AI is moving from helping marketers analyze campaigns to making media-buying decisions itself. That promises faster optimization and lower operational costs, but it also creates a serious problem: an algorithm can optimize for performance without fully understanding the quality of the traffic behind it.
Artificial intelligence has already changed how performance marketers work. Campaign teams use AI to generate creatives, analyze audiences, predict conversion rates and identify underperforming placements. The next stage is considerably more significant. AI agents are beginning to participate directly in media buying, making decisions that previously required a human campaign manager: where to allocate budget, which traffic sources to scale, when to reduce spending and how aggressively to optimize toward a particular result.
This shift is usually presented as an efficiency story, and there is a good reason for that. A machine can analyze far more campaign data than a human media buyer, react to changes almost instantly and continuously optimize thousands of variables across multiple sources. For advertisers operating at scale, the potential savings in time and operational costs are substantial.
But there is another side to autonomous media buying that is receiving much less attention. If an AI system is given a performance target and allowed to optimize toward it, how does it know that the traffic producing those results is actually good traffic?
That question was raised this week in a MarTech article published on October 8, featuring the perspective of PropellerAds on the security risks surrounding agentic media buying. The concern is straightforward: AI can become extremely good at finding traffic that appears to perform well according to the metrics it has been given, while still failing to recognize signals that an experienced media buyer or security system might consider suspicious.
AI optimizes what it can measure
Performance advertising has always been driven by optimization. Give a media buyer a target CPA, conversion rate or return on advertising spend, and the natural objective is to identify the sources producing the strongest numbers and move more budget toward them.
AI can perform the same process much faster. If one source begins delivering conversions at €12 while another produces them at €25, an autonomous system can detect the difference and immediately increase spending on the cheaper source. At first glance, this is exactly what advertisers want.
The problem appears when the metric being optimized does not tell the full story.
A cheap conversion does not necessarily mean high-quality traffic. A placement generating impressive click-through rates may be using misleading creative. A source delivering large volumes of traffic may include suspicious redirects, manipulated landing pages, malware exposure, invalid activity or other characteristics that are difficult to detect through campaign performance data alone.
An AI agent focused primarily on reaching a predefined KPI can therefore make a perfectly logical decision based on incomplete information. If a questionable source produces better numbers, the system may not simply continue buying it. It may actively scale it.
This is what makes agentic media buying fundamentally different from using AI as an analytics tool. When AI recommends an action, a person can still examine the context before making a decision. When an autonomous agent is authorized to act, the time between identifying an opportunity and spending more money on it can become extremely short.
Speed, one of AI's greatest advantages, can also accelerate mistakes.
Bad traffic is becoming an AI problem
The advertising industry is already familiar with invalid traffic, click fraud, domain spoofing, malicious redirects and other forms of manipulation. None of these problems appeared because of artificial intelligence. What changes with agentic buying is the speed and scale at which an automated system can respond to the signals these environments produce.
Traditional fraud prevention generally focuses on identifying whether traffic is legitimate. Agentic media buying introduces an additional layer: whether an autonomous buyer understands enough about the environment to avoid scaling traffic that technically satisfies its optimization criteria but creates other risks for the advertiser.
The issue is particularly relevant in performance marketing and affiliate advertising, where campaigns frequently operate across multiple networks, publishers, placements and traffic types. Buyers may test a large number of sources simultaneously and move budget rapidly toward those producing results. An AI agent is extremely well suited to this environment, but only if the information available to it extends beyond price and conversion performance.
This means future media-buying systems may need to evaluate a broader set of signals before making allocation decisions. Traffic origin, redirect behavior, landing-page integrity, historical source reputation, abnormal conversion patterns, brand-safety indicators and fraud signals could become part of the decision-making process alongside CPA, CTR and ROAS.
Without those additional signals, autonomous optimization risks becoming a system that is very efficient at buying exactly what the advertiser asked for, without understanding whether the advertiser would actually want it. We made a similar point about optimization targets in Media Buying Is Moving From Dashboards to Prompts.
The problem is not AI. It is the information AI receives.
There is an important distinction here. The conclusion should not be that autonomous media buying is inherently unsafe or that humans are naturally better at identifying traffic quality. Human media buyers also make poor decisions, miss fraud signals and scale sources that later turn out to be problematic. Much of the modern advertising ecosystem is already too complex for manual oversight alone.
The more useful question is what information an AI agent needs before it should be trusted to make independent purchasing decisions.
An agent that sees only campaign performance will optimize campaign performance. An agent that also receives reliable information about traffic provenance, source reputation, fraud risk and advertising quality can make a much more informed decision.
This shifts part of the responsibility from the AI itself to the infrastructure around it. Autonomous buying requires verification systems, transparent traffic data and clearly defined rules determining which sources an agent is allowed to use. The quality of an AI buyer will ultimately depend not only on the intelligence of the model but also on the quality of the market information available to it.
That principle becomes increasingly important as advertising moves toward more autonomous systems. AI does not magically eliminate the information asymmetry that has existed between traffic buyers and sellers for years. If anything, it can magnify it. A human buyer might hesitate before dramatically increasing spending on an unfamiliar source. An automated system may have no reason to hesitate if every metric it sees is positive.
Transparency becomes more valuable when the buyer is a machine
This development has an interesting implication for traffic marketplaces. As media buying becomes more automated, structured and verifiable information about traffic sources becomes more valuable, not less.
A human media buyer can ask questions, research a publisher, speak with a traffic provider and use experience to judge whether an opportunity looks credible. AI agents need those same signals in a form they can interpret systematically.
Where does the traffic originate? What format is being offered? Which geographies are available? How was the audience acquired? Is the traffic exclusive or resold? What historical performance information exists? Are there restrictions on verticals, creatives or landing pages? What verification is available? These are the same questions behind lead provenance in insurance and lending.
These questions have always mattered, but autonomous buying changes their importance. In a market increasingly operated by machines, transparency needs to become part of the data itself.
For Traffberry, this is closely connected to the logic behind a traffic marketplace. The objective is not simply to create another place where advertisers can purchase volume. The more useful role is to make traffic opportunities understandable: allowing buyers and sellers to see what is being offered, communicate directly and evaluate whether a source is appropriate before money starts moving. That is why every listing on the marketplace is moderated, and every fact in the ad networks directory links to its source.
As AI becomes more involved in that process, marketplaces may eventually need to make the same information readable not only by people but also by buying agents.
AI could make the traffic market better, but only with the right guardrails
There is a positive side to this development. Autonomous buying could eventually improve parts of the traffic market that are currently inefficient. AI agents can compare opportunities continuously, identify unusual patterns faster than human teams and potentially incorporate fraud and quality signals into purchasing decisions at enormous scale.
The same technology capable of scaling questionable traffic in seconds could also stop spending on suspicious sources in seconds.
The difference lies in what the system is designed to optimize.
If autonomous media buying is built primarily around obtaining the lowest possible conversion cost, advertisers may discover that efficiency and quality do not always move together. If traffic verification, security and source transparency become part of the optimization framework, AI could make buying both faster and more disciplined. The scale of the problem is real: as we noted in AI Can Buy Traffic. But Can It Tell Good Traffic From Bad Traffic?, a large share of measured open-programmatic app traffic is classified as invalid.
This is likely to become one of the defining questions of the next stage of AdTech. The industry has spent years developing algorithms capable of deciding which advertisement should be shown to which user and at what price. Now those algorithms are evolving into agents capable of making broader commercial decisions on behalf of advertisers.
The technology to automate more of the buying process is arriving quickly. The infrastructure required to verify what those systems are buying needs to develop just as quickly.
Because when AI can move budgets faster than any human media buyer, discovering that the traffic was bad after the money has already been spent is no longer good enough.
The next competitive advantage in automated advertising may not be the AI that buys traffic fastest. It may be the AI that knows which traffic not to buy.
Sources and further reading
- MarTech: Agentic media buying needs malware guardrails: A PropellerAds view (October 8, 2026)