State of Agentic Traffic: Agentic Traffic Tripled in September as Meta’s Muse debuts at No. 1

State of Agentic Traffic is HUMAN’s monthly benchmark on how AI agents are showing up across the web, reporting on agent mix, sector destinations, and activity patterns observed across the Human Defense Platform. Each edition builds on findings from the annual State of AI Traffic & Cyberthreat Benchmark Report and is produced with analysis from HUMAN’s Satori Threat Intelligence Team, which investigates agentic traffic and other automated traffic to identify emerging trends, developments, and threats.

September brought the largest single-month increase in agentic traffic HUMAN has recorded this year. Total volume rose 209.7% over August. The largest gain in the prior seven months was 43.4%, in March. Passed and blocked requests both climbed with it.

Two developments account for most of the growth.

The first is Meta’s Muse, which launched on September 8 and ended the month as the largest agent by traffic share. Muse reaches consumers through Facebook, Instagram, Messenger, and WhatsApp, and its free tier includes a browser agent by default, where most other agents put browser automation behind a paid subscription. That combination of distribution and low cost of entry may make Muse the first mass-market moment for agentic AI.

The second is OpenAI. ChatGPT Agent’s traffic reached 4.5 times its August level, Codex Browser’s reached 3.3 times, and Codex Chrome Extension entered the top 10 in its first month.

Blocked requests also reached a new high in September, and Muse is the main reason. HUMAN Sightline Cyberfraud Defense blocks Muse traffic by default. Customers fully control which agents are allowed on their sites, and each one must choose to allow Muse. Because the agent is so new, many are still blocking it.

The top 10 agent rankings changed significantly in September. Meta’s new agent, Muse, launched on September 8 and quickly rose to the top spot, capturing 40.3% of total traffic and surpassing many well-established agents along the way. This report includes a spotlight review of Muse’s traffic patterns during its first month.

Another meteoric rise this month came from ChatGPT Agent, whose share grew from 19.8% to 28.6%, moving it into second place. While this month-over-month growth in relative share is impressive, it does not fully reflect the agent’s growth in absolute traffic. In September, ChatGPT Agent’s traffic reached 4.5 times its August level. Even allowing for the traffic it absorbed from Atlas, which was sunset in August, this is a large jump.

These two rapid rises pushed the long-reigning Comet Browser from the top spot into third place. Its share plunged from 40.7% to just 9.4%, while its absolute traffic declined by 28.3% from its August level, continuing the downward trend that began in August.

OpenAI’s Codex products collectively accounted for 11% of total traffic, surpassing Comet Browser’s share. Codex Browser’s relative share remained broadly stable at 7.7%, compared with 7.3% in August; however, its absolute traffic grew to 3.3 times its August level. Codex Chrome Extension made its first appearance in this report, entering the list in sixth place with a 3.4% share. Altogether, OpenAI products accounted for 39.7% of agentic traffic—nearly matching Meta’s Muse alone.

Claude Chrome Extension remained stable in absolute traffic, but its relative share fell sharply from 22% in August to 7.1% in September. Genspark’s months-long downward trend also continued, with its share declining from 1.6% to just 0.5%.

In absolute terms, Browserbase and AgentCore each grew to 2.6 times their August levels. Their relative shares remained broadly stable, with Browserbase moving from 2% to 1.7% and AgentCore from 1.2% to 1%. BrowserUse closed this month’s top 10 with a 0.2% share, pushing Hyperbrowser off the list.

The surge in agentic traffic increased absolute traffic across all sectors, and sharply widened the gap between ecommerce and media. Ecommerce traffic grew to four times its August level, making it the dominant sector in September with 60.4% of total agentic traffic, up from 46% in August. Media grew to 1.8 times its August level, but its share fell from 36.2% to 21.1%. 

Travel grew to 3.2 times its August level but remained in third place, with a 17.1% share compared with 16.3% in August. SaaS, financial services, and streaming and gaming each roughly tripled their absolute traffic while maintaining similar relative shares of 0.6%, 0.4%, and 0.3%, respectively.

Although the surge in agentic traffic increased absolute traffic across all page categories, their relative shares remained broadly in line with August, suggesting that users’ patterns of agent use have not materially changed as individual agents come and go. 

Discovery and research continued to dominate: product and search routes, including browsing listings, reading articles, and executing queries, accounted for 81% of total activity, up from 80% in August. This reinforces the ongoing trend of agents being used primarily for research rather than transaction completion. 

After several months of growth, user-account interactions declined, possibly reflecting the increase in miscellaneous agentic requests from 3% to 5.8%. User account management fell from 7.1% to 4%, while authentication declined from 5.5% to 3.1%.

As discussed in the Muse spotlight below, Muse users have not yet widely connected the agent to their personal accounts or authorized it to manage them, likely due to limited initial trust and confidence. Given Muse’s substantial share of overall traffic, increased adoption and user trust could help account-related activity recover in the coming months.

Meanwhile, content engagement rose from 2.2% to 3.6%, while checkout and payment increased slightly from 2.1% to 2.4%.

Meta announced the release of its new AI agent, Muse, on September 8, 2026. HUMAN gained dedicated detection visibility into Muse traffic on September 22, 2026. Our first look at Muse traffic, published September 28th, covered its opening days, where Muse accounted for up to 74% of daily agentic requests. This spotlight extends that analysis through the end of the month. 

By the end of September, Muse was already generating an average of millions of requests per day.

In less than a month, Muse entered HUMAN’s top 10 agent list and peaked at first position, surpassing several long-standing, well-established agents. For comparison, Muse’s traffic size was 1.4 times that of the second agent on the list, ChatGPT Agent. 

Muse’s daily adoption rate (daily delta) reached 27.5%, meaning that, on average, each day during its first month saw 21% more traffic attributed to Muse than the day before.

Compared with overall agentic traffic, Muse traffic was more concentrated in ecommerce and advertising and underrepresented in media and travel.

Ecommerce led both Muse and overall agentic traffic, but accounted for a much larger share of Muse traffic: 76.5% versus 60.4% overall, approximately 1.3 times higher. Advertising represented 4.3% of Muse traffic but hardly registered in overall agentic traffic (0.001%).

Conversely, Muse had a smaller presence in media and travel. Media accounted for 12.4% of Muse traffic versus 21.1% overall, while travel represented 5.9% versus 17.1%. Overall agentic traffic therefore had approximately 1.7 times the media share and 2.8 times the travel share of Muse traffic.

SaaS and streaming and gaming were similarly represented in both groups, at 0.4% and 0.3%, respectively. Financial services, however, accounted for 0.2% of Muse traffic versus 0.02% overall,a 9.6-fold difference.

Muse’s traffic category profile broadly aligned with overall agentic traffic, but it was underrepresented in account-related actions.

Search and product routes accounted for 84.3% of Muse traffic, compared with 81% of overall agentic traffic, indicating that Muse is primarily used for research and product discovery.

Content engagement and checkout and payment activity were broadly comparable to the overall distribution. Content engagement represented 3.5% of Muse traffic versus 3.6% overall, while checkout and payment activity accounted for 2.6% versus 2.4%.

The largest gap appeared in personal account management. User-account routes represented 1.7% of Muse traffic, compared with 4% overall, approximately 2.4 times higher across all agents. Authentication routes accounted for 1.1% of Muse traffic versus 3.1% overall, or approximately 2.8 times Muse’s share.

Given Muse’s substantial contribution to total agentic traffic, its underrepresentation in these categories may be contributing to their declining shares overall. However, as a newly launched agent, Muse may see increased account-related activity over time as users become more familiar with it and develop greater trust in its ability to manage personal accounts.

For more on Muse, watch HUMAN’s on-demand webinar, Meta Muse 101: The Agentic Internet is Here. Now What?

Data reflects agentic traffic observed across HUMAN Sightline Cyberfraud Defense during September 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.

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 identifies and verifies Muse and dozens of other AI agents using proprietary research and digital signatures that cannot be spoofed. 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, down to whether a given agent is allowed to add to cart or log in.

The first step is visibility. To see what Muse and other AI agents are doing across your business, book an AgenticTrust Demo today.