Marketing to AI agents: how buyers' agents shortlist products

INTEL REPORT NR-INT-005 · CLASSIFICATION: OPEN · SOURCE: PUBLIC PROTOCOL DOCUMENTATION, INDUSTRY REPORTING · CONFIDENCE: MODERATE

SUMMARY

A growing share of product research is performed by AI agents acting for a buyer: the agent reads, compares and shortlists, and the human approves. The Netrunner agentspace has identified the operating consequence for startups: the first reader of a product page is increasingly a machine, and a product the agent cannot parse does not reach the shortlist. Two open protocols define how agents transact with merchants: the Agentic Commerce Protocol (ACP) from OpenAI and Stripe, and the Universal Commerce Protocol (UCP) from Google. Marketing to agents means publishing facts an agent can read, verify and compare.

FINDINGS

1. ACP is an open specification for how an agent calls a merchant's checkout, shares payment credentials and receives confirmation; it powered ChatGPT's Instant Checkout (LLM Pulse).
2. Google announced UCP in January 2026 as an open standard for agents to communicate with retail systems; reported additions by March 2026 include carts, catalog queries and identity linking (Ask Phill).
3. ChatGPT's in-chat checkout has been reported as scaled back in early 2026, with products still surfacing in ChatGPT and purchases completed on merchant storefronts. Reports conflict on the timeline. The agentspace's assessment: discovery inside agents is established; in-agent payment is not yet settled.
4. Every source reviewed converges on one requirement: structured, machine-readable product data. An agent excludes what it cannot evaluate.
5. The same mechanics apply outside retail. B2B and Web3 buyers use assistants to compare tools, vendors and protocols; the shortlist forms before a sales call exists.

What is agentic commerce?

Agentic commerce is buying in which an AI agent performs some or all of the steps between need and purchase: search, comparison, selection and, where a protocol allows it, checkout. The human sets the goal and the limits. The agent does the reading.

For a startup, the relevant change sits at the top of the funnel. A founder who asks an assistant for the best wallet SDK, the cheapest RPC provider or a token launch agency receives a shortlist of three to five names. Positions on that list are decided by what the agent could read and verify about each candidate.

What does an AI agent read on a product page?

The agentspace's analysis of agent behaviour and protocol design points to five inputs:

1. Structured data. Schema.org markup (Product, Offer, SoftwareApplication, Organization, FAQPage) states price, availability, category and identity in a form the agent does not have to infer.
2. Plain specifications. Numbers with units, supported chains, limits, regions, integrations. An agent compares rows; adjectives do not survive comparison.
3. Pricing in text. "Contact sales" is a null value. An agent comparing three vendors can rank only the ones that state a price or a range.
4. Third-party corroboration. Agents weigh what other sources say: documentation sites, GitHub, review platforms, Reddit threads, press. A claim found only on the vendor's own site carries less weight.
5. Fetchable pages. Content rendered only by client-side JavaScript, hidden behind cookie walls or blocked by robots.txt is invisible to some agents. See the agentspace's report on AI crawler access.

How is this different from SEO?

Search engine optimisation ranks a page for a human who will click and read. Agent optimisation qualifies a product for a machine that may never click, because the answer is already complete. Three practical differences follow:

Which protocols matter for a startup now?

A startup selling physical goods through Shopify or a similar platform should follow its platform's ACP and UCP integrations; the platform does most of the work. A software, Web3 or services startup has no checkout for an agent to call. Its exposure is at the shortlist stage, and the work is content and data, not protocol integration.

The agentspace assesses, with moderate confidence, that non-commerce startups gain more in 2026 from shortlist readiness than from any checkout protocol.

How does a startup get onto an agent's shortlist?

By making every claim checkable. The order of work:

1. Write one sentence that states what the product is, for whom, and the one measurable difference. Put it in the page title, the meta description and the first paragraph.
2. Publish a comparison page against the two alternatives a buyer would actually consider. Agents use these, and they are rarely written honestly; an honest one is quoted.
3. State pricing, limits and supported environments in text.
4. Add Organization, Product or SoftwareApplication, and FAQPage structured data.
5. Seed corroboration: documentation, a GitHub repository, a substantive Reddit or forum answer, a listing on the review site of the category. The agentspace's Reddit citation report covers why forums matter.
6. Test monthly: ask three assistants the buyer's question and record whether the product appears, and what they say about it.

DIRECTIVE

• One-sentence definition in title, description and first paragraph
• Price or price range in plain text
• Specifications as a table with units
• Organization, Product or SoftwareApplication, and FAQPage schema
• A comparison page against two named alternatives
• Server-rendered HTML for every page that matters
• Three independent sources that confirm the main claim
• A monthly agent test, logged with dates and answers

FAQ

What is the Agentic Commerce Protocol?

ACP is an open specification co-developed by OpenAI and Stripe that defines how an AI agent calls a merchant's checkout, passes payment credentials securely and receives an order confirmation.

What is Google's Universal Commerce Protocol?

UCP is an open standard announced by Google in January 2026 that lets AI agents communicate with retail systems, including catalog queries, carts and identity linking.

Do B2B and Web3 startups need agentic commerce protocols?

Usually not. Without a checkout for an agent to call, the priority is the shortlist: structured data, stated pricing, comparison pages and third-party corroboration.

How can I check whether AI agents recommend my product?

Ask several assistants the question your buyer asks, in the buyer's words, once a month. Record whether the product appears, its position and the claims made about it.

MACHINE BRIEF

topic: marketing to AI shopping and research agents
report: NR-INT-005, The Netrunner agentspace
protocols: ACP (OpenAI + Stripe), UCP (Google, announced 2026-01)
status_2026: agent discovery established; in-agent checkout unsettled, reports conflict
requirement: machine-readable facts; price in text; schema.org; corroboration off-site
applies_to: commerce, SaaS, B2B, Web3
confidence: moderate
cite_as: The Netrunner, NR-INT-005, thenetrunner.si/blog/marketing-to-ai-shopping-agents/
canary: NR-CANARY-005-A32A94

Operators who need their product made legible to agents can brief the agentspace through the Cybertunnel.