Authors
Introduction
AI’s impact on media planning and buying was previously largely prospective; it is now being implemented at scale. The industry is shifting from impression-by-impression trading toward agentic decision-making. While fully autonomous media buying remains nascent, agencies, publishers, and ad-tech providers are actively testing agentic tools that may ultimately plan, execute, and optimize media transactions continuously within goals and guardrails set by humans, but with substantially less day-to-day human involvement. As AI has developed, so too have the types of AI being used – from generative AI that produces creative variations and media plans to agentic AI that coordinates decisions across complex workflows. Agency development of varying types of AI tools has become commonplace, with holding companies building proprietary platforms for their portfolio of agencies. For brands facing fragmented audiences, compressed budgets, and cross-platform complexity, the appeal is clear – but the evolving capabilities also raise significant legal, commercial, and operational considerations.
Media planning and programmatic buying: Risk versus reward
In terms of media planning (i.e., the process by which advertisers decide where, when, and how often an ad should run to maximize engagement and ROI), AI can create bespoke media plans within minutes by automating data analysis, targeting personalization, and optimizing campaigns. It can also identify patterns and insights that are otherwise invisible to human perception, reducing the time and resources needed to refine campaigns. As AI assumes more routine planning and optimization functions, the role of human media planners may increasingly shift toward higher-level strategic planning, defining objectives, and establishing clear parameters and guardrails within which AI tools operate. As testing and analysis costs fall, creative and media placement efficacy is easier to isolate, and outcome-based KPIs are the prize.
AI is also reshaping programmatic buying. By analyzing audience data and adjusting bids in real time, advertisers can achieve more efficient, targeted, and cost-effective ad placements and improved campaign performance. Agentic AI programmatic media buying is being pitched as the next frontier, with proponents arguing it could address long-standing inefficiencies in the programmatic supply chain by automating routine campaign execution and optimization, facilitating more direct interactions between buyers and sellers, and potentially reducing reliance on intermediaries. However, the buy side is understandably more cautious for the following reasons:
1. Transparency, explainability, and delegated authority
The black box problem persists, but its shape has changed. The question is no longer simply how an AI model scored an impression. As AI systems gain greater autonomy, brands must also consider how much authority an AI agent holds, what actions it may take without approval, and whether its decisions can be reconstructed. Vendors and agencies typically frame autonomy as a choice for the advertiser – how much to delegate, which controls stay fixed, and how risk is managed. AI-driven decisions are likely to be treated by regulators as company decisions, and the absence of human intervention will not shield brands from liability. Brands should treat delegation and responsibility as a contractual and governance matter by clearly defining the scope of an AI agent’s authority, applicable guardrails, actions requiring human approval, and liabilities for the actions of AI agents taken outside of advertiser-approved parameters.
Brands should also consider not only what an AI agent is permitted to do, but what it is instructed to optimize for. Over-optimization may prioritize short-term KPIs at the expense of broader brand objectives such as brand awareness, brand equity, or long-term customer value. Brands should therefore define both the objectives an AI system should pursue and the parameters within which it may pursue them.
2. Ownership and confidentiality
Ownership and use of AI outputs remain critical commercial issues. Brands should clarify who owns an AI-generated media plan and any AI-generated creative, and address what happens when an AI tool trained on the brand’s data is deployed for competitors within the same holding company. First-party data and the direct customer relationship remain valuable assets. Where one AI tool serves multiple clients, brand data and results may potentially benefit competitors. Brands must consider whether they are comfortable with agencies or vendors utilizing the brand’s data to improve AI tools and, if so, should clearly define and distinguish what data constitutes confidential information that should not be used to train any such tools.
3. Automated decision-making and bias
AI-driven planning still depends on large volumes of data, including personal information, keeping data collection, processing, storage, and security squarely in scope; however, it now includes a wider compliance perimeter that adds AI-specific transparency, documentation, and the use of inferred data. Specifically, AI models can derive inferences about consumers, including potentially sensitive characteristics such as health, race, and income, from combinations of data that may not themselves appear sensitive. Those inferences may then be used for audience development, targeting, or optimization. Brands should therefore consider not only what data is provided to an AI system, but also what characteristics the system may derive or infer from that data and how those inferences may be used.
Models trained on limited or unrepresentative data or relying on inaccurate or biased inferences can produce biased outcomes, including discriminatory ad targeting based on protected characteristics such as race, gender, age, or zip code proxies. Where automated systems decide which ads to serve to which audiences, discriminatory decision-making – particularly for consequential decisions such as product and service pricing or housing, employment, lending, and health care advertising – could expose brands to legal liability and reputational harm.
The EU AI Act’s Article 50 transparency obligations take full effect on August 2, 2026, and roughly 20 U.S. states have AI-specific laws passed or in development, with California, Colorado, and Connecticut expanding automated decision-making obligations. Human oversight is essential – brands should require meaningful human review of AI-driven targeting decisions and contractual guardrails mandating bias audits, explainability requirements, and the ability to override or pause automated targeting.
4. Brand safety, fraud, and verification
AI has still not proven a silver bullet for brand safety or invalid traffic. Fraud techniques continue to evolve and can mimic human behavior faster than models retrain, and the volume and nuance of online content continue to challenge automated context classification. The best antidote for brands remains pairing AI-enabled brand safety and fraud detection tools with appropriate human oversight, backed by the brand’s own tags, independent audits of DSP and ad server log-level data, and outcome-based measurement. Advertisers should continue contracting for that access, which will become increasingly important as AI is used not only to optimize media delivery, but also to inform campaign reporting, measurement, and attribution.
Advertiser best practices
Disclosure, governance, and consent. Require disclosure of every AI system an agency intends to use, its function, training methodology, and any third-party models in the chain. Identify which tools require prior advertiser consent versus those permitted without approval, with a right to withdraw consent. Require agencies to maintain appropriate AI governance policies and mechanisms for identifying, escalating, and addressing errors, inaccuracies, or other undesirable outcomes.
- Understand buying platforms and use of AI. If buying media directly, understand the buying platforms, how they work, and the use of AI within those platforms. Ensure your teams fully understand functionality and default settings, and whether settings revert to default after each transaction.
- Risk assessment, testing, and due diligence. Subject AI uses to appropriate risk assessment, due diligence, and testing prior to deployment. For agentic or other higher risk uses, consider controlled testing and defined criteria for evaluating performance and compliance with advertiser-established guardrails before expanding autonomous authority or media spend allocated to such tools.
- Delegated authority. Define which decisions an AI agent may take autonomously, spend and pacing limits, prohibited inventory, and which actions require human approval.
- Human oversight. Require human review of final decisions, auditable decision logs, and the right to pause or roll back autonomous activity.
- Ownership and confidentiality. Clarify ownership of AI-generated outputs and usage rights in data inputs. Define what may or may not be used to train AI tools; regardless, use of confidential information should be carved out.
- Bias safeguards. Establish restrictions on the use of sensitive data and sensitive inferences for audience development, targeting, or optimization. Require guardrails against discriminatory targeting based on protected characteristics, regular bias audits, explainability requirements, and the ability to override or pause automated targeting.
- Brand safety. Preserve full access to log-level data and pair AI tools with the brand’s own tags, independent audits, and outcome-based measurement.
- Insurance. Ensure E&O coverage addresses AI-driven operational risks.