when AI provides a "perfect" recommendation, 95% of consumers still perform at least one manual verification step — checking third-party reviews, visiting the brand's website, or cross-referencing prices — before buying. The 2% AI agent conversion rate (from OpenAI's study of 1.5 million ChatGPT conversations) reflects a trust deficit, not a technical limitation. His most compelling comparison: Amazon's Just Walk Out technology. It was technically brilliant — automated checkout using computer vision and sensors. Consumers experienced what Iglesias calls "phantom friction" — anxiety from lacking a tangible transaction moment. Amazon pivoted the technology away from its own stores toward B2B licensing. The parallel to AI-mediated commerce is direct: solving the technical problem doesn't solve the psychological one.

These aren't fringe positions. Seufert has two decades in e-commerce. Iglesias's analysis is data-driven and precise. Emily Pfeiffer, a principal analyst at Forrester, told GeekWire: "The experiences that are out there today, in my opinion, are extremely premature."

This analysis does not claim to refute these arguments. What we can say is this: the DISCOVERY impact is real and measurable now. Adobe's 670% traffic growth is not a projection — it's observed data from over a trillion visits to US retail sites. The 44–60% of consumers using AI for product research is documented by McKinsey and BCG/Mirakl independently. Whether AI ever completes transactions autonomously at scale is genuinely uncertain. But AI is already shaping which products consumers discover, which businesses they consider, and which purchases they make through conventional channels. The honest position holds both facts simultaneously: the autonomous transaction future is speculative, and the AI-influenced discovery present is measurable.

What should you actually do about this?

The evidence supports three layers of preparation, sequenced by urgency. The first two are actionable now. The third is worth monitoring.

Layer 1: Discoverability (do now)

This is where current AI visibility and future agent-readiness overlap most heavily — and where the moderate counter-argument has the most force. If you're already doing this well, you have a genuine head start.

• Structured data and schema markup. IBM's analysis is explicit: agents need machine-readable product data, standardised attributes, and clear metadata. If your product information exists only as prose on a webpage, agents can't parse it.
• Reviews and community presence. BCG and Mirakl found that AI agents are "highly sensitive to product reviews" — positive reviews increase recommendation probability even at higher price points. Presence on Reddit, forums, and review sites provides the third-party validation that agents weight heavily.
• Consistent business information. NAP consistency, verified Google Business Profile, attributed claims. The signals Onely identified as driving AI citations — attributed statistics (+22% lift), expert quotations (+37%) — are the same signals agents use for recommendation decisions.
• Per-platform optimisation. The Two Realities framework still applies. Google AI and Perplexity use different source preferences. A Google-only strategy creates a platform-sized blind spot.

Layer 2: Evaluability (build toward)

This is where agent-readiness diverges from traditional search optimisation.

• Problem-solving content framing. SAP's NRF 2026 guidance was direct: tag products by the problems they solve, not just their attributes. An AI agent evaluating whether to recommend a product needs to match it against a consumer's stated problem, not a keyword.
• Rich product data beyond keywords. Google's new Merchant Center attributes — answers to common questions, compatible accessories, substitutes — represent a richer data model than standard product feeds. Businesses that populate these fields become evaluable by agents; those that don't become invisible at the evaluation stage.
• Trust signals structured for machine reading. As Chris Riedy, CRO of Ibotta, told Retail Customer Experience: "AI bots aren't looking at display ads, but are looking at the inherent quality and metadata of the product, including its price." The trust signals that matter to agents are verifiable, structured, and attributable — not brand narrative.

Layer 3: Transactability (prepare for)

This layer is genuinely early. It's where the fierce counter-arguments have the most weight, and where caution is warranted.

• Product feeds compatible with ACP and UCP. If you're on Shopify, this integration is being built for you. If you're not, implementation will require dedicated resources — Stripe noted that initial integration may require additional staffing.
• Checkout infrastructure agents can interact with. This means API endpoints that can receive and respond to purchase requests programmatically — a capability most small and medium businesses don't have today.
• Payment processing supporting delegated tokens. Stripe's Shared Payment Token, PayPal's agent-ready solution, and the emerging crypto rails (Coinbase x402) all provide mechanisms for agents to transact without seeing payment credentials. The infrastructure exists. Consumer adoption doesn't — yet.

What does this analysis raise but cannot answer?

This analysis identifies one question we're currently researching and don't yet have sufficient evidence to answer.

If AI agents don't see display ads — and both ACP and UCP default to organic, unsponsored product rankings — what replaces the $600 billion digital advertising ecosystem? Google's Direct Offers pilot (exclusive discounts within AI Mode) and Criteo's new Agentic Commerce Recommendation Service suggest early experiments with agent-era monetisation. Chris Riedy's observation — that agents evaluate inherent product quality and metadata, not advertising — implies a structural challenge with no clear solution yet.

We don't have sufficient evidence to answer this, and anyone claiming they do is speculating. We're researching it. When the evidence is sufficient, we'll publish our findings.

---

Honest caveats

This analysis should be read with three disclosures in mind.

Conflict of interest: Findcraft is an AI visibility consultancy. We provide services related to what this article discusses. The analysis that AI-influenced discovery is growing could serve our commercial interests. We name this because you deserve to evaluate the argument knowing the incentive behind it.

Methodology: This analysis is produced using AI assistance following our M.A.R.C. methodology (Machine-Assisted, Research-driven, human-Curated content). The research is human-curated from 26 sources across seven reference documents, the synthesis is machine-assisted, and the editorial judgment is human. Every source is named and linked where possible.

Source bias: Many of the sources cited in this analysis have commercial interests in agentic commerce succeeding. McKinsey advises retailers on digital transformation. Google and OpenAI built the protocols. Stripe processes the payments. Adobe and Salesforce sell the analytics. We've balanced these with independent and sceptical sources — Seufert, Iglesias, Pfeiffer/Forrester, and Flagship Advisory Partners — but the weight of available industry analysis skews toward optimism. Sceptical voices are harder to find, which is itself worth noting.

---

Further reading

These are independent sources — none of them are affiliated with Findcraft:

• Eric Seufert — "Agentic Commerce Is a Collective Hallucination" — The strongest sceptic case: every "agentic" capability already exists in non-agentic form, and the voice commerce parallel should give everyone pause.
• Roi Iglesias / invidis — "Agentic Commerce: Why AI Is Winning the Search War but Losing the Checkout Battle" — The "verification tax" concept and the Amazon Just Walk Out comparison. The most current independent sceptic analysis available (February 2026).
• Flagship Advisory Partners — Agentic Commerce Market Tracker, January 2026 — Independent data showing AI agents remain under 1% of retail traffic, with growth tapering by late 2025.
• Adobe — Cyber Monday 2025 Report — The 670% AI traffic growth figure, based on over a trillion visits to US retail sites.
• Morgan Stanley — Agentic Commerce Impact Outlook — The $190–385 billion projection and the 23% consumer adoption survey.

---

Frequently asked questions

Findcraft provides AI visibility services for businesses seeking to be discovered by AI search and commerce agents. This article reflects our analysis of publicly available research — including the substantial evidence that autonomous AI purchasing is not yet happening at meaningful scale.

Content produced through the M.A.R.C. methodology — our framework for evidence-based, ethically-governed content. Research drawn from 26 sources across seven reference documents, with source quality assessed against a published hierarchy.