Yes, AI agents can buy retail media through AdCP, the Ad Context Protocol, wherever a retail media network runs an AdCP sales agent. The protocol has a dedicated retail_media channel and a Commerce Media guide that models sponsored products, on-site display and video, off-site audience extension, premium placements and in-store screens as products a buyer agent can discover, buy and measure. The retail media seller specialism is still marked Preview in AdCP 3.2, so support varies from one network to the next.

TL;DR. AdCP, published by AgenticAdvertising.org, treats each retailer as a separate sales agent and each retail media offer as a product with its own format, catalog requirements and measurement. Brands sync product feeds to the retailer account, buyer agents find eligible items through GTIN matching, and delivery reports carry closed-loop metrics such as return on ad spend and in-store conversions. People still own the retailer relationship, the trade-offs between retailers and the approval of commitments that matter.

Can AI agents buy retail media using AdCP today?

AdCP can carry a retail media buy from discovery to delivery reporting, but only with retail media networks that have stood up an AdCP sales agent. The AdCP Media Channel Taxonomy defines retail_media as "retail media networks and commerce marketplaces", and notes that one retail media budget can buy sponsored products on websites, display in apps, off-site ads using retailer data and in-store digital screens.

Maturity is the caveat. According to What's new in AdCP 3.2, the sales-retail-media specialism covers catalog sync, keyword sponsored search and item availability, and its status is Preview. The AdCP compliance catalog explains that a preview specialism's ID and scope are reserved while the underlying protocol surface stabilises, and that its conformance test reports a preview result rather than a verified pass or fail.

The protocol underneath is generally available. The AdCP versions page lists 3.2 as the current production release, published on 30 September 2026, with 3.3 in beta.

Because retail support is at preview stage, testing is part of the answer. For example, Alkimi is neutral on standards and can work with agents built on AdCP or IAB Tech Lab's Agentic Advertising Management Protocols (AAMP). In its Gym, a retail media network can build its seller agent and a brand can build its buying agent, set the rules each works to and let them spar before any real money moves, while Concourse benchmarks a buying agent without asking which standard it uses.

What does the AdCP documentation say about retail media?

The documentation's central argument is that a retail media network should stand up an "agentic storefront" rather than build another self-service platform. The Commerce Media guide says that "when agents handle discovery, the advantage shifts from UI quality to data quality", which is where most retailers already have a moat. The retailer defines the products and rules, buyer agents handle discovery and execution, and every transaction runs through the retailer's own server.

Mike Feldman, SVP of commerce at Flywheel, told Digiday in January 2026 that what Walmart uniquely brings to AI-first retail media "always comes back to their scale of first-party retail data and closed-loop measurement".

The guide maps retail media concepts to protocol objects: a retail media network becomes a sales agent, GTIN and SKU selection become catalog selectors, and closed-loop measurement becomes declared metrics on each product. Each retailer is a separate sales agent, so a cross-retailer campaign is several conversations held in one format rather than one conversation with a single intermediary. The guide is candid that this does not replace everything: brands that are not using buyer agents still need basic self-service access.

How could a retail media network expose its inventory to agents?

A retail media network exposes inventory to agents by modelling each offer as a separate product in the retail_media channel. The Commerce Media guide recommends distinct products for sponsored products, on-site display, off-site extension, premium placements and in-store digital, because that is "how buyer agents comparison-shop across retailers". Sponsored product listings are rendered by the retailer from its own catalog data matched by GTIN, so the brand supplies a product feed rather than finished creative.

Catalog matching does the work a sales call used to do. A brand syncs its product and inventory feeds to the retailer account through sync_catalogs, and the retailer returns a catalog_match field showing which of the brand's items it carries. The guide compares this to a sales rep saying "we carry 3 of your 1,200 products", except that it happens automatically at query time.

Physical stores are treated as the retailer's data, not the brand's. According to the guide, store locations are seller-provided, with catchment areas that buyer agents can reference for proximity targeting.

Under AdCP 3.2, sellers advertise the lifecycle tools they serve, and where a seller does not, buyers fall back to the get_products, create_media_buy and update_media_buy facades, which stay available throughout 3.x. A buyer agent should read each network's advertised tools before assuming one flow works everywhere.

How do retail signals and shopper data reach a buyer agent?

Retailer shopper data reaches buyer agents in two ways under AdCP: inside the retailer's own products, and as signals the retailer publishes. The signals ecosystem guide gives retail media networks a dual role, as a publisher selling inventory and as a data provider.

As a data provider, a retail media network can publish signal definitions alongside its properties in the same adagents.json file, according to the AdCP data provider guide. That guide says retail signals are "especially valuable because they're deterministic", based on actual purchases rather than modelled behaviour. How get_signals and activate_signal move signals into a buy is covered in How Does AdCP Handle Audiences and Signals?

Real-time product choice is a newer layer. AdCP's Trusted Match Protocol is marked experimental, and its retail media page describes a retailer asking the buyer which catalog items to feature on a search or category page, within packages already agreed. The same page notes that today buyers "have limited real-time control over which specific products appear in which contexts".

How is retail media measured when an agent buys it?

Retail media measurement under AdCP is closed-loop and usually run by the retailer. The Commerce Media guide describes deterministic purchase attribution, matched through loyalty data, login state and point-of-sale records, as the asset that sets retail media apart. A product that declares platform-managed conversion tracking needs no buyer-side pixel, because the retailer provides always-on measurement.

Delivery reports carry return on ad spend, attributed sales and new-to-brand rate, with a breakdown separating website, app and in-store conversions. Each report states its attribution window, because retailers use different lookbacks, and the guide says this lets buyers "normalize ROAS comparisons across retailers". AdCP 3.2 adds a vendor_relationship field that, in the words of the 3.2 overview, "discloses seller-owned measurement".

That disclosure matters because in retail media the seller is usually also the measurer. Reconciliation between a brand's records and a retailer's report still sits with the commercial relationship, a point covered for other channels in Can AdCP Buy Beyond Programmatic?

What stays with people when agents buy retail media?

Agents can take on the parts of retail media few people enjoy, such as matching SKUs across retailers and pulling reports into one format. People keep the judgement. The AdCP media buy specification lets a sales agent hold a task for internal human review, for example insertion order signing, before it commits.

Several retail decisions sit outside the ad buy altogether. According to the Commerce Media guide, enhanced product content and brand stores are managed through the retailer's content systems, not through the ad buy, and retailers commonly require co-branding and restrict landing pages to their own site. The account relationship with each retailer and the choice of which retailers deserve budget remain human calls.

A sensible pattern is earned autonomy, in which an agent moves through five stages: observe, recommend, draft, human-approved action and, only then, bounded automatic action within limits a person has set. How governance checks budgets and approvals before an agent spends is covered in How Does AdCP Governance Check Budgets, Brand Safety and Human Approval Before an Agent Buys?

What should brands and retail media networks check first?

Before relying on AdCP for a retail media buy, confirm these points with each counterparty.

Whether the network runs an AdCP sales agent, and on which version.
Whether it claims the sales-retail-media specialism, which is still Preview.
Which lifecycle tools it advertises.
Which attribution window it reports, and whether measurement is seller-owned.
Which steps it holds for human review.

For the wider business case for agents in this channel, see What Is Retail Media in Agentic Advertising?

What does this mean for teams testing retail media agents?

Retail media suits agent buying because its inputs are structured, but the protocol's retail specialism is still at preview stage, so testing against realistic counterparties before live spend matters. Alkimi offers three products for that stage: Concourse benchmarks any buying agent without asking which standard it uses, the Gym trains and spars agents regardless of protocol, and the Marketplace is neutral, works with or without a DSP, and keeps humans' approval rights at the deal level.

Within the Marketplace, the DealSheet is written at agreement, timestamped and held by both parties, so either side can retrieve what was agreed independently.

Frequently asked questions

Can AI agents buy retail media using AdCP?

Yes, where a retail media network runs an AdCP sales agent and models its offers as retail_media products. The retail media seller specialism is marked Preview in AdCP 3.2, so coverage differs from network to network.

What is an agentic storefront in AdCP?

It is the documentation's term for a retail media network exposing its inventory through an AdCP sales agent instead of asking brands to learn another self-service platform. The retailer defines the products and rules, and buyer agents handle discovery and execution.

Does AdCP support sponsored product ads?

Yes, the Commerce Media guide models sponsored product listings as products the retailer renders from its own catalog data, matched by GTIN. Brands sync product feeds through sync_catalogs and pick items with catalog selectors.

How does AdCP report in-store sales?

Delivery reports can break conversions down by action source, including website, app and in-store. The retailer usually runs this measurement itself, and each report states its attribution window.

Do people still approve retail media buys made by agents?

They can, because the AdCP specification lets a sales agent hold a task for internal human review before it commits. Account relationships, brand content and the choice of retailers also stay with people.