DEV Community

Cover image for Agentic Commerce Optimization: What 4,491 Merchants Reveal About UCP Readiness
Benji Fisher
Benji Fisher

Posted on • Edited on • Originally published at ucpchecker.com

Agentic Commerce Optimization: What 4,491 Merchants Reveal About UCP Readiness

Agentic Commerce Optimization: What 4,491 Merchants Reveal About UCP Readiness

Every UCP technical guide tells you how to get UCP ready. We decided to measure who actually is.

Since UCP launched, UCP Checker has tracked 4,491 merchants — 4,024 of which are verified and actively serving UCP endpoints. We maintain the largest UCP index of live merchant implementations, and the data tells a story that no theoretical guide can. We've run over 1k agent testing sessions in UCP Playground, consumed 43 million tokens doing it, and watched real AI agents attempt to browse, cart, and buy products across every major ecommerce platform. The result isn't a theoretical framework for agentic commerce optimization. It's a field report.

And the field looks very different from what the guides tell you.

What "Agentic Commerce Optimization" Actually Means When You Have Data

The term "agentic commerce optimization" — or ACO — has entered the SEO lexicon as a catch-all for making your store ready for AI-powered shopping agents. Most of the early writing treats it like a checklist: add Schema.org markup, update your Merchant Center feed, structure your product data. That advice isn't wrong. It's just incomplete, because it's built on assumptions about how agents will behave rather than observations of how they actually do.

ACO, measured empirically, is the practice of optimizing your ecommerce stack for the specific patterns that AI agents exhibit when they interact with UCP endpoints. Those patterns are surprising. Agents don't browse the way humans do. They don't use carts the way humans do. And the failure modes that block them from completing purchases are not the ones you'd predict from reading the spec alone.

The data we've collected across 4,024 verified UCP merchants tells a concrete story about what matters, what doesn't, and where the real optimization opportunities are hiding.

UCP Stack Layers — capability adoption across verified merchants

The Real State of UCP Readiness

Let's start with what's working. Of the 4,024 verified merchants in UCP Registry — the open UCP directory where agents discover merchants — capability adoption breaks down like this:

  • Checkout: 4,003 merchants (99.5%)
  • Cart: 3,987 merchants (99.1%)
  • Product discovery: Near-universal
  • Identity: 3 merchants
  • Payment: 0 merchants

Read those last two numbers again. Three merchants support identity. Zero support native payment. This is the defining feature of UCP's current state: the bottom of the funnel is wide open, but the capabilities that would make agentic commerce truly autonomous — knowing who the customer is and processing payment without a handoff — are functionally nonexistent.

The spec migration numbers are more encouraging. When the v2026-04-08 specification dropped, 3,994 out of 4,022 tracked merchants had migrated within four days. That's a 99.3% adoption rate in under a week, which speaks to the platform-driven nature of UCP rollout. Most merchants aren't manually implementing UCP. Their platform is doing it for them, and the platforms shipped the update fast.

Platform-by-Platform Reality

UCP Transport Comparison — REST vs MCP vs Embedded by platform

The theoretical guides will tell you that UCP readiness is about your structured data and feed configuration. In practice, it's mostly about which platform you're on. Here's what we've seen across the major players.

Shopify: The Default Winner

Shopify accounts for roughly 74% of identified platforms in our dataset (898 of the platform-identified merchants). This dominance isn't because Shopify merchants are more proactive about UCP — it's because Shopify rolled out UCP support at the platform level, giving every store baseline compliance automatically.

Out of the box, a Shopify store gets functional product discovery, cart, and checkout endpoints. The Schema.org markup is handled. The Merchant Center feed attributes are populated. For the average merchant, getting UCP ready on Shopify means verifying that your product data is clean rather than building anything from scratch.

The downside: Shopify's one-size-fits-all approach means limited customization of UCP behavior. If you need to implement conversational commerce attributes like substitution logic or compatibility data, you're working within Shopify's constraints. But for baseline agentic commerce readiness, nothing else comes close to the out-of-the-box experience.

WooCommerce: Flexible but Inconsistent

WooCommerce stores show the widest variance in UCP readiness. The open-source model means implementation quality depends entirely on which plugins a merchant has installed and how they've configured their stack. We've seen WooCommerce stores with excellent structured data and smooth agent interactions right next to stores where basic product attributes are missing or malformed.

The flexibility is a genuine advantage for merchants who want to implement advanced ACO features — conversational attributes, detailed return policies, rich product relationships. But the inconsistency is a problem for agents, which need predictable data structures to operate reliably. If you're on WooCommerce and serious about agentic commerce optimization, an audit of your specific UCP endpoint output is essential, not optional. Run your store through UCP Checker and see what an agent actually encounters.

BigCommerce: Strong APIs, Broken Images

BigCommerce has a genuine technical advantage in its API architecture. The platform's API-first design translates well to UCP's endpoint model, and the stores we've tracked generally produce clean, well-structured UCP responses.

But there's a specific, persistent issue: BigCommerce's S3-hosted image URLs break agent image parsing. This is a real failure mode we've observed in Playground sessions. When an agent can't parse product images, it loses a significant input signal for product matching and variant selection. For a platform that otherwise has strong UCP fundamentals, this is an unfortunate gap — and one that BigCommerce merchants should pressure their platform to fix. For now, it's worth investigating whether your image delivery pipeline produces URLs that agents can reliably consume. Our BigCommerce guide walks through the specifics.

Magento (Adobe Commerce): Enterprise Muscle, Enterprise Complexity

Magento implementations tend to be enterprise-grade, which means the UCP output is thorough but the setup complexity is high. These stores generally have rich product data, detailed catalog structures, and the kind of attribute depth that agents love. But the implementation burden falls more heavily on the merchant's development team compared to Shopify or BigCommerce, where the platform handles the heavy lifting.

If you're on Magento and aren't UCP ready yet, expect a meaningful engineering investment. If you have started, you're probably in good shape — the platform's data model maps well to what UCP expects, especially for multi-variant products and complex catalog hierarchies. See our Magento guide for implementation specifics.

What Agents Actually Do (vs. What Guides Tell You to Optimize For)

Agent Shopping Flow — MCP tool call sequence

Here's where our data diverges most sharply from the advisory content circulating about UCP preparation.

Agents Skip the Cart

The conventional model of ecommerce — browse, add to cart, review cart, checkout — doesn't describe how AI agents behave. In our Playground data, we've recorded 395 checkout operations versus just 104 cart operations. Agents are going direct to checkout nearly four times more often than they're using the cart.

This has major implications for agentic commerce optimization. If you've invested heavily in cart-level features — upsells, cross-sells, minimum order messaging, cart-based promotions — agents are likely bypassing all of it. The checkout endpoint is where the action happens. Your optimization effort should weight accordingly — compare your store against competitors to see where you stand: make sure checkout handles single-product and multi-product flows cleanly, with clear variant specification and unambiguous pricing.

Variant Mismatches Are the Top Failure Mode

Cart variant mismatches remain the most common reason agent sessions fail to complete a purchase. An agent selects a product, identifies the desired variant (size, color, configuration), and submits a cart or checkout request with a variant ID that doesn't match what the endpoint expects. The session stalls or errors out.

This isn't an agent intelligence problem — it's a data clarity problem. Stores with clean, unambiguous variant structures and consistent ID schemes see dramatically higher agent completion rates. Stores with complex variant matrices, inconsistent naming, or variant IDs that change between API responses create confusion that even the best models struggle to resolve.

If you do one thing for ACO today: audit your variant data. Make sure every variant has a stable identifier, a clear human-readable name, and consistent representation across your discovery and checkout endpoints.

Token Consumption Tells You Where Agents Struggle

We've consumed 43 million tokens over 1,000 Playground sessions. The per-session cost varies dramatically based on store complexity and model choice, but a telling pattern emerges in checkout flows: completing a purchase takes approximately 55,000 tokens with the best-performing models.

That number is a proxy for friction. A 55K-token checkout means the agent is making multiple round-trips, parsing product data, resolving variants, handling errors, and re-trying. Stores that produce clean, predictable UCP responses see lower token counts — which directly translates to faster agent interactions and lower cost for the platforms running these agents at scale.

Model Performance Varies Significantly

Not all AI models handle UCP interactions equally. Claude Sonnet 4.5 leads our Playground leaderboard with 205 sessions, and the checkout completion rate across all sessions sits at 41%. That might sound low, but consider what it represents: four out of ten fully autonomous purchase attempts succeed end-to-end, without any human intervention, across a diverse set of merchants with varying UCP implementation quality.

The model performance gap matters for merchants because it signals where your UCP implementation has rough edges. If top-tier models struggle with your checkout flow, every agent will struggle. Testing your store in UCP Playground with multiple models gives you a direct read on where your implementation creates unnecessary friction.

The Capabilities Gap That Will Define Winners

Go back to those adoption numbers: identity at 3 merchants, payment at 0. These aren't just gaps — they're the entire frontier of competitive differentiation in agentic commerce.

Right now, every UCP checkout ends with a handoff. The agent gets the customer to the point of purchase, then drops them into a traditional checkout flow to enter their identity and payment information. That handoff is where conversion dies. Every redirect, every form field, every authentication step is a chance for the customer to abandon.

The merchants who figure out identity and payment first — who let an agent complete a purchase end-to-end without a handoff — will have a structural conversion advantage that no amount of Schema.org optimization can match. This is where UCP's roadmap points: loyalty integration, post-purchase management, multi-vertical capabilities. But the foundation is identity and payment.

We don't yet know what the winning implementation pattern looks like for these capabilities. The spec supports them, but the ecosystem hasn't built them. This is the space to watch, and the space where early investment will pay disproportionate returns.

An Optimization Checklist Grounded in Data

Most ACO checklists are derived from the spec. This one is derived from watching >1,000 agent sessions succeed and fail across 4,024 merchants. Here's what actually moves the needle, ranked by observed impact:

1. Fix your variant data first. Stable IDs, clear names, consistent representation across endpoints. This is the single highest-impact fix based on our failure-mode analysis.

2. Optimize for direct-to-checkout flows. Agents skip the cart. Make sure your checkout endpoint handles product selection, variant specification, and pricing in a single clean interaction.

3. Audit your product images. If you're on BigCommerce or any platform using CDN-hosted images with complex URL structures, verify that agents can parse your image URLs. Broken image parsing degrades product matching accuracy.

4. Migrate to the latest spec version immediately. The v2026-04-08 migration happened in four days across the ecosystem. If you're still on an older version, you're already behind 99.3% of verified merchants.

5. Test with actual agents, not just validators. Schema validation tells you if your markup is syntactically correct. It tells you nothing about whether an agent can actually complete a purchase. Run your store through UCPPlayground and watch what happens.

6. Validate your full UCP endpoint output. Use UCPChecker to see exactly what your store exposes to agents — capabilities, product data, structured attributes — and where the gaps are.

7. Clean up your Merchant Center feed. Return policies, product identifiers, and the native commerce attributes that feed into UCP discovery. This is table-stakes, but our data confirms that stores with complete feed data see higher agent engagement in discovery flows.

8. Start thinking about identity and payment. You won't implement these today — almost nobody has. But understanding the spec's identity and payment capabilities now positions you — our April ecosystem report tracks adoption monthly to move fast when the ecosystem catches up. The jump from 0 to first-mover will be worth more than incremental improvements to discovery or checkout.

9. Monitor your platform's UCP updates. If you're on Shopify, WooCommerce, BigCommerce, or Magento, your platform is doing most of the UCP work. Stay current with their releases — set up domain alerts to get notified when your store's status changes. Platform-level updates drove 99.3% spec migration in four days — the single most effective "optimization" most merchants can do is simply keeping their platform current.

10. Get listed in the UCP directory. UCPRegistry is the open UCP index where agents discover merchants. Your listing is what agents see when deciding which merchants to route a customer to. Make sure you're listed, your data is accurate, and your capabilities are competitive with peers in your vertical.

The Bottom Line

Agentic commerce optimization isn't a theoretical exercise anymore. UCP ecommerce is live, it's measurable, and it's growing fast. Our UCP index tracks 4,024 verified merchants serving UCP endpoints today. AI agents are completing purchases 41% of the time. The gap between being UCP ready and being UCP optimized is measurable in variant data quality, checkout flow design, and capabilities adoption.

The merchants who treat ACO as a data problem — not just a markup problem — are the ones who'll convert when agents come shopping. And agents are already shopping. We've got 43 million tokens of proof.


Check if your store is UCP ready at UCPChecker.com. Browse the UCP directory at UCPRegistry. Test agent interactions in UCPPlayground. Platform-specific implementation guides: Shopify · WooCommerce · BigCommerce · Magento.

Top comments (7)

Collapse
 
toshihiro_shishido profile image
toshihiro shishido

@benjifisher, the 0.07% identity-ready and 0% payment-ready numbers across 4,003 merchants are the kind of stats that should reframe every "we're AI-ready" deck I've seen this year — checkout-ready ≠ agent-pickable. A few angles I'd add for anyone trying to build the bridge from UCP coverage to agent-driven revenue:

  1. The 41% checkout completion rate is the unit-economics ceiling, not the floor. With agents averaging 55K tokens per checkout and round-trips inflating that further, the cost-per-completed-purchase on agent traffic is non-trivial. For low-margin SKUs that's a breakeven ROAS conversation by another name — your tier-1 cost per agent purchase has to clear 1 ÷ gross margin × 100, except now the "ad spend" denominator is platform infrastructure cost rather than ad budget.

  2. Variant quality scoring belongs in the bid feed, not just the catalog. If platform-level implementation is 75% of outcome quality, merchants on weaker platforms can't fix it — but they can deprioritize variants with low completion likelihood from agent surfaces. Effectively a "POAS for agents": down-weight low-margin SKUs with messy variants, let agents converge on high-completion-probability + high-margin pockets.

  3. @jasper_wang's selection-fragmentation point is the harder downstream problem. "How do agents decide who to choose" looks a lot like ranked retrieval — embedding quality of product descriptions, mention frequency in trusted corpora, and prior agent success rate per merchant become the new SEO. Worth thinking about agent-discovery quality as a separate measurable metric from UCP technical readiness.

For the merchants in your dataset, did you see any correlation between UCP readiness scores and checkout completion rates at the merchant level — or was completion rate dominated by platform/category effects independent of merchant implementation quality? That split would tell teams whether to invest in their own UCP work or push their platform to fix it.

A genuinely sobering read — appreciate you publishing the 0.07% number rather than rounding it to "early days." The empirical floor is more useful than another aspirational forecast.

Collapse
 
benjifisher profile image
Benji Fisher

@toshihiro_shishido Really sharp framing on the unit economics. You're right that 41% is a ceiling — and the cost equation has a model-selection dimension most merchants won't know exists. We see significant token variance across models for the same stores, which means the agent platform's model choice directly affects the economics of completing a purchase. That's a cost lever merchants can't even see, let alone control. We track model performance across our agent sessions on our model profiles page.

On your direct question — right now, 3-4 months into the protocol, platform effects dominate. Shopify's platform-managed UCP with automatic updates produces consistently faster response times and cleaner data than self-managed implementations on WooCommerce or BigCommerce. A well-configured WooCommerce store can match Shopify on data quality, but the infrastructure layer (response times, caching, endpoint lifecycle management) is harder for individual merchants to fix. Rough split today: platform choice is ~75% of the outcome, merchant-level implementation quality is the remaining 25%. That ratio will shift as platforms mature and merchant-level optimization becomes the differentiator. We break down what each platform delivers out of the box on our platforms page.

On variant quality as a bid signal — that's where we think this goes next. We're building a UCP score surface that factors in completion-probability signals per merchant, not just spec compliance. The stores with clean variant data and fast responses will naturally rank higher when agents start choosing between merchants. Agent-SEO, as you put it.

The 0.07% identity number is the one I keep coming back to. Checkout is table stakes now — 99.5% have it. The competitive differentiation will be in identity and payment, and right now almost nobody's there. We documented the first autonomous AI agent purchase using identity linking, and the newly expanded Tech Council — now including Amazon, Meta, Microsoft, Salesforce, and Stripe — has identity linking as a top priority. First movers on identity and payment will have a structural conversion advantage that no amount of catalog optimization can match.

Collapse
 
toshihiro_shishido profile image
toshihiro shishido

The 75/25 shift is the part most threads miss. Once platforms converge on "good enough," the variance moves to the merchant layer — and merchants not baselining now will have nothing to compare against when that flip lands. Identity as the next moat tracks too — checkout became infra, identity stays competitive.

Collapse
 
jasper_wang profile image
jasper wang

This is a great breakdown — especially the “POAS for agents” framing.

It feels like we’re converging on something important here:

Right now we have multiple partial signals:

– UCP readiness (can the agent transact)
– completion rate (does it actually succeed)
– cost per checkout (is it economically viable)
– mention / embedding signals (is it even discoverable)

But they’re all disconnected.

What’s missing is a unified decision layer where agents (or platforms) can:

→ compare merchants across these dimensions
→ normalize them into a comparable score
→ and optimize for a target (conversion, margin, reliability)

Without that, selection is still effectively binary:

→ “works” vs “doesn’t work”

Which explains why so much of the current behavior looks like availability-first routing.

The interesting question is:

do you see this decision layer emerging at the platform level (e.g. UCP Registry evolving into ranking + routing),

or at the model layer (where agents themselves learn to weight these signals)?

That split probably determines where most of the value accrues.

Collapse
 
jasper_wang profile image
jasper wang

this is a really interesting shift

if agents become the buyers, the question isn’t just “are you UCP-ready”

it’s:

how do agents decide who to choose?

feels like there’s a missing layer around “decision signals”

today that’s fragmented across:

– user conversations
– product mentions
– real usage signals

curious if you’ve seen any patterns in what agents actually rely on when ranking merchants

Collapse
 
benjifisher profile image
Benji Fisher

Good question. From over 1,000 agent sessions in UCP Playground, the pattern is clear: agents prioritize product discovery and checkout above everything else. They search the catalog, validate variants and pricing, then go straight to purchase.

What agents almost never do: check store policies. Return policies, trust signals, brand reputation — basically invisible to agents right now. They're optimizing for "can I find the product and complete the transaction," full stop.

That's exactly the gap you're pointing at. Today an agent picks a merchant based on whether the UCP endpoint responds and has the product. There's no quality ranking, no success-rate weighting, no way to distinguish a merchant that completes 80% of agent purchases from one that completes 20%. We're building toward that at UCP Registry — not just a directory of who supports UCP, but a quality signal layer for agent routing.

If you want to see how two merchants compare side by side on implementation quality right now, we just shipped that: UCP Checker side-by-side compare. It's a start toward making those decision signals

Collapse
 
jasper_wang profile image
jasper wang

This is super interesting — especially the part about agents ignoring policies and trust signals.

It almost feels like we’re in a “pre-ranking” phase right now:

→ if the endpoint works, you get selected
→ if not, you don’t exist

But once that layer stabilizes, it seems inevitable that agents will need some kind of selection signal beyond availability.

What’s interesting is that those signals might not come from inside the system alone.

A lot of “preference” today already lives outside:

– what users are discussing
– which products get mentioned in context
– which solutions show up repeatedly in problem-driven queries

Feels like there’s a second layer forming:

→ infra signals (what you’re building)
→ external signals (what people / systems keep referencing)

Curious if you’ve seen any early hints of agents starting to incorporate those external signals yet, or if everything is still strictly endpoint-driven for now?