Meta’s Muse drove 59% of agent requests in HUMAN’s October data

src-4b9eba3d

Meta’s Muse accounted for 59% of agent requests recorded on websites and apps protected by HUMAN Security in October, while agents from Meta and OpenAI together generated nearly 90%, according to HUMAN’s monthly State of Agentic Traffic report. The figures point to a sharp increase in automated browsing activity, concentrated among a small number of providers.

Total agent requests in HUMAN’s measured network were approximately eight times higher than in August and about 2.6 times higher than in September. The report covers requests reaching HUMAN-protected properties, not all internet traffic. It counts each request as an agent action, such as loading a page or conducting a search; it does not count each request as a unique visitor or completed transaction.

Muse expands its lead

Muse’s share of agent requests rose 19 percentage points from September, and its request volume nearly quadrupled. HUMAN reported that the agent generated 3.7 times as many requests as ChatGPT Agent, the next-largest individual agent in the October data. Meta launched Muse on September 8.

OpenAI represented about 30% of requests through three agents ranked in the top 10. ChatGPT Agent’s share fell from 28.5% in September to 15.9% in October, even though its request volume increased 44%. The change illustrates how a growing overall pool of agent activity—and Muse’s faster growth in particular—can reduce an agent’s share while its own volume rises.

OpenAI’s Codex Browser application increased its share from 7.7% to 8.8%, while the Codex Chrome extension rose from 3.3% to 5.2%. The Claude Chrome extension’s share dropped from 7% to 3.5%, although its request volume increased 30%. Google’s Gemini agent, which HUMAN says launched October 8, ranked eighth with 0.8%. xAI’s Grok Agent entered the top 10 in ninth place, at 0.7%.

Standalone agent browsers lose ground

Browser-based agents that operate as standalone products lost share in HUMAN’s tracking. Perplexity’s Comet accounted for 41% of agent requests in August, then fell to 9.4% in September and 3.6% in October. It was the only top-10 agent whose request volume declined during October, with a decrease of approximately 30% since August. Genspark’s browser-based agent no longer appeared in the top 10. OpenAI discontinued its standalone ChatGPT Atlas browser in August.

HUMAN attributed the broader movement toward agents connected to familiar platforms or companies’ other products to users’ existing familiarity and product relationships. The report’s data documents shifts in traffic share; it does not establish why individual users selected a particular agent.

Most activity reached product and search pages

Product and search pages received 85% of agent requests in October, up from 81% the previous month. The share reaching checkout and payment pages declined from 2.4% to 1.9%. HUMAN said agents still browse much more than they buy, while noting that the number of requests reaching checkout has been increasing rapidly.

Those page categories describe where a request arrived, not what action followed. A request to a checkout page does not demonstrate that an order was placed. Ecommerce sites accounted for a stable share of about 60% of agent requests in the period. Travel was the second-largest category, ahead of media for the first time in the report series.

Identification depends on more than signatures

HUMAN said it identifies agent activity using behavioral data, user-agent strings and, when available, publisher-level integrations, while excluding crawler activity. The company said its AgenticTrust system identified and verified Muse through proprietary research and digital-signature validation. The report did not break down what proportion of Muse requests was identified by each method.

Separately, Meta has said Muse’s browsing will appear to websites as user activity. DataDome reported that Muse moved its Web Bot Auth signatures from Meta’s key directory at meta.com to muse.ai between November 7 and November 9. Web Bot Auth signatures can allow a website to verify an agent’s identity. DataDome also reported that, as of November 10, half of Muse requests on its network were unsigned. Requests without signatures cannot be verified through that mechanism, and sites configured to accept signatures tied only to meta.com may not recognize requests using the new directory.

For ecommerce and travel operators, the figures highlight a practical measurement and access-control challenge: a large share of observed agent requests comes from a few providers, while conventional site analytics may not distinguish agents from people. Teams reviewing bot rules should consider which identification methods their protections support and how changes to signature directories affect verification. HUMAN said a forthcoming report would cover agent activity around Black Friday and Cyber Monday, providing another view of how requests develop during a major shopping period.