● Crawling collapses, Agent shopping explodes, API protocol war reshapes commerce
Agent Shopping Era Begins: “Exposure” No Longer, “Invocation” Determines Revenue
In today’s piece, I will summarize it into exactly four takeaways.
1) Why the “commerce standard protocol” announced by Google and OpenAI will upend the commerce landscape
2) What GEO (Generative Engine Optimization) and UCP/ACP will each change
3) The data/content/organizational change roadmap brands and retailers must prepare immediately
4) The “real core point risks and opportunities (hidden battlegrounds)” that other news/YouTube rarely cover
1) One-line breaking news: Shopping shifts from “crawling → protocol (API)”
core point change is that we are moving from an era where AI “reads” web pages (crawling)
to an era where AI “communicates directly” with company systems to “receive” data (protocol/API).
This change is nuclear because the control over the purchase journey shifts from “human clicks” to “AI agent execution.”
Search → comparison → checking reviews → inventory check → payment → delivery
Entire intermediate steps become automated in this new structure.
2) How the consumer experience changes: “Conversation + Delegation” rather than “Search”
Before: Consumers searched directly, moved between multiple sites to decide and pay
Change: Consumers state their conditions, and AI recommends and completes the purchase on their behalf
Example (original core point)
When asking “Can I buy Jo Malone near Euljiro?”
Today you might get an answer like “I heard it’s available at some store,”
but in a protocol-based future AI will check and complete the purchase including
– how many units are in stock right now
– the likelihood it will remain in stock if delivered in 40 minutes
– what discounts/promotions apply after 6 PM
– the fastest pickup/delivery option and the cost
all verified and processed by AI up to payment.
3) Two pillars driving AI commerce: GEO vs UCP/ACP
3-1. GEO = “the battle to get on the recommendation list (qualifiers)”
GEO is, simply put, the work that makes a generative AI confident that “this brand fits this situation.”
Whereas SEO used to be the fight to “appear at the top of search results,”
GEO will be about “getting into the AI’s recommendation candidate pool.”
Important point here.
Providing good data via protocol does not guarantee AI recommendations.
AI cross-verifies layers such as reviews/community/social reactions,
and only includes brands that it concludes are “trustworthy” in the candidate pool.
3-2. UCP/ACP = “the battle to complete the purchase (finals)”
Protocols like UCP/ACP show their value after a brand has made it into the GEO candidate list.
This is the stage where AI executes “Okay, I’ll make the purchase right now under this user’s conditions from here.”
To summarize in two stages
1) GEO: “Can I recommend you?”
2) Protocol: “Then I’ll proceed to payment/delivery under optimal conditions now”
Because of this structure, brands must prepare both brand trust (content/reputation) and purchase execution data (API/inventory/price/delivery).
4) From a corporate perspective, the definition of “marketing” changes: the data communication war
The sharpest point in the original text was this.
A protocol is ultimately a “language for exchanging data,” and
from the AI’s perspective it prefers suppliers that have rich data needed to satisfy consumers.
So marketing shifts from simple advertising/campaigns to
providing “AI-executable information” including structured product data, conditions/policies, and real-time inventory/delivery.
This direction reduces friction in customer experience (CX),
and over the long term will reshape customer acquisition cost (CAC) structures.
5) Offline is not an exception: “dynamic pricing + congestion” enters via API
People often think AI commerce is just “online shopping automation,” but
the truly large market is the digitalization of offline.
Examples given include
– food court discounts (50% after 6 PM)
– current waiting line/congestion
– seat availability (comfort)
– personal preference/budget
AI can take these real-time inputs and change behavior like “today Lotte is the better fit.”
In other words, it’s not about inflation-style price swings,
but a structure where “situation-based conditions (congestion/time/stock/promotion)” are directly reflected in perceived product value.
6) Korea market watching points: the “protocol war” is not as simple as overseas
The U.S. is likely to go to a three-way competition among OpenAI/Google/Amazon,
but Korea already has strong players like Naver and Coupang that control payment/logistics/search touchpoints.
From the original perspective,
– even if overseas protocols (OpenAI/Google) enter,
– domestic platforms are likely to defend their ecosystems (closed/own specs)
This means for brands,
it will not be “connect once and be done” but
a multi-protocol (overseas + domestic) 대응 strategy will likely be necessary.
7) Practical checklist (roadmap) brands must prepare immediately
7-1. For GEO (getting into recommendation candidates): design content that occupies the “meaning space”
Goal: Build the evidence so AI can say “this brand fits this situation”
– Define category entry points (when consumers look for this product)
– Content on comparison criteria (decision factors): “A vs B”, “recommendations by situation”, “fail-safe choices”
– Plan for spreading reviews/community/UGC (own online store alone is insufficient)
– Strengthen brand trust signals (expertise/experience/evidence/transparency from an E-E-A-T perspective)
This continues from SEO perspectives, and as search moves to generative AI it’s increasingly important how the brand is invoked in context.
7-2. For protocols (purchase execution): API-ize product/inventory/price/delivery data
Goal: Enable immediate purchase when a recommendation occurs
– Structure pricing/discount conditions (subscriptions, bundles, time-based, etc.)
– Real-time inventory/store inventory/pickup availability
– Delivery SLA (same-day/next-day/express) and costs
– Standardize return/exchange policies and customer service policies
AI agents dislike “ambiguous information.”
Brands with clear, structured data will be recommended more often.
7-3. Prepare a “brand agent”: the next version of D2C
Brands will not want to be dependent solely on platform agents.
So after D2C comes the brand agent.
A brand agent goes beyond CRM automation.
– Predict customer usage cycles (consumables/filters/subscription reminders)
– Recommend according to customer situation (budget/taste/purpose)
– Negotiate terms with external shopping agents (price/delivery/bundles)
Ultimately, it becomes the weapon to protect the “customer touchpoint.”
8) The “most important points” other news/YouTube rarely cover
8-1. Many brands may attach protocols but see no revenue
People mistakenly think
that just connecting via API will cause AI to recommend them.
But AI first filters on “is this brand trustworthy?”
In other words, protocols are the final tool and GEO is the qualifier.
8-2. The battleground is “policy” not “advertising” (price/delivery/returns/inventory)
In agent shopping, what determines conversions is not ad copy but
policy data like “make sure I receive it tomorrow,” “is it in stock now?”, “is returns easy?”
The company that provides this in real time wins.
8-3. Why “2–3 years” is the real golden time: meaning space preemption effect
The original text had an important hypothesis.
Although agents may seem different, the underlying meaning spaces are likely to be shared by a few large models.
Brands that secure a lot of early training/citation/recommendation advantages may be favored.
This could create a structure where “catching up later doesn’t easily reverse the lead,” similar to market interest rates.
9) One-line economic outlook: Commerce will reorganize from “platform fees” to “agent fees”
Until now, platforms (marketplaces) held distribution power,
but going forward agents will hold purchase decision power and
fees, exposure, and customer data power may shift again.
In this process,
– global supply chain reorganization (which logistics/inventory connect faster)
– digital transformation investment (data/API/governance)
– big tech regulation (data monopolies, payment dominance)
will also grow as macro issues.
From a blog SEO perspective, it is no longer about simple keywords but
about creating “information structures that get cited” in generative AI search.
< Summary >
The commerce standard protocols from Google and OpenAI turn crawling-based shopping into API-based “agent shopping.”
Commerce will operate in two stages: GEO (getting into AI recommendation candidates) and protocols (UCP/ACP, purchase execution).
Brands must prepare both trust-building content (to occupy meaning space) and structure/API-ize price, inventory, delivery, and policies.
The real battleground is policy data rather than advertising, and offline factors like congestion/discounts/inventory will be integrated via APIs and reshuffle the market.
[Related posts…]
- GEO era: How brands enter AI recommendation lists
- Understanding AI commerce protocols (UCP/ACP) at a glance
*Source: [ 티타임즈TV ]
– 에이전트 쇼핑시대, 브랜드는 어떻게 해야하나? (박세용 대표, 강정수 박사)
