AI agents are becoming a measurable source of web traffic, and their rapid growth is beginning to raise an uncomfortable question for advertising: what is a visit worth when the visitor is not human?
For most of the internet's history, the concept of traffic has been fairly straightforward. A website visit represented a person arriving somewhere online to read, search, compare, shop or simply browse. Advertising technology became increasingly sophisticated in measuring what happened around that visit, but the basic assumption remained the same: somewhere behind the impression, click or session was a human being.
That assumption is becoming less reliable.
On October 5, HUMAN Security published its latest analysis of agentic web activity, reporting that traffic generated by AI agents increased by 209.7% in September compared with August 2026. The data comes from HUMAN Sightline Cyberfraud Defense and therefore represents traffic observed across HUMAN's own infrastructure rather than the entire internet, but the scale and speed of the increase are significant. In practical terms, the company observed more than three times as much agentic activity in September as it had only one month earlier.
The composition of that traffic is also changing quickly. Meta Muse, which launched on September 8, accounted for 40.34% of the agentic traffic HUMAN observed during the month, immediately becoming the largest source in its dataset. ChatGPT Agent followed with 28.6%, while OpenAI's Codex products collectively represented approximately 11%. Perplexity's Comet Browser accounted for 9.4% and the Claude Chrome Extension for 7.1%.
These percentages should not be confused with shares of overall internet traffic. What they show is the distribution within the agentic activity HUMAN was able to observe. Even with that qualification, the numbers point to something important for the advertising industry: AI is moving from being a technology that produces information for users to one that increasingly navigates the web on their behalf.
AI agents are already entering the commercial internet
The destinations of these agents make the trend particularly relevant to AdTech. HUMAN found that 60.4% of observed agentic traffic went to e-commerce websites, compared with 21.1% to media and 17.1% to travel. Around 81% of agent activity involved product discovery and search-related pages, while checkout and payment pages accounted for only 2.4%.
For now, this suggests that AI agents are much more active in the research stage of the customer journey than at the point of transaction. They search, compare and collect information, while humans still complete most purchases. But that separation may not last very long. An assistant asked to find the best laptop under €2,000, compare several hotels in Barcelona or identify the cheapest suitable flight can already perform a significant part of the work that previously required a consumer to visit multiple websites personally. As agentic systems become capable of completing more complex actions, research and transaction are likely to move closer together, a shift we looked at in The Click Is No Longer the Conversion.
This creates an interesting problem for the way the advertising market thinks about traffic. A traditional website session is valuable partly because it creates an opportunity to influence a person. The user sees the page, encounters advertising, responds to design and messaging, clicks through offers and leaves behavioural signals that can later be measured. An AI agent behaves differently. It may visit dozens of pages, extract exactly the information it needs and make a recommendation without ever engaging with the advertising formats designed for human attention.
The traffic is real, but its economic meaning is different.
Imagine a person visiting an online store, viewing several products and leaving without making a purchase. Now imagine an AI assistant accessing the same store, comparing its prices with several competitors and recommending one of its products to a user who eventually buys it. Existing analytics may provide a much clearer picture of the first journey than the second, even though the second could have contributed directly to revenue.
This is where the growth of agentic traffic starts to become more than an AI story. It becomes an attribution, measurement and monetization problem.
The advertising industry may need a new definition of valuable traffic
Digital advertising has spent years building systems around impressions, clicks, sessions, conversions and increasingly sophisticated attribution models. Agentic browsing introduces another participant into that system: software acting on behalf of a consumer.
Not every visit generated by an AI agent should automatically be considered valuable. Agents can crawl pages, retrieve information, automate repetitive processes or generate activity with little commercial significance. Treating one thousand AI visits as equivalent to one thousand human visitors would therefore make little sense. At the same time, classifying all automated traffic as undesirable or worthless could become equally problematic if some of those agents are researching products with genuine purchase intent on behalf of real customers.
The distinction matters because the advertising industry has traditionally divided traffic largely into human activity and invalid or automated activity. Agentic AI complicates that model. A bot scraping a website, an AI assistant comparing insurance products for a customer and a human browsing those same products are three very different types of traffic, even though all three can appear as requests hitting the same website infrastructure.
For publishers, this could eventually become a monetization issue. If AI agents increasingly consume information from websites while bypassing advertising designed for human audiences, publishers will need to understand how much value those visits create and whether existing business models capture any of it. For advertisers, the challenge is attribution: if an AI system discovers a product, evaluates competing offers and influences the final purchase, traditional click-based measurement may capture only part of the customer journey.
There is also a broader implication for the traffic market. Buyers increasingly need to understand not simply how much traffic they are purchasing, but where it originated, what generated it and what kind of intent sits behind it. The familiar distinction between "good traffic" and "bad traffic" may become considerably more nuanced when automated traffic can represent both meaningless machine activity and a highly relevant consumer acting through an AI intermediary. We covered the quality side of this in AI Can Buy Traffic. But Can It Tell Good Traffic From Bad Traffic?
A new layer of the traffic economy is emerging
It is too early to claim that AI agents are transforming the economics of the entire internet. HUMAN's September data covers a specific observed environment, and one month of rapid growth does not establish a permanent trajectory. What it does provide is an unusually clear signal that agentic activity is becoming large enough to measure and commercially relevant enough to watch.
For the traffic industry, this development deserves particular attention because it challenges one of the assumptions on which the market has been built: that understanding traffic ultimately means understanding the human visitor behind it. In an agentic internet, the visitor, the decision-maker and the buyer may no longer be the same entity.
That could eventually require a new layer of traffic classification. Advertisers may want to know whether a visit came from a human, a search crawler, an autonomous AI agent or an assistant operating with explicit commercial intent. Publishers may need different rules for how those visitors access and use their content. Measurement companies will have to determine how AI-assisted discovery contributes to conversions, while marketplaces connecting buyers and sellers of traffic will increasingly need transparency about what is actually being traded, as AI agents start to act as media buyers themselves.
For Traffberry, this is the part of the trend worth watching. The future of the traffic market will not be determined simply by how many visits a source can deliver. Origin, quality, intent and transparency are becoming more important, and AI agents add an entirely new dimension to all four. On the Traffberry marketplace, buyers can compare traffic offers by GEO, format and pricing model, and check ad network profiles where every fact links to its source.
The internet is not suddenly becoming "post-human", and human attention remains the foundation of the advertising economy. But another type of visitor has clearly arrived, and it is growing quickly. The next question for AdTech is no longer whether AI will generate web traffic. It is how the industry will decide what that traffic is worth.
Sources
- HUMAN Security: State of Agentic Traffic: Agentic Traffic Tripled in September as Meta's Muse Debuts at No. 1 (October 5, 2026)
- HUMAN Security: Is Meta's Muse the first mass-market AI agent? What our traffic data shows (September 2026)
- HUMAN Security: The 2026 State of AI Traffic and Cyberthreat Benchmark Report