Briefings
Obenan Briefings
Original analysis on the infrastructure layer between merchant systems and AI commerce.
More briefings
Four AI engines can name four different local winners
Appearing in one AI assistant is not proof that you appear in the others. And looking once is not measuring.
An AI assistant can now shop across the web, if the customer approves
Stripe says Meta’s Muse assistant can buy on a customer’s behalf at more than a million businesses that accept Link, and can pay with a single-purchase virtual card elsewhere. Stripe says the customer approves the total before payment. That is one payment route for US consumers, published by the company that built it.
Payment networks are learning to identify the agent
Software is starting to buy things on a customer’s behalf. For such a payment to be worth trusting, someone has to be able to establish which piece of software is acting, which person it acts for, and what that person allowed it to do. One payment provider has now published a protocol aimed at carrying that information, and says work on an interoperability framework with two card networks has begun.
Building the commerce agent is getting easier. Connecting it to a real merchant is still the hard part.
Anthropic's commerce blueprint helps teams build shopper and merchant agents. Real catalogs, checkout, permissions and approval still need integration.
A visible AI answer is not a customer
Being named in an AI answer is one event. Winning a customer is another. A US Google Search experiment measured clicks to publishers, not customers.
Making the list is not the same as being chosen
Being named by AI, named first and described correctly are separate problems. A seven-day study of two Bali submarkets helps measure each. No ranking recipe.
From a question in Maps to an order in the kitchen
Square says eligible US restaurants can appear in Ask Maps and turn a question into an order. The restaurant still owns the records and the promises behind it.
AI Search is available in France. Visibility is not automatic.
Google announced AI Search in France. Eligibility, appearance, citation, recommendation and customer action are separate states. Check each; do not assume.
Search visibility is not AI visibility
We asked Google Maps, ChatGPT and Claude the same restaurant question in three cities on the same day. They did not return the same places.
Google AI Mode can find an available restaurant table in South Africa. Being visible now includes being bookable.
Google's AI Mode in Search can check live table availability for restaurants in South Africa and hand the diner to a booking partner to finish the reservation. If your booking information is stale or disconnected, the journey can break right after a customer finds you.
On July 1, 2026, Cloudflare split AI traffic into Search, Agent, and Training.
A major edge platform stopped treating AI access as one on-or-off switch. AI traffic is now three separate operating decisions: stay discoverable, govern real-time agents, and limit training crawls, with a crawl-to-referral ratio to test whether that traffic sends anything back.
The committability ladder
Agentic commerce is not one step. It is five: discovery, validation, commitment, execution, and settlement. Networks, payment service providers, and open protocols are advancing the surfaces around the merchant. The stage in the middle, where a local service authors the commitment it can make and keep, is still the merchant's own work.
AI shopping visibility is moving from brand mentions to product routing
AI shopping surfaces used to be measured by how often they mentioned your brand. They now route buyers to specific products by reading product feeds. The part you control is catalog and feed truth. Ranking and recommendation stay on the platform side.
Ask Maps shows local AI rollout is market-specific
Google Ask Maps is arriving market by market: the US and India first, then Brazil in Portuguese. Availability of a local AI discovery surface is market-specific. That is not the same as a ranking change, and it reveals nothing about how any business is surfaced.
On June 17, 2026, Adobe turned Semrush's GEO data into a closed-loop AI visibility platform
Adobe Brand Visibility packages prompt intelligence, edge deployment, bot verification, and revenue attribution as one workflow. The category signal is broader than the product: enterprise GEO is moving from dashboards toward measure, deploy, verify, and attribute.
Agentic commerce stacks are moving upstream. Merchant commitment for local service is still missing.
Between Google's Universal Commerce Protocol rollout, Visa's Intelligent Commerce expansion, and the June 16, 2026 launch of Adyen Agentic, the feed, cart, readiness, trust, and orchestration layers of agentic commerce are arriving fast. The layer none of them author is the one a restaurant, clinic, salon, or hotel needs most: the commitment a local service must make and keep.
Google turned AI visibility into a control plane. A control plane is not a verdict.
On June 3, 2026, Google published Search Console generative AI performance reports for Search and Discover, plus an eligibility control with include, exclude, and inherit states, effective for the first cohort on June 17, 2026. AI visibility is now first-party, reportable, and governable. Include does not guarantee appearance, ranking, or recommendation. What the report counts is impressions, not clicks, rankings, recommendation quality, business impact, or whether the facts in the answer were correct.
Agentic checkout tolls are becoming negotiable. Merchant readiness is the durable control point.
AI commerce surfaces are not converging on one checkout fee or one custody model. They diverge on what the merchant pays, who stays merchant of record, and where payment and fulfillment sit. The part that travels across all of them is current, authorized merchant readiness: the catalog, pricing, availability, policy, location scope, and acceptance state an agent reads before it acts.
Merchant-owned agents are becoming an acceptance surface. Merchant truth is still the hard part.
In May 2026, Mastercard, PaymentAuth, and Circle moved agentic commerce toward merchant-owned surfaces and standardized payment objects. None of it yet proves that a local service can be committed to, held, and honored. That gap sits upstream of payment, and it is where the merchant still carries the risk.
Mastercard Agent Pay went live in Portugal. Merchant commitment is still the unsolved part.
On May 5, 2026, Mastercard said it ran a first live Agent Pay transaction in Portugal with a local partner, PayOS, and opened a Lisbon Centre of Excellence for Innovation. It is one more stop in a country-by-country European rollout. Each new market proves an agent can pay. None yet proves the local service behind the purchase can be committed to and honored.
AI visibility is now measurable. Measuring it is not the same as fixing it.
Between March and April 2026, two independent tools turned AI visibility into something you can count. Peec moved citation diagnostics to the URL level and then shipped Agent Analytics for first-party crawl logs. Ahrefs launched Bot Analytics and exposed the hidden fan-out queries behind AI answers. The dashboards now show whether AI systems reach you and cite you. They do not change what those systems find about you when they arrive.
Google is hardening who can edit your business facts. Merchant-authored truth is the answer.
On April 16, 2026, Google added Gemini-based blocking of policy-violating Business Profile edits, faster review-scam detection, and proactive owner alerts. The gate around merchant facts is tightening, which raises the value of truth a merchant authors and can defend.
The web is being scored for agent readiness. For a local business, the score still rests on merchant truth.
On April 17, 2026, Cloudflare launched an Agent Readiness score and a Radar dataset that turn agent-facing web standards into something operators can measure. Its scan of 200,000 domains showed how rare the new signals still are. The score grades whether a site can be discovered, fetched, and read by agents. It does not grade whether the local business behind the page tells the truth.
Local AI visibility is moving from answers to actions.
By late May 2026, the two largest consumer AI surfaces with local intent had publicly moved their local layer from recommendation only toward recommendation plus action. The single citation number stops predicting outcomes.
Paid agent actions are becoming infrastructure. Merchant commitment is still unsolved.
Agent platforms are turning paid APIs, paid MCP servers, paid web content, and machine payments into native runtime infrastructure. That answers how an agent pays. It does not answer whether the merchant on the other side can honor a specific commitment at the moment money moves.
Prepaid local services: a practical wedge for agentic commerce
Bounded local commitments are where agentic discovery meets agentic payment. The merchant side still has to be true before money moves.
AI agents are starting to pay. Merchants still need to prove the offer is real.
Visa, Mastercard, and American Express are building ways for AI agents to pay. But a payment connection cannot tell a customer whether a product, service, price, or booking slot is actually available. The merchant must keep those facts current.
Mastercard's Thailand ride-booking test gives agentic commerce a real local-service proof point
On April 7, 2026, Mastercard publicly showed a live agentic ride-booking transaction in Thailand, making local-service committability more concrete than generic checkout demos while still leaving merchant-side truth and service semantics unresolved.
Before an agent can book, order, or pay, it needs merchant truth
Execution layers are advancing fast. Merchant truth is not. Identity, availability, acceptance, and current business state remain the weakest link in agentic commerce.