AI and advertising traffic: when the click does not come from a person

According to Gartner, by 2026, the volume of traditional Google searches will fall by 25%, largely replaced by AI chatbots and virtual agents. This means that a growing part of the traffic that comes to your ads is no longer born from a person who types a search, but from an automatic system that acts on its behalf. If the campaigns continue to treat this traffic as if it were always human, they risk optimizing on increasingly less reliable signals. It is therefore necessary to recognize and correctly interpret this new type of traffic generated by AI agents.

AI agents are not passive viewers of the web, but actors who use mouse, keyboard and even visual browsers to interact with pages, compile forms, book services and complete entire tasks. Companies like OpenAI, Anthropic and Microsoft have developed systems that allow these agents to do exactly what a person would do, but with different precision and speed. This means that agents not only read ads, but often click them, evaluate them and arrive until conversion. In the most recent tests, some of these agents showed very similar behavior to that of a real consumer, with even very high conversion rates.

Not necessarily. Some conversions generated by AI agents reach success rates of 100%, a data that looks great but hides a problem: that click or that tracked conversion can look perfect for Google Ads or analytics, but without reflecting a true human purchase intention. An agent could simply compare offers, test the experience of the site or simulate a purchase path, without any real will to buy. For this reason it is important not to blindly trust the number of conversions, but also to understand the nature of the traffic that generates them.

Some studies have shown that AI agents tend to ignore purely visual elements, such as GIF or text-only banner, preferring buttons and links semantically clear in the page code. When a call to action like "Order Now" was just a text in an image, no agent clicked, while turning it into a real HTML button clicks increased significantly. The agents also show a tendency to "satisficing": they stop as soon as they find a pretty good answer, without exploring the page as a human user would often. This behaviour very different from the human one has an important impact on how digital advertising is read and interpreted.

In 2024 automated traffic exceeded human traffic for the first time, with bad bots representing 37% of total internet traffic. According to HUMAN Security, the traffic generated specifically by AI agents increased by 7,851% only in the last year, a data that gives the idea of the speed with which this phenomenon is expanding. These agents often browse product pages, log in to accounts and cross checkout flows, acting very similar to a real user. This makes the boundary between legitimate automation and traffic more subtle to monitor carefully.

Yes, the ability of AI agents to complete economically sensitive actions, such as subscription offers or participate in promotions, opens delicate scenarios for those who manage advertising campaigns. It is not necessarily aft or direct abuse, but an automation that crosses commercial funnels and generates conversions that, for advertising systems, seem completely genuine. The real challenge becomes therefore to distinguish between useful automations, risky automations and even fraudulent behaviors. For this reason, monitoring and interpreting this type of traffic is becoming increasingly important to protect the advertising budget.

In part, yes, but not entirely. Many AI agents openly declare their identity through the browser user agent they use, so technically it is possible to recognize them and distinguish them from human traffic. The problem is that this statement is not mandatory or always reliable: an agent can also disguise itself as a normal browser. Also when identity is declared correctly, it remains to decide what to do with that traffic, because simply excluding it completely risks penalizing legitimate automations, such as price comparisons authorized by users themselves.

No, and that's exactly the delicate point. Some AI agents act on behalf of a real user who has given a precise instruction, such as "find me the cheapest flight" or "compared with these three products", so they still represent a genuine interest, though mediated by a software. Others are simple scraping bots or automated tests without any real user behind. For advertising campaigns the difference is huge: in the first case there is a potential economic value, in the second is only a cost without return.

Advertising platforms are still adapting their systems to this relatively new phenomenon, so at the moment they offer partial protection. The existing filters are mainly designed to recognize traditional bots and basic automated traffic, not necessarily the most advanced AI agents that better imitate human behavior. With time it is likely that Google and Meta will strengthen specific controls on this type of traffic, but in the meantime a dedicated monitoring helps to bridge the gap and protect the budget immediately.

Not entirely, and it would also be counterproductive to do so. Some AI agents represent legitimate automations, requested directly by a real user interested in your product or service, and blocking them a priori would mean losing that value. The most effective solution is not a total block, but an analysis that distinguishes useful agents from harmful or irrelevant to your business, evaluating the context and final outcome of the interaction, not only its automated origin.