How do AI shopping and media-buying agents negotiate prices with publishers?

AI buying agents negotiate prices with publishers through a structured sequence: request, evaluation, counter-offer and commitment, all in machine-readable message exchanges. The process mirrors a human deal negotiation but operates at machine speed and volume. The critical governance requirement is that the final agreed terms are recorded in a format both sides can verify independently. Without that, the negotiation produces no enforceable commitment.

AI buying agents negotiate prices with publishers through a structured exchange of machine-readable messages. The sequence is: the buying agent sends a deal enquiry specifying inventory requirements and price parameters; the publisher agent evaluates against available inventory and terms; the publisher agent responds with an offer or a counter-offer; the buying agent evaluates the response against its mandate; and the exchange continues until both agents commit to terms or one party declines.

This mirrors a human deal negotiation in structure. The differences are speed, volume and the requirement that every step produces a machine-readable record.

At a glance

Deal enquiry: A structured message a buying agent sends to a publisher agent specifying the inventory it wants to access, the price range it will accept, and the constraints it requires.

Counter-proposal: A publisher agent's response to a deal enquiry — modified terms (different price, volume, or publisher environment) that the buying agent evaluates against its mandate.

Mandate ceiling: The maximum price or volume a buying agent is authorised to commit to without escalating to a human approver.

Machine-readable deal terms: Deal parameters structured as data fields rather than prose — parseable by both agents without human interpretation.

What does a buying agent's opening position contain?

A buying agent's initial deal enquiry, in a system conforming to the Agentic Advertising Management Protocols, contains the buyer's targeting requirements, a price range, volume parameters (minimum guaranteed impressions or spend), brand safety constraints and any approval conditions the human principal has defined.

The price range is not a single bid price: it is a mandate with a floor the agent will not go below and a ceiling it will not exceed without human escalation. The agent's negotiation logic operates within that range. Where the deal settles within the range depends on inventory scarcity, competitive pressure from other buyers and the publisher's own floor constraints.

How does a publisher agent respond?

A publisher agent evaluating an incoming deal enquiry checks the requested inventory against availability, applies its own floor pricing, and evaluates the brand safety and targeting parameters against what the inventory can technically deliver. If the buyer's requirements can be met within acceptable pricing, the publisher agent may accept the opening terms. More commonly, it returns a counter-offer.

A counter-offer modifies one or more elements of the original request: the price, the volume commitment, the targeting scope or the delivery period. The buyer agent evaluates the counter-offer against its mandate and responds accordingly. This exchange can run for multiple rounds in rapid succession before both agents commit.

What makes a negotiated price enforceable?

The negotiated price becomes enforceable when both agents sign the deal record. In a well-implemented agentic marketplace, the deal record is a structured document that contains the final agreed terms, a timestamp, and tamper-evident signed commitments from both agents confirming acceptance. The record is written to the marketplace at the moment of commitment and cannot be modified by either side unilaterally after that point.

This is the governance requirement that many current programmatic platforms do not meet. An SSP that returns a deal acceptance message via an API has not necessarily created a bilaterally verifiable record. The acceptance is in the buyer's log. The deal confirmation is in the SSP's database. If those records diverge later, there is no authoritative reference.

The distinction matters at scale. A buying agent executing a large number of deals per week will encounter a rate of delivery discrepancy that makes manual reconciliation impractical. The deal record is the resolution mechanism.

Can a buying agent go above its price mandate?

No. The price mandate is set by the human principal and enforced by the agent's operating parameters. An agent that exceeds its mandate has violated its instructions and triggered a governance failure. In a well-constructed agentic system, any deal that requires exceeding the mandate is escalated to a human for approval before commitment.

This is one of the earned autonomy principles described in the Agentic Advertising Organisation's 2026 operating model: bounded automatic action. The agent acts automatically within defined bounds. Actions outside those bounds require human approval. A marketplace that allows agents to commit deals outside their stated mandate without human escalation is not a safe environment for buyer deployment.

What happens when both agents agree?

When both agents agree terms, the deal record is committed in the marketplace. Both the buying agent and the publisher agent receive a reference to the signed deal. Delivery begins against the committed terms. The deal record is available to both sides for the duration of the campaign and provides the reference point for any delivery query or discrepancy resolution.

In this model, the negotiation is complete when the deal record exists, not when an API call returns a success status. Alkimi's DealSheet is an implementation of this approach: a signed bilateral record committed to the marketplace at the moment both agents accept, accessible to both sides independently, not when an API call returns a success status. The existence of the signed record, accessible to both sides, is the commitment.

Frequently asked questions

What is A2A protocol and how is it used in advertising?

A2A (Agent-to-Agent) is Google's 2025 specification for peer-to-peer communication between AI agents. In advertising, it provides the messaging layer over which buying agents and publisher agents exchange deal enquiries, counter-proposals, and commitments.

What protocols let advertising agents transact directly with publisher agents?

The primary standards are A2A for agent communication and AAMP (Agentic Advertising Management Protocols) for deal term structure. A2A handles the messaging layer; AAMP defines what a machine-readable deal record must contain for both sides to verify what was agreed.

How does agent-to-agent negotiation reduce intermediaries in ad buying?

Agent-to-agent negotiation allows a buying agent and a publisher agent to commit to a deal directly, recording terms in a bilateral deal record. The neutral marketplace holds the record; separate SSP and DSP intermediaries are not required to relay terms between the two sides.