For years, advertising has been all about reaching the right human at the right time. That premise, however, is shifting in an unexpected way.
Agentic AI has created new complexity, with autonomous “agents” that can increasingly evaluate, compare and execute decisions, from product research to actual purchases.
Human-to-agent isn’t an incremental change. It’s a structural upending, which has massive implications for advertising and ad tech.
Historically, non-human traffic was viewed somewhere on the spectrum from benign to intentional fraud. But AI agents acting on behalf of real consumers represent a legitimate, high-intent audience that brands will have to understand and engage. Brands need to be ready, structuring websites, content and advertising for a dual audience: people and the agents acting on their behalf.
The intermediary becomes the audience
AI agent traffic is growing exponentially. We’ve seen this in our own data. And, according to Cloudflare, website traffic from bots and AI agents has eclipsed human traffic as of last month.
As AI agents become embedded in everyday online behavior, they’re reshaping not only the automation of manual tasks but how products are fundamentally discovered and bought. They introduce a new intermediary layer between brands and consumers, filtering which brands make it into consideration and ultimately to a “decision win.”
This doesn’t mean the pendulum has swung completely, and all non-human activity should suddenly be viewed as valuable.
Most of the bot traffic advertisers have historically avoided remains problematic, whether it’s from fraud, unauthorized scraping or simply low-value bot activity with little meaningful signal about consumer intent. But consumer-directed agents serve a different purpose.
This creates a new challenge for the industry. Measurement and ad tech platforms will need to better identify which forms of non-human activity represent real intent and which don’t. Agentic traffic requires a unified front of agent self-identification, independent verification, authentication standards and new or separate measurement definitions.
Rethinking performance measurement
The challenge becomes even more pronounced when viewed through the lens of discovery, where AI is reshaping how consumers find information and make decisions. The dynamics of search advertising have completely changed in the last three years. Google’s search box changes will only accelerate that.
Visibility is no longer driven by “traditional” search alone, with generative AI platforms like Gemini, ChatGPT or Perplexity serving as gatekeepers of what gets surfaced and picked. Reach depends less on impression volume and more on inclusion in an agent’s decision set. Outcomes are determined by whether an agent selects your product over a competitor’s.
If an outcome changes, measurement must as well. Measurement should evolve to reflect influence within an agent-driven decision, as the funnel itself is being reimagined from end (discovery) to end (decision). This fundamentally changes how success is defined and how we all, in advertising, evaluate the totality of that process.
It’s also worth thinking philosophically here, rather than strictly from a channel perspective. How does an agent think? How does that compare to a human? That question steers everything downstream. To measure influence or outcomes, you first have to understand how the agent operates. Everything else follows from there.
Credibility drives discovery
Agents aren’t just assessors. They evaluate information, compare sources and actively determine what’s credible based on signals across reviews, social media and online sentiment.
When brand claims don’t align with real consumer experiences, that gap can be surfaced and acted upon by agents. If a hotel claims to be “family-friendly,” but reviews complain about a lack of kid-friendly amenities, an agent may discount that claim when recommending options to a parent.
This degree of scrutiny will only increase as agent systems become more sophisticated. In a recent paper on multi-agent systems, engineers at Kennesaw State University note that one agent can support a claim while others challenge it, helping improve accuracy in decision-making. As these systems evolve, brand claims will become subject to a more rigorous process of evaluation, challenge and validation.
This means measurement also has to evolve to account for credibility, to consider not just whether a brand was seen but whether its claims were understood, validated and trusted by the systems influencing discovery. Brands that prove their value with clarity and transparency will gain an advantage. In this way, trust moves from a softer brand attribute to a more concrete ranking signal.
Measuring two audiences, not one
Just as marketers learned to differentiate performance across channels and devices, they will now need to understand performance across human and agent-driven interactions. We will also need to rethink attribution models to better evaluate how agent interactions steer consideration, shape decisions and drive outcomes.
For years, the industry treated non-human activity as something to avoid, filter or block. In an agentic world, that approach is too blunt. We need to distinguish between fraudulent or wasteful non-human traffic and the trusted agents acting on behalf of real people. This will require new language, because “invalid traffic” does not capture the full reality of what non-human activity means today.
The next era of advertising won’t just be about reaching people; it will be about understanding and reaching new AI systems that decide what people ultimately see, trust and buy.
“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Follow DoubleVerify and AdExchanger on LinkedIn.
For more articles featuring Mark Zagorski, click here.
