Ultimate Guide to Agentic Commerce

1 in 5 Cyber Week orders last year involved an AI agent. That’s roughly $70 billion in GMV in a single week. And that’s not a projection. That’s Salesforce’s actual data from November 2025.

Here’s what it means: AI isn’t just helping people find products anymore. It’s starting to buy them.

Instead of a customer searching Google, clicking through to your site, and adding something to their cart, an AI agent now handles the intent matching and the transaction — sometimes without the customer ever setting foot on your website.

ChatGPT Instant Checkout has been live since September 2025. Google announced its own protocol in January 2026. The infrastructure is already built. The question is whether your products are set up to be found and purchased through it.

Let me walk you through what’s actually happening, which protocols matter, and what you should be doing right now.

What Is Agentic Commerce?

Why “conversational commerce” is the more accurate frame

You’ll hear “agentic commerce” used to describe a future where AI agents autonomously buy things on your behalf with zero human input. That’s mostly hype right now.

Full autonomy works at the extremes. A $1,000 mattress involves too many variables for any agent to handle alone. A replacement razor cartridge you order every month is already on Subscribe & Save. The real action is in the high-consideration middle ground: the $200 running shoe, the $400 air purifier, the $150 skincare set. These are the purchases where AI eliminates hours of research and consolidates the decision into a single conversation.

By the time a customer clicks a product recommendation from ChatGPT, they’ve already refined their needs through multiple conversations. They’re past awareness and consideration. They arrive ready to buy, not browse. That’s why the conversion data looks the way it does — I’ll get into that in detail later.

ACP vs. UCP: The Two Protocols Explained

Two open standards are competing right now to define how AI agents plug into your store. They are not interchangeable, and confusing them will cost you.

ACP: Agentic Commerce Protocol (OpenAI + Stripe)

ACP was co-developed by OpenAI and Stripe and launched September 29, 2025. It’s what powers ChatGPT Instant Checkout and Microsoft Copilot Checkout.

The core idea: the AI agent handles the checkout UI inside the conversation, and your backend processes the transaction through four API endpoints: CreateCheckout, UpdateCheckout, CompleteCheckout, and CancelCheckout.

If you’re on Shopify, this is already available to you. Activation is automatic through the Shopify catalog integration. Etsy U.S. sellers are already live too, enrolled automatically through Offsite Ads. The implementation lift for Shopify merchants is close to zero.

The catch is the economics. OpenAI charges a confirmed 4% transaction fee on every completed purchase, on top of standard Stripe processing (~2.9% + $0.30). On a $100 order, you’re looking at about $7.20 in combined fees. And that’s before you account for the email restriction.

UCP: Universal Commerce Protocol (Google)

Google’s Universal Commerce Protocol is a bigger play. Where ACP solves checkout inside ChatGPT, UCP covers the full commerce lifecycle: discovery, purchase, and post-purchase — across Google Search AI Mode, Gemini, and Lens.

It was announced at NRF 2026 in January and co-developed with Shopify, Etsy, Wayfair, Target, Walmart, and 20+ other partners including Adyen, American Express, Best Buy, Mastercard, Stripe, The Home Depot, and Visa.

The key difference for you: with UCP, you keep the full customer relationship. Email rights, loyalty data, merchant-of-record status. All yours. The trade-off is that you’re now competing for one of three recommendation slots in an AI Overview instead of ten blue links.

ACP vs. UCP: side-by-side

Will one protocol win?

I don’t think so. Major retailers are already running both. Walmart joined ChatGPT (ACP) and announced a Google Gemini partnership (UCP) at the same conference. Shopify co-developed UCP with Google while simultaneously enabling 1M+ merchants on ChatGPT Instant Checkout.

The consumer doesn’t care which protocol powers their purchase. They care whether their AI assistant can find, recommend, and buy the product without friction.

What This Changes (And Who It Hurts)

Agentic commerce doesn’t affect every business equally. Here’s an honest breakdown of who wins, who loses, and why.

Marketing-first brands are in trouble

The last decade rewarded eCommerce brands that could outspend competitors on ads and out-optimize them on landing pages. Product quality was secondary to marketing execution.

LLMs don’t work that way. They read your specs, cross-reference your claims against other sources, and synthesize third-party reviews. If you claim to be the “best running shoe for flat feet” but your arch support specs don’t hold up under scrutiny, the model doesn’t just rank you lower. It recommends a competitor instead.

You cannot market your way out of inferior product data. That’s a hard shift for brands that built on arbitrage.

What actually changes, stakeholder by stakeholder

Buyers

Buyers win clearly. The research phase that used to take an hour now takes a two-minute conversation. Instead of tabbing between product pages, reading spec sheets, and decoding review sites, the buyer describes what they need and gets a recommendation that already accounts for their constraints. For complex purchases, that’s a genuinely better experience.

Merchants

The economics depend entirely on your product type and margin structure. On ChatGPT you gain access to 900M+ weekly users but lose the direct customer relationship. On Google UCP you keep the customer but face tighter competition for fewer recommendation slots. Morgan Stanley estimates agentic shoppers could drive $190–385 billion in U.S. e-commerce spending by 2030. Capturing that requires being the product an AI recommends, not just a product it can find.

Affiliates and content publishers

Review sites and affiliate publishers are already feeling this. When an AI answers a product question without sending the user to a review article, the click that funded that article never happens. Some publishers are moving behind paywalls. Others are shifting to direct merchant partnerships where brands pay for coverage instead of earning it through commissions. Neither outcome is great for the open web, but that’s where the economics are pointing.

Amazon

Amazon has not joined ACP or UCP. They’re building proprietary AI shopping tools (Rufus, Alexa+, Buy for Me) inside their own walled garden.

Here’s their bind: their most profitable business is advertising, which requires shoppers to browse, scroll, and encounter sponsored placements. A one-prompt purchase skips all of that. Joining agentic commerce means cannibalizing their own ad revenue. Staying out means ceding the discovery layer to competitors like Walmart who have already moved.

Google

Google is the best-positioned platform through this transition. They can monetize both the old model and the new one at the same time. AI Overviews are already generating ad revenue comparable to standard search results. And because the buyer arriving through an AI Overview is typically further down the funnel, conversion rates are stronger — which means advertisers bid higher per click even as overall click volume adjusts.

We found that AI Overviews now appear on 14% of shopping queries, up 5.6x in just four months. That’s not slowing down.

The real competitive shift

Your competitive advantage is shifting from “best website” to “best structured data.” The brands that win in agentic commerce are not the ones with the most sophisticated ad stacks. They’re the ones whose product data AI agents can actually read, verify, and trust.

Product Feeds: The Most Overlooked Factor

I’ll be direct: this is the single most impactful thing most eCommerce stores haven’t done yet. And it’s the prerequisite for almost everything else in this guide.

AI assistants like ChatGPT and Google’s AI Overviews rely heavily on product feeds when recommending products. A Semrush experiment confirmed what we suspected: the top product in ChatGPT and Google Shopping overlapped 75% of the time.

ChatGPT uses Google Shopping results to form its product recommendations. That means your Google Shopping feed is your ChatGPT presence. We’ve seen this play out across client accounts. Stores with incomplete feeds are simply invisible to AI assistants, even when they rank fine in traditional search. If you haven’t submitted a comprehensive feed, your products aren’t showing up. Full stop.

The binary reality

In traditional search, a weak product page might rank on page 3. In AI search, there is no page 3. If your data is incomplete or inconsistent, the model doesn’t rank you lower — it skips you entirely. Missing attributes (no sizing info for apparel, no dimensions for furniture, no shipping speed) can disqualify you from entire categories of queries without any signal that traffic is being lost.

What your feed needs to include

Attribute	Why It Matters for AI	Where to Set It
Title	Must include brand, product type, and key differentiators. Generic titles don’t match natural language queries.	Shopify: Admin > Products > Title
Price + Currency	Required for real-time inventory matching and checkout.	Shopify: Admin > Pricing
Availability	Out-of-stock items damage AI recommendation trust. Agents avoid recommending them.	Shopify inventory tracking
Images (800×800px+)	Multiple angles improve recommendation confidence.	Shopify: Admin > Media
Description	Agents read this to validate category-specific claims. Treat it as a spec sheet, not a sales pitch.	Shopify: Admin > Description
GTIN / UPC / MPN	Required for Google UCP checkout eligibility.	Shopify: Admin > Variants > Barcode
Brand Name	Critical tiebreaker when AI surfaces familiar brands over unknowns.	Shopify: Admin > Vendor
Shipping Speed	Agents filter by delivery time, especially for high-intent buyers.	Google Merchant Center > Shipping
Return Policy	Google UCP requires this for checkout eligibility.	Merchant Center > Return policies

Schema markup: don’t skip it

Beyond your feed, implement schema.org markup on every product page. Fabrice Canel, Principal Product Manager at Microsoft Bing, confirmed that schema markup helps Copilot understand your content.

Other platforms haven’t publicly confirmed the same, but the citation data tells a similar story. SE Ranking found that 65% of pages cited by Google AI Mode and 71% cited by ChatGPT include structured data.

At minimum, implement:

• Product schema
• AggregateRating schema (review apps like Judge.me and Yotpo handle this automatically)
• Offer schema
• Organization schema with sameAs links to your brand profiles on major platforms

One more thing: most AI crawlers cannot execute JavaScript. If your product data loads client-side, it may be invisible to LLMs entirely. Make sure all product information is server-side rendered.

Google UCP: Merchant Center setup

For Google’s UCP, you need more than a standard feed. Google’s Merchant Center implementation guide specifies additional requirements for checkout eligibility:

• Create or update your Google Merchant Center account
• Submit your product feed and verify website ownership
• Configure your return policy in Merchant Center — this is a hard requirement and must include return cost, return window, and a link to the full policy
• Set customer support information (used to generate the Contact Merchant link on order confirmation pages)
• Add the native_commerce: TRUE attribute to eligible products — do this via a Supplemental Feed, not your primary feed, to avoid disrupting regular product ingestion
• Integrate Google Pay
• Test checkout in AI Mode at search.google.com and in the Gemini app

Products ineligible for UCP checkout include subscriptions, personalized or custom goods, age-restricted items, pre-orders, and anything blocked by Google Shopping Policy. Set native_commerce: FALSE on those SKUs.

How to Get Recommended by AI Search Engines

Getting into the checkout flow requires being recommended first. Here’s how that recommendation process actually works, and what you can do to influence it.

How LLMs actually pick products

There are two stages happening every time someone asks an AI for a product recommendation.

Stage 1: Retrieval. The LLM queries external search engines (Google, Bing, Amazon) and collects the top results. Traditional SEO governs this stage. If your product doesn’t rank in the top results, nothing else you do matters.

Stage 2: Synthesis. The LLM takes those results and generates a ranked recommendation. This is where most brands are losing ground right now. Their pages rank in traditional search but get filtered during synthesis because the content doesn’t give the model what it needs to construct a recommendation rationale.

A University of Illinois study (the CORE study) tested 3,000 products across 15 categories on GPT-4o, Gemini 2.5, Claude, and Grok. Products at the bottom of search engine retrieval results had a 0% chance of appearing in the AI’s final recommendations. Not low. Zero. Ranking on page 1 is still the prerequisite for everything.

What LLMs want to see on your product pages

LLMs are trained on the consensus of the internet. If your product page says what every other product page says, you provide no value. The model already knows it. You need to provide information the model can’t get elsewhere.

What works:

• Detailed product specifications in structured HTML (tables, definition lists), not paragraph-form marketing copy
• Ingredient and materials breakdowns for consumable products — models read these to verify claims
• Proprietary test results with specific numbers (e.g., “independently lab-tested for 99.7% filtration,” “rated to -40°F”) — the CORE study found that products with structured reasoning content and quantitative differentiators significantly outperformed those with generic descriptions
• Use-case specificity: “Best for flat-footed runners who overpronate,” not “Best running shoe”
• Comparison tables with honest trade-offs — models that detect exaggeration discount the entire listing

What doesn’t work:

• Generic superlatives (“premium,” “best-in-class”) without data to back them up
• Sustainability claims without documentation
• Marketing copy that describes how a product makes you feel rather than what it does

Third-party mentions: the real trust signal

Here’s the finding that surprised me most: across multiple studies, AI systems consistently pull the majority of their citations from third-party sources, not brand websites. Muck Rack analyzed over a million AI citations and found 82% came from earned media.

The reason makes sense when you think about how LLMs work. They actively scrape Reddit, specialized forums, and expert review sites to form a consensus about which products are trustworthy. Your own website is one input among many — and not the most trusted one.

What this means practically:

• Identify the relevant review publications and niche blogs in your category and make sure your products are covered
• Send products to expert reviewers and set up Google Alerts for your brand and product names
• Implement verified purchase reviews on your product pages (Judge.me, Loox, Yotpo) and push for detailed, use-case-specific reviews — star ratings alone are weak signals
• Respond to negative reviews — models read merchant responses as evidence of how you treat customers

Brand familiarity is a genuine tiebreaker. When an AI recommends a brand the user already recognizes, conversion rates improve significantly. Building your brand outside of LLMs directly improves your performance within them.

Organic search is still the gatekeeper

I want to be direct about this because I see a lot of content suggesting you can optimize for AI search independently of traditional SEO. You can’t. Not fully.

Most AI systems use Retrieval-Augmented Generation (RAG), which means they fetch from the top search results before synthesizing an answer. If you’re not on page 1, you’re not in the pool the model draws from. The CORE study confirmed retrieval order as the hard prerequisite, and we’re seeing the same pattern in our own work. Products that aren’t retrieved cannot be recommended, regardless of how good their content is.

Think of traditional SEO as the ticket into the room. You still need it. But being in the room is just the start. What happens once you’re retrieved is a different game entirely.

Implementation Checklist

Work through Phase 1 before worrying about protocol-specific setup. Too many brands jump to ChatGPT integration before verifying their product feeds are complete. That’s backwards.

Phase 1: Foundation (do this first)

• Confirm U.S. business presence (current requirement for all platforms)
• Verify you are the merchant of record for your products
• Confirm payment processing infrastructure is in place (Stripe, PayPal, or Google Pay)
• Audit your product catalog completeness against the feed attribute table above
• Ensure all product content is server-side rendered, not loaded via JavaScript

Phase 2: ChatGPT ACP

Shopify merchants:

• Enable Shopify Catalog: Admin > Settings > Apps and Sales Channels > Shopify Catalog > Enable
• Verify product sync is active
• Test the single-item checkout flow end-to-end
• Model the P&L: 4% commission + no email remarketing — does it work for your AOV and margins?
• Build a fallback remarketing strategy on other paid channels to replace the email revenue you’ll lose

Non-Shopify merchants:

• Review OpenAI ACP documentation
• Implement ACP API endpoints: CreateCheckout, UpdateCheckout, CompleteCheckout, CancelCheckout
• Configure checkout webhook handlers
• Test end-to-end

Phase 3: Microsoft Copilot Checkout

Same ACP protocol as ChatGPT, but with an important difference: you retain customer email marketing rights on Copilot.

• If using Shopify, Stripe, or PayPal: activation is automatic through partner platforms
• Verify customer data access is configured — you do receive customer data on Copilot
• Set up post-purchase email capture and remarketing sequences
• Test checkout on Copilot.com, Bing, and Edge

Phase 4: Google UCP

Follow Google’s official Merchant Center implementation guide throughout this phase.

• Create or update Google Merchant Center account
• Submit product feed and verify website ownership
• Pass automated account review and enable free listings
• Configure return policy (return cost + window + policy link required)
• Set customer support information
• Create Supplemental Feed adding native_commerce: TRUE for eligible products, FALSE for ineligible categories
• Integrate Google Pay
• Test checkout in AI Mode (search.google.com) and Gemini

Which Protocol Should You Prioritize?

The honest answer is: it depends on your product type, your AOV, and how much of your LTV comes from post-purchase email. Here’s how to think about it.

Prioritize ChatGPT / Copilot ACP if:

• Your AOV is $100 or higher
• Your products are genuinely differentiated — models recommend “best,” not just “cheapest”
• You have strong third-party reviews and brand presence already
• Email marketing is a smaller percentage of your LTV
• You’re on Shopify (near-zero implementation cost)
• You’re targeting early-adopter demographics

Prioritize Google UCP if:

• You sell commodity or heavily price-compared products
• You have a competitive pricing or fulfillment speed advantage
• Email is essential to your LTV model — UCP lets you keep it
• You already have an established Google Merchant Center presence
• You’re a newer brand that needs to build early visibility while accumulating conversion history

Common mistakes I see brands making

• Waiting for the “perfect” implementation. Launch fast and iterate. Early movers accumulate conversion data that compounds into recommendation advantage.
• Treating both protocols identically. The strategies are different.
• Not modeling the email restriction impact before scaling ACP. The 15–20% AOV loss is real.
• Ignoring third-party reviews. This is the strongest AI trust signal and takes the longest to build.
• Assuming your SEO ranking is sufficient. You need to rank AND have content that survives the synthesis stage.

ROI: Does Agentic Commerce Actually Convert?

We analyzed 12 months of GA4 data across 94 eCommerce stores, pulling data from January through December 2025. To keep the comparison clean, we isolated non-branded organic search as the baseline, excluding branded queries, direct traffic, and paid. ChatGPT traffic was identified via the chat.openai.com and chatgpt.com referral sources.

Key findings from our 94-store GA4 study:

• ChatGPT traffic converts 31% better than non-branded organic search (1.81% vs. 1.39% conversion rate)
• Revenue per session is 10.3% higher for ChatGPT vs. organic ($3.65 vs. $3.30)
• AOV is 14.3% lower for ChatGPT traffic ($204 vs. $238) — users tend to buy more affordable options
• ChatGPT traffic grew 1,079% over the 12-month period (January to December 2025)
• Still a small share: ChatGPT generated $474K vs. organic’s $32.1M — just 1.48% of non-branded organic revenue

The growth numbers are real, but 1.48% of revenue means you shouldn’t be abandoning your traditional SEO foundation to chase this. The trajectory matters, but so does the current scale.

Why ChatGPT traffic converts better

The buyer’s journey explanation is straightforward. ChatGPT users have typically refined their needs across multiple conversations before clicking a product recommendation. By the time they arrive on your site, they’ve already moved through awareness and consideration. They’re evaluating a specific product, not browsing a category.

In contrast, traditional search traffic is often exploratory. Users scan multiple results, bounce between sites, and take more time to commit. That’s not a problem — it’s just a different stage of the funnel.

The GA4 attribution problem

Here’s something important: the data above almost certainly understates ChatGPT’s real impact. The most common ChatGPT shopping funnel doesn’t show up as ChatGPT referral traffic in Google Analytics at all.

Here’s how it plays out:

1. A shopper asks ChatGPT which brand of air purifier is best for allergies
2. ChatGPT names your brand as the top pick
3. The shopper types your brand name into Google to find the site
4. They click the branded search result and land on your site
5. GA4 records the sale as branded organic search — ChatGPT gets no credit

The fix is simple: add a single question to your order confirmation page. Something like “Where did you first hear about us?” with AI assistants as one of the options. It takes an afternoon to set up and gives you a much clearer picture of how much of your revenue is actually starting in ChatGPT, Perplexity, or Gemini before showing up as branded search in your analytics.

The market opportunity

Morgan Stanley estimates that agentic shoppers could represent $190–385 billion in U.S. e-commerce spending by 2030, representing 10–20% of online retail market share. eMarketer projects AI platforms will account for $20.9 billion in retail spending in 2026, nearly quadrupling 2025 figures. This is early, but it’s accelerating.

Final Thoughts

The infrastructure is built. The protocols are live. The conversion data confirms the opportunity is real, even if it’s still early.

What to take away from this guide: agentic commerce is not going to replace everything you’re doing in SEO tomorrow. It won’t. Non-branded organic search is still 47x larger than ChatGPT referral traffic as of Q4 2025. Traditional SEO remains the dominant channel. Don’t let the hype convince you otherwise.

But the trajectory is clear. AI Overviews now appear on 14% of shopping queries, up 5.6x in four months. ChatGPT traffic grew 1,079% in 2025 and is still growing. The window for early-mover advantage is open.

The brands that win in this environment aren’t necessarily the ones with the biggest budgets. They’re the ones with the most complete, accurate, and machine-readable product data. The ones that have earned real third-party validation. The ones whose product specs can be verified, not just marketed.

Get your product feed right. Get your schema markup in place. Build third-party review presence. Make sure your organic search foundation is solid. Then layer the protocol-specific implementation on top.

That’s the order of operations. Don’t skip steps trying to get to the exciting part.

---

Sources

• Visibility Labs: ChatGPT Traffic Converts 31% Better than Non-Branded Organic Search
• Visibility Labs: AI Overviews Now Appear on 14% of Shopping Queries
• Visibility Labs: 40-Point eCommerce AI SEO Checklist
• Salesforce: 2025 Holiday Shopping Data
• Morgan Stanley: Agentic Commerce Impact Could Reach $385 Billion by 2030
• eMarketer: Agentic Commerce Reports and Analysis
• Semrush: ChatGPT Searches Google Shopping: Experiment Findings
• SE Ranking: How to Use Structured Data for SEO and LLM Visibility
• Muck Rack: AI Citation Analysis
• University of Illinois: CORE Study
• OpenAI ACP Developer Documentation
• Google Developers: Prepare Your Merchant Center Account for UCP