State of Agentic Traffic is HUMAN’s monthly benchmark on how AI agents are showing up across the web. Each edition reports on agent mix, sector destinations, and activity patterns observed across the Human Defense Platform, extending findings from the annual 2026 State of AI Traffic & Cyberthreat Benchmark Report.
This is a monthly report by HUMAN’s Satori Threat Intelligence Team, which monitors all agentic traffic and other observed automated traffic and produces insights surrounding the trends, the developments, and the threat landscape.
Key Findings
AI agent traffic by operator
In August, Comet Browser remained the largest source of agentic traffic, accounting for 40.7% of total volume. However, this total reflects a significant decline of nearly seven percentage points from July. The gap was largely filled by the resurgence of ChatGPT Agent, which rose from 6% in July to 19.8% in August after months of continuous decline, capturing nearly one-fifth of all agentic traffic.
One possible explanation is that ChatGPT Agent absorbed traffic from Atlas, which OpenAI sunset on August 9, 2026. OpenAI, which operated both platforms, shifted Atlas’s browser-based agentic capabilities into the official ChatGPT desktop app. Accordingly, August should be Atlas’s final appearance in this report finishing with 4.1%.
Following Atlas’s sunset, another OpenAI agentic platform—Codex—appeared, launching strongly with 7.3% of total traffic. Codex is OpenAI’s AI software-engineering agent and includes both an in-app browser and a Chrome extension. Altogether (ChatGPT Agent, Codex, and Atlas), in August, OpenAI products account for ~31% of agentic traffic.
Among the remaining platforms, the Claude Chrome Extension remained relatively stable at 22%, despite a slight decline. Genspark continued its downward trend, falling from 2.4% to 1.6%. Browser-agent infrastructure platforms also declined: Browserbase fell from 2.6% to 2%, while Browser Use Cloud decreased from 1% to 0.7%.
Industry shares of AI agent traffic
After being overtaken by the Media sector last month, E-commerce businesses reclaimed the top position from the “which industry received the most agentic traffic” perspective, increasing its share from 42% to 46%. This represents a 38.5% increase in traffic volume compared with July. Media moved in the opposite direction, declining from 43.5% to 36.2%. Travel remained in third place and grew modestly, from 13.4% to 16.3%.
Most smaller sectors also increased in absolute volume, although their relative shares of total traffic remained limited. Streaming and Gaming nearly tripled in size but still accounted for just 0.3% of total agentic traffic. SaaS grew by 63% and represented 0.67%, while Financial Services increased by 31.5% and accounted for 0.4%.
What agents are doing: Agentic traffic by page category
Discovery and research continued to drive the vast majority of agentic activity, with this focus becoming even more pronounced in August. Product and search routes—including browsing listings, reading articles, and executing queries—reached an all-time high of 80% of total activity, reinforcing the ongoing trend of agents being used primarily for research rather than transaction completion.
User-account interactions continued July’s upward trend, rising from 6.4% to 7.1% and indicating a growing willingness among users to connect agents to their personal accounts. In contrast, authentication activity declined from July’s peak of 7% to 5.5%, returning to pre-July levels. Together, these trends may suggest that users still prefer to sign in themselves but are becoming less reluctant to let agents manage their accounts once authenticated.
Other route categories remained largely unchanged from July, including content engagement at 2.2%, miscellaneous activity at 3%, and checkout and payment at 2.1%.
Methodology
Data reflects agentic traffic observed across HUMAN Sightline Cyberfraud Defense during August 2026. Agent identification combines behavioral signal analysis, user-agent attribution, and publisher-level integration where available. Sector and route categorization reflects the destination endpoint, not the agent operator’s intent.
See agentic traffic on your own site
The benchmarks above show agentic activity observed by HUMAN Sightline Cyberfraud Defense and AgenticTrust. On your own site, most analytics tools can’t distinguish AI agents from human visitors, let alone identify specific agents or classify their intent.
AgenticTrust is HUMAN’s trust and control layer for AI agents. It gives security, fraud, commerce, and marketing teams continuous visibility into every agent session, classifies intent as the session unfolds, and enforces adaptive policies based on what agents are actually doing. Book a demo to learn more.
Get visibility and control over AI agents and agentic browsers on your website.
