# When the Buyer Is a Bot > Agentic commerce, the protocol stack running underneath it, and the infrastructure work nobody wants to budget for. When the thing evaluating your product reads JSON and times out at 200ms, the storefront was never the point. *Source: https://rawmktg.com/blogs/when-the-buyer-is-a-bot · rawmktg. by Vinayak Ravi* For thirty years the job was to make a human want something. Now the thing evaluating your product reads JSON, ignores your hero image, times out at 200 milliseconds, and never sees a single pixel you paid a designer to make. This is what changes, why it changes, and what you actually have to build. ## 01. What changes when the buyer is a bot? **Everything the last thirty years optimised, the search-and-scroll funnel, gets deleted, not shortened. Awareness, consideration, cart and checkout collapse into a single programmatic execution against your API.** Online retail ran on one model since the web began: a person types into a box, scans a page, clicks a category tree, gets nudged by a banner, and grinds through a checkout form. Every layer of the modern stack, visual merchandising, trust badges, cart-abandonment emails, exists to serve that sequence. Agentic AI does not improve it, it deletes it. Ask a user states an outcome → Interpret agent weighs trade-offs → Query backends in parallel, under 200ms → Execute payment tokenised, no form → Done no storefront rendered Figure 1, the old funnel gave you six chances to influence a decision. The new one gives you one, and it is your API response. Your homepage is not the product surface anymore. Your catalog endpoint is. When the buyer is software, the graphical storefront is bypassed entirely. What replaces it is unglamorous: API machine-readability, structured catalog schemas, sub-second response times, and protocol compliance, the same shift that made [winning Google stop being winning AI](/blogs/why-traditional-seo-is-no-longer-enough), taken to its logical end. ## 02. How big is agentic commerce, and who stands to lose? **Between $3 and $5 trillion of transaction value moves onto agent rails by 2030, and the marketplaces that monetise human visual attention face structural disintermediation.** US agent-mediated retail alone is projected at $900 billion to $1 trillion a year against roughly $6.3 trillion of legacy web and mobile commerce. This is not a niche channel forming at the edges; it is a reallocation of gross merchandise value onto a different set of rails, and it is moving faster than web or mobile did because agents need no new physical infrastructure. Figure 2, agentic commerce is not additive demand. It is existing demand rerouted through a machine intermediary. Source: McKinsey, industry projections. Traditional e-commerce concentrated demand into destination platforms (Amazon, Expedia) that aggregated intent and charged for access. Agentic commerce runs horizontally: a personal AI concierge resolves intent at the point of origin and talks to several merchant backends at once. The aggregator loses its position as the front door, and any channel whose model depends on monetising human visual attention faces disintermediation, because the buyer no longer has eyes. What actually changes between the two eras | Dimension | Traditional e-commerce | Agentic commerce (2025-30) | What it means for you | | --- | --- | --- | --- | | Orchestrated value | ~$6.3T on web and mobile | $3.0T-$5.0T on agent rails | GMV reallocates fast; early protocol adopters catch it first | | Primary interface | Graphical UI in a browser | Conversational AI + headless endpoints | Visual UX budget shifts to machine-readable API structure | | Who you sell to | A human, visually and emotionally | An autonomous agent, logically and on data | Persuasion loses to verifiable factual proof | | Navigation model | Vertical silos (Amazon, Expedia) | Horizontal agent ecosystems | Destination portals get disintermediated | | Main conversion killer | Cart abandonment and UX friction | Data fragmentation and API latency over 200ms | Infrastructure performance decides if you are even considered | ## 03. How does a bot actually meet your store? **Three ways, and they impose different engineering requirements: Agent-to-Site, Agent-to-Agent, and Brokered Agent-to-Site. Most teams build for one and get blindsided by the other two.** Get the topology straight before touching any protocol. Three interaction shapes, three requirements | Topology | What it is | What it demands of you | | --- | --- | --- | | Agent-to-Site (A2S) | A consumer agent hits your web store or public APIs directly, parsing markup or driving your flow. | Parseable structured markup and fast headless endpoints. Your site is read, not viewed. | | Agent-to-Agent (A2A) | The buyer's agent negotiates natively with your inventory/sales agent in a standard message format. | A negotiating counterpart, a selling agent of your own. No human on either side. | | Brokered A2S (BA2S) | A brokerage validates identity, normalises payloads, and aggregates feeds before passing execution to you. | Accepting a middleman between you and your buyer, and the risk of losing the customer entirely. | ## 04. What's in the agentic-commerce protocol stack? **Five complementary open protocols: UCP for the shopping lifecycle, ACP for in-chat checkout, AP2 for delegated payment authority, and MCP plus A2A for context and inter-agent messaging.** They are complementary, not competitive, and you will likely implement several. They divide up cleanly by layer. ### UCP: the discovery and lifecycle layer Co-developed by Google and Shopify (NRF, January 2026), the Universal Commerce Protocol is an open, end-to-end standard layered like TCP/IP: a Shopping Service layer of transaction primitives, a Capabilities layer (Catalog, Cart, Checkout, Identity) versioned independently, and an Extensions layer for domain schemas. Merchants publish what they support at a fixed URI, and if it does not exist, an agent has no idea what you can do and falls back to guessing: Discovery, the /.well-known/ucp capability profile ``` GET https://yourstore.com/.well-known/ucp { "ucp_version": "1.0", "merchant": { "id": "urn:merchant:yourstore", "name": "Your Store", "merchant_of_record": true }, "capabilities": [ { "name": "com.yourstore.catalog", "version": "2.1" }, { "name": "com.yourstore.cart", "version": "1.4" }, { "name": "com.yourstore.checkout", "version": "2.0" }, { "name": "com.yourstore.identity", "version": "1.0", "auth": { "type": "oauth2", "scopes": ["profile", "loyalty"] } } ], "extensions": [ { "name": "com.yourstore.fulfilment.split", "version": "1.0" } ], "endpoints": { "catalog": "https://api.yourstore.com/ucp/v2/catalog" }, "sla": { "p95_response_ms": 140 } } ``` When an agent initiates, your system computes the mathematical intersection of the two capability profiles; whatever both sides support becomes the operating envelope, everything else is silently dropped. That decentralised negotiation is what lets both sides upgrade on their own schedule. UCP also ships an explicit state machine, incomplete, then ready\_for\_complete, then requires\_escalation when a risk score trips, with a human escape hatch rendered inside the agent via the Embedded Checkout Protocol. Build the escalation path first, because it is where your legal exposure concentrates. ### ACP: in-chat checkout Launched September 2025 by OpenAI and Stripe (Apache 2.0), the Agentic Commerce Protocol specialises in conversational, human-in-the-loop checkout inside workspaces such as ChatGPT. Raw card credentials never reach the model; a scoped token does. The commercial term to internalise: participation costs 4% of the completed order. ACP, creating a checkout session (note the loyalty\_discount) ``` POST /agentic_checkout/sessions HTTP/1.1 Authorization: Bearer Content-Type: application/json { "items": [ { "sku": "TRK-42-BLK-M", "quantity": 1 } ], "buyer": { "identity_token": "eyJhbGciOi...", // OAuth-linked account, keeps you the MoR "shipping_address": { "postal_code": "560001", "country": "IN" } }, "payment": { "delegate": "spt_1QX7mF...", // shared payment token, scoped "scope": { "merchant": "yourstore", "currency": "INR", "amount_max": 899000 } } } --- 200 OK --- { "session_id": "acs_9f2b...", "status": "ready_for_complete", "totals": { "subtotal": 799000, "tax": 143820, "shipping": 0, "loyalty_discount": -79900, "currency": "INR" } } ``` The one field that decides your margin Notice the loyalty\_discount in that response, it only exists because the buyer identity token was present. That single field is the difference between competing on your real offer and competing on list price. Identity linking is not a nice-to-have; it is the mechanism that keeps you from racing to the bottom. ### AP2: delegated payment authority Introduced by Google Cloud (September 2025) with 60+ financial institutions including Mastercard, Visa, PayPal and Adyen, the Agent Payments Protocol restores the trust that card networks assumed from a present, approving human. It uses two cryptographically signed, non-repudiable Mandates: an Intent Mandate encoding the guardrails (max spend, categories, time window), and a Cart Mandate locking specific items and price. Together they form an unalterable audit trail: An Intent Mandate, signed as a Verifiable Credential ``` { "type": ["VerifiableCredential", "IntentMandate"], "issuer": "did:web:wallet.example.com", "credentialSubject": { "principal": "did:key:z6MkhaXg...", // the human, cryptographically "agent": "did:key:z6MkjR9pQ...", // the delegate "constraints": { "max_total_minor_units": 1200000, "currency": "INR", "categories": ["travel.flight"], "merchant_allowlist": ["*.iata-verified"], "valid_until": "2026-08-15T00:00:00Z", "requires_human_signature_above": 900000 } }, "proof": { "type": "Ed25519Signature2020", "jws": "eyJhbGciOiJFZERTQSJ9.." } } ``` Underneath, two more protocols hold the rest up: the [Model Context Protocol (MCP)](/blogs/does-llms-txt-do-anything-yet), donated by Anthropic to the Linux Foundation, standardises how models access live catalog data and invoke backend tools; and the Agent-to-Agent (A2A) protocol lets agents from different vendors discover each other and negotiate over JSON-RPC. Which do you need? If buyers reach you through chat, ACP puts you in the transaction; if you sell across agent platforms, UCP is the broader surface; AP2 is not optional either way, because it is what lets an issuer approve an automated payment without treating it as fraud. The protocol stack at a glance | Protocol | Lead ecosystem | Layer | Cost | | --- | --- | --- | --- | | UCP | Google, Shopify, Etsy, Target | Full shopping lifecycle | Open standard, platform-free | | ACP | OpenAI, Stripe, Etsy, Shopify | In-chat conversational checkout | Apache 2.0, 4% fee on orders | | AP2 | Google Cloud, Visa, Mastercard, PayPal | Trust, security, payment rails | Open standard, normal processing fees | | MCP | Anthropic, Linux Foundation | System data access + memory | Open standard | | A2A | Open community, cross-industry | Multi-agent coordination | Open standard | ## 05. How do money and identity work when the buyer has no hands? **Through multi-party token delegation and OAuth identity linking, and the failure mode nobody plans for is the anonymity gap that quietly strips you of the customer relationship.** Letting an agent pay without exposing a raw card number needs a tokenisation model that did not previously exist. Stripe's Shared Payment Token is scoped to one merchant, one currency, one amount, so the blast radius of a compromise is a single transaction; Visa and Mastercard's Agentic Network Tokens carry the delegated-authorisation proof and a risk score so issuers can approve good automated transactions instead of blanket-declining anything robotic. The anonymity gap, and why it should worry your CRM team An agent buys from you, the purchase completes, and you have no idea who bought it. Agent-mediated checkout degenerates into anonymous transactions that strip you of customer data and block the buyer from benefits they already earned, and the agent platform becomes the only party that knows anything. The fix is Identity Linking (OAuth 2.0, built into both UCP and ACP): the consumer links their store account to their concierge once, and from then on the agent presents a cryptographic identity token so you can compute tier pricing, apply retention offers, and keep the relationship, the same [entity and account resolution problem](/blogs/becoming-an-entity) in a payments context. A newer category, Know Your Agent (e.g. Skyfire's KYAPay), assigns verified identities to the agents themselves. ## 06. What's the infrastructure bill nobody budgets for? **A MACH architecture, a machine-readable catalog, a sub-200ms latency SLA, and one unified promotions endpoint. Legacy monolithic stacks simply cannot participate.** This is the section skipped in strategy decks that then kills the project in month four. There is no prompt-engineering workaround for a slow database. Figure 3, mature MACH adopters (microservices, API-first, cloud-native, headless) hit a 77% successful AI-deployment rate; legacy monoliths, 36%. The gap is not AI talent, it is whether the system can be called programmatically at speed. AI reasoning engines evaluate products through structured data, not marketing copy. If your feed has unstructured text blobs or inconsistent variant identifiers, agents skip your listing and pick a competitor whose data parses cleanly, there is no appeal process. Three commitments are non-negotiable: explicit granular attributes as distinct schema fields, real-time inventory and pricing sync (a stale price that fails at checkout gets your domain down-ranked), and a strict sub-200ms latency SLA, because discovery engines query competing backends in parallel and time out anything slower. This is the machine-readable, [chunk-extractable structure](/blogs/how-your-page-gets-retrieved) retrieval has always rewarded, now enforced at the API. JSON-LD, a Product and Offer entity doing its job ``` ``` And centralise your incentives. Most enterprises fragment promotions across CRM, POS, the commerce platform and a loyalty system; humans hunt for a coupon, agents do not. If they cannot find one API endpoint that evaluates every eligible incentive in real time, they default to your base list price and compare you on that number. Competing purely on price in an agentic environment is a margin-eroding race to the bottom, so expose value the price field cannot capture (priority support, extended returns, VIP access) as structured signals agents fold into their evaluation. ## 07. How do you rank when the reader is a retrieval system? **By optimising structured content for extraction and citation, not clicks: high information density, schema as entity resolution, and a refresh cadence, because AI citations decay on a roughly 13-week half-life.** A RAG pipeline grades four things: semantic relevance (vector distance), information density (verifiable facts per token, padding hurts), citation authority and E-E-A-T, and structural extractability. If a fact is only legible because of where it sits on the page, it is invisible, write self-contained blocks that answer the question in the first sentence, the [anatomy of any high-citation page](/blogs/anatomy-of-a-high-citation-page). Schema in JSON-LD is the primary data language of GEO, but the purpose has shifted from decorating a results page to resolving entity ambiguity when a model synthesises an answer from a dozen disagreeing sources, your schema is how the model knows which "Apex" you are. FAQPage blocks in particular get extracted almost verbatim, the same [structured-data playbook that earns AI citations](/blogs/schema-markup-ai-citations-2026). Figure 4, dark-web discussion of AI agent tooling surged over 450% in six months, and malicious bot-initiated transactions rose 25% globally and 40% in the US. The offensive side industrialised before most merchants finished their first pilot. Source: Visa PERC. Citation decay: your best page has a shelf life Roughly 50% of content cited in AI answers is less than 13 weeks old. A page that dominated in March can be invisible by July without anything about it changing. A content library is not an asset that appreciates, it depreciates on a ~13-week schedule, and the maintenance budget has to be real, which is exactly what [the 30-day content half-life](/blogs/30-day-content-half-life-recency-ai-ranking-signal) measures and the [content recency decay estimator](/tools/content-recency-decay) quantifies. Three decay vectors, three different fixes | Decay type | What is happening | The countermeasure | | --- | --- | --- | | Statistical | Prices, stock counts and benchmarks age out; fresher competitors displace you. | Automated quarterly data refreshes + real-time dateModified in schema. | | Structural | The platform changes its extraction preference (prose to bulleted lists) and down-ranks your format overnight. | Modular content: concise definitions, bulleted specs, and data tables in one doc. | | Competitive | A rival publishes higher-density, more authoritative coverage and outranks you on merit. | Monitor citation share; enrich with original research and updated specs. | ## 08. What breaks: liability, identity, and the attack surface? **The identity gap (which human authorised this?), the legal fact that an AI agent is not a legal agent, and a threat surface that industrialised before merchants finished their pilots.** Delegate purchasing authority without a verifiable human root credential and you create legal ambiguity and a systemic attack surface at once, exposed to synthetic-identity probing, restriction bypass, and repudiation cascades where a buyer disputes a valid charge by claiming the bot exceeded instructions. The most expensive misconception hides inside the word "agent". Legal agency requires a consensual relationship between two legal persons with enforceable fiduciary duties; software has none of that. In the US, the Uniform Electronic Transactions Act (49 states) binds a person or corporation to contracts their automated system executes, even on an erroneous outcome, provided it operated within its deployed scope. So the scope you deploy an agent with is not a product decision, it is a liability boundary, and your Terms of Service needs an automated-agent clause now. The threat landscape moved first. Three vectors define the risk, and the first is genuinely novel: agent-targeted prompt injection, where attackers plant instructions inside product descriptions, reviews or HTML metadata that override the shopping agent's logic when it reads the page, your user-generated content is now an executable surface, the retail edition of [hallucination-proofing your brand](/blogs/hallucination-proofing-your-brand). What a poisoned review looks like in your database ``` -- Customer review, rendered on your public product page -- and read verbatim by any agent parsing the DOM "Great shoe, held up well on wet rock. Would buy again." -- Mitigations, in order of effectiveness: -- 1. Strip HTML comments + control sequences at ingestion, not render -- 2. Serve agent-facing content from structured fields only, never raw UGC -- 3. WAF rules matching instruction-shaped patterns in submitted text ``` The other two: dark-agent SEO and synthetic merchants (storefronts engineered with immaculate schema and far-below-market prices to harvest payment tokens, so the better your competitor's data hygiene, the more suspicious an unusually good offer should look), and bot-to-bot collusion (delegated pricing and ordering settling into price-fixing loops no human agreed to). Add concentration risk: ~90% of autonomous coding agents default to Stripe, and defaults in agentic systems are near-total market allocation, not mild preferences. ## 09. What's the roadmap, in order? **Four phases run strictly in sequence: foundation, protocol integration, GEO and incentive alignment, then risk hardening. Rushing protocol adoption before the data and API foundation exists produces failure, exposure and margin erosion, in that order.** Sequence matters more than speed, because each phase depends on the one before it actually working. 1. Foundation sub-200ms APIs + readable catalog → 2. Protocols publish /.well-known/ucp, ACP, OAuth → 3. GEO + incentives citation share, unified promo engine → 4. Risk hardening injection filters, KYA, monitoring Figure 5, the order is not arbitrary. Foundation removes timeout rejections; protocols remove channel isolation; GEO removes commoditisation; hardening removes injection and legal exposure. ## 10. What do you actually do on Monday? **Measure p95 catalog latency, audit one page's structured data, count where promotions are decided, add a ToS agent clause, sanitise UGC at ingestion, and set a refresh cadence, none of it waits for a standard to finalise.** Strip away the acronyms and the work reduces to a short list, all of which has value even if agentic commerce underdelivers on its projections. The Monday checklist | Do this | Why it matters | | --- | --- | | Measure your p95 catalog API latency (not average) | Above 200ms is your first quarter of engineering work, agents time you out. | | Audit one product page's JSON-LD | Could an agent answer three buying questions from the data alone? If not, your PIM is the bottleneck. | | Count every system that decides a promotion | More than one, and agents compare you on list price; every discount you fund is invisible. | | Add an automated-agent clause to your ToS | Cheap, fast, and the legal position on third-party bot access is unsettled, plant your stake. | | Sanitise user-generated content at ingestion | Strip HTML comments and control sequences before storage, reviews are an executable surface. | | Set a refresh cadence on high-value content | A 13-week half-life means an unmaintained library is a depreciating one; update dateModified. | | Check what your platform already ships | Shopify, commercetools and others abstract much of this, building it yourself is the common expensive mistake. | Free Tool · Diagnostic Score your agentic-commerce readiness Run the Monday checklist as a scorecard, latency, schema, incentives, identity and risk controls, and see your biggest gaps ranked. Agentic readiness 0/100 Answer the checks At riskEmergingAgent-ready Biggest gaps [Open the full tool →](/tools/agentic-commerce-readiness-scorecard) Free tools from this piece Four browser-based tools built from this teardown: the [Agentic Commerce Readiness Scorecard](/tools/agentic-commerce-readiness-scorecard), the [Product Schema Auditor](/tools/product-schema-auditor) to check whether an agent can read a product page, the [UGC Prompt-Injection Scanner](/tools/ugc-prompt-injection-scanner) to catch poisoned reviews, and the [Product/Offer JSON-LD Generator](/tools/product-offer-jsonld-generator). All free, all run in your browser. ## 11. What's the takeaway? **Agentic commerce rewards exactly the work that has always been easy to defer: clean data models, fast APIs, honest specifications, consistent identifiers. The arbitrage of hiding a weak backend behind a great front end is closing.** For thirty years a good-enough backend could hide behind a great front end. That arbitrage is closing, because the bot does not care how the page looks. The bot only cares whether the answer is there, whether it is true, and whether it arrived in time. Which is, if you think about it, the market finally rewarding the thing good operators wanted to build all along. Frequently Asked Questions ### What is agentic commerce? Agentic commerce is online buying carried out by autonomous or semi-autonomous AI agents rather than humans clicking through a storefront. A user states an outcome ("cheapest direct flight under a budget"), and the agent interprets intent, queries live inventory and pricing across several merchant backends in parallel, executes a tokenised payment, and tracks fulfilment, often without consulting the human after the first sentence. Projections put agent-orchestrated transaction value at $3-5 trillion globally by 2030. The graphical storefront is bypassed; your catalog API becomes the product surface. ### Which agentic-commerce protocols do I actually need? It depends on how buyers reach you, but they converge on the same requirements. If buyers transact through chat interfaces like ChatGPT, implement ACP (OpenAI/Stripe, 4% fee). If you sell across multiple agent platforms and care about discovery and post-purchase, UCP (Google/Shopify) is the broader surface. AP2 (Google Cloud, Visa, Mastercard) is effectively required either way, since it is the trust layer that lets an issuer approve an automated payment. MCP matters for live catalog querying and A2A once you run a selling agent. Underneath all of them: fast structured APIs, provable authorisation, and accurate real-time data. ### Why does the 200ms latency wall matter so much? Because agents query competing merchant backends in parallel and time out anything that fails to return a structured payload within roughly 200 milliseconds, excluding it from the evaluation set entirely. Unlike a human waiting on a page, the agent is running a race against five other merchants answered simultaneously. Measure your p95 (not average) catalog API latency; if it is above 200ms, that is your first block of engineering work, and no protocol adoption will compensate for it. ### Is an AI shopping agent a legal agent? No. Legal agency requires a consensual relationship between two legal persons with enforceable fiduciary duties; software models have no legal personhood and cannot owe a duty to anyone, legal scholars compare them to trained animals or industrial machinery. In the US, the Uniform Electronic Transactions Act (adopted in 49 states) binds a person or corporation to contracts their automated system executes within its deployed scope, even on an erroneous outcome. Practically: the scope you give an agent is a liability boundary, and your Terms of Service needs an explicit automated-agent clause. References The specifications, research and legal analysis this teardown draws on. 1. [Agentic Commerce, UCP, MCP, and the Product Data Layer AI Agents Need. Crystallize.](https://crystallize.com/blog/agentic-commerce) 2. [What Is the Agentic Commerce Protocol? Future of Online Shopping. Acodez.](https://acodez.in/agentic-commerce-protocol/) 3. [Agentic commerce: How agents are ushering in a new era. McKinsey.](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity-how-ai-agents-are-ushering-in-a-new-era-for-consumers-and-merchants) 4. [What Is Agentic Commerce? The 2026 Guide. Fin.ai.](https://fin.ai/learn/what-is-agentic-commerce) 5. [Agentic Commerce Protocol. OpenAI.](https://developers.openai.com/commerce) 6. [Supporting additional payment methods for agentic commerce. Stripe.](https://stripe.com/blog/supporting-additional-payment-methods-for-agentic-commerce) 7. [A guide to agentic commerce: how AI shopping agents are reshaping brand loyalty. Talon.One.](https://www.talon.one/blog/agentic-commerce) 8. [Agentic Commerce: The Case For Foundational Readiness. commercetools.](https://commercetools.com/blog/agentic-commerce-the-case-for-foundational-readiness) 9. [Generative Engine Optimisation (GEO) and AI SEO for Ecommerce Brands. Charle.](https://www.charle.co.uk/articles/geo-ecommerce-optimisation-guide/) 10. [What is Generative Engine Optimization (GEO)? 2026 Guide. Frase.](https://www.frase.io/blog/what-is-generative-engine-optimization-geo) 11. [Five Key Technical SEO Factors for AI Search (GEO). Adcetera.](https://www.adcetera.com/insights/five-technical-seo-factors-for-ai-search-geo) 12. [10-step framework for generative engine optimization. Profound.](https://www.tryprofound.com/articles/generative-engine-optimization-geo-guide-2025) 13. [Agentic Commerce Has An Invisible Identity Gap. Forbes.](https://www.forbes.com/sites/forbesbooksauthors/2026/08/07/agentic-commerce-has-an-invisible-identity-gap/) 14. [Legal Liability and Agentic AI: How the Law Applies When Bots Go Rogue. Duke Law.](https://law.duke.edu/news/legal-liability-and-agentic-ai-how-law-applies-when-bots-go-rogue) 15. [From Chatbot to Checkout: Who Pays When Transactional Agents Play? Future of Privacy Forum.](https://fpf.org/blog/from-chatbot-to-checkout-who-pays-when-transactional-agents-play/) 16. [Agentic AI Commerce: The Next Wave of Online Shopping and Retailer Risk. Sheppard Mullin.](https://www.sheppard.com/insights/blogs/agentic-ai-commerce-the-next-wave-of-online-shopping-and-retailer-risk) 17. [Agentic Commerce: Threats and Risks. Visa.](https://corporate.visa.com/en/sites/visa-perspectives/security-trust/the-threats-landscape-of-agentic-commerce.html) 18. [Agentic Commerce: Risks, liability and trust. Shopware.](https://www.shopware.com/en/news/agentic-commerce-paul-krauss-interview-part-2/) 19. [Generative Engine Optimization: How to Dominate AI Search (arXiv:2509.08919). Further reading.](https://arxiv.org/html/2509.08919v1) 20. [Agentic commerce in 2026: Why delivery decides who wins. nshift. Further reading.](https://nshift.com/blog/agentic-commerce-future-of-ecommerce) About rawmktg. rawmktg. publishes data-driven teardowns and technical playbooks on GEO, agentic commerce and B2B AI-search visibility. Method: same data, same lens, every time. Contact: vinayak@rawmktg.com Sources: the published UCP, ACP, AP2, MCP and A2A specifications and vendor documentation, McKinsey and Visa PERC research, and legal analysis, 2025-26. Code samples are illustrative reference implementations.