What happened
HUMAN’s October 2026 State of Agentic Traffic, published November 9, reports that detected agentic traffic was 2.6 times September’s level, a 157.5% monthly increase, and about eight times August’s. The company says activity had been roughly flat through the summer before accelerating in September and October. Its monthly benchmark draws on the HUMAN Defense Platform and analysis by the company’s Satori Threat Intelligence Team.
Meta’s Muse represented 59% of October agent requests, up 19 percentage points from September; HUMAN says Muse’s traffic nearly quadrupled month over month. It generated 3.7 times as many requests as ChatGPT Agent, the next-largest agent in the report. OpenAI products collectively accounted for about 30%: ChatGPT Agent had 15.9%, Codex Browser 8.8%, and the Codex Chrome extension 5.2%. ChatGPT Agent’s share fell from 28.5%, even as its request volume rose 44%. Codex Browser traffic nearly tripled, while the Chrome extension moved from sixth to fourth place. HUMAN reports smaller October shares for Perplexity Comet (3.6%), Anthropic’s Claude Chrome extension (3.5%), Google Agent (0.8%) and xAI’s Grok Agent (0.7%).
The report also describes a change in browser-agent shares. Perplexity Comet fell from 41% in August to 9.4% in September and 3.6% in October; HUMAN says its request volume declined about 30% from August. Genspark’s agentic browser dropped out of the top ten. HUMAN contrasts those standalone consumer browsers with agents built into platforms consumers already use, while noting that developer-focused Codex Browser traffic grew. The report says Google launched Google Agent on October 8 and xAI launched Grok Agent on August 11; Claude Duet entered research preview in October, with very low traffic while access remained limited.
By destination, ecommerce accounted for about 60% of agent traffic, travel 19.3%, and media 19.1%. SaaS was 1.2%, though its traffic nearly quadrupled in October and was about 15 times its August level. Agents mostly researched rather than transacted: 85.2% of requests went to product and search pages, while 1.9% went to checkout and payment. Checkout’s share fell from 2.4% in September, but its absolute request volume rose 117% month over month. HUMAN counts each agent action on a site, such as loading a page or running a search, as one request.
What the archive adds
New Runtime’s earlier coverage on November 12 reported that Meta Muse began signing some web traffic with Web Bot Auth on November 6, while Grok Bot traffic shifted toward Cloudflare IP space (The Latest in Agent Traffic). HUMAN’s report is a new fact on that broader story: it quantifies October request shares and destinations, rather than describing those later traffic-signing and routing changes.
Trend signal
This does not yet fit a pattern New Runtime tracks. HUMAN’s figures are a vendor-observed snapshot of agent requests and market mix; they do not establish a durable change in agent protocols, identity, or the software boundaries websites expose. The distinction between identifying an agent and inferring what it intends to do remains important: HUMAN categorizes traffic by destination endpoint, not operator intent.
Why now
HUMAN says the sharp increase followed the arrival of Meta’s Muse in September, with growth accelerating in October. It also points to the upcoming Black Friday and Cyber Monday period: the next monthly report is expected to show how agents behaved during those shopping events. That timing makes the reported rise in checkout requests especially worth checking against later measurements.
Connections to other events
The earlier New Runtime note on November 12 described Meta Muse signing some traffic and a shift in Grok Bot’s route. Together, the two accounts point to distinct questions for site operators: how much agent activity arrives, and whether they can identify its operator or route. HUMAN’s October report addresses volume and destination mix, not the subsequent signing and routing details.
Why it matters
For a team running an online store, a surge in agent requests may affect analytics, site capacity, fraud controls, and how product information is presented to automated visitors. But the headline growth rate is not a conversion rate: HUMAN counts requests, not people, sessions, intent, or completed purchases. The report’s 85.2% share for product and search pages, compared with 1.9% for checkout and payment, suggests that much of the observed activity is still discovery. The 117% increase in checkout request volume is a reason to watch transaction paths, not proof that agent-driven sales are already large.
The reported concentration also matters operationally. If one agent accounts for most observed requests, changes in that agent’s behavior could materially alter a monitored site’s traffic profile. Teams should avoid treating HUMAN’s shares as universal market shares, however: they describe requests detected across applications protected by HUMAN. A useful response is to establish site-specific baselines, distinguish requests from sessions and transactions, and test whether existing bot, analytics, and checkout systems classify agent traffic consistently.
Limits and uncertainty
The benchmark covers agentic traffic detected across applications protected by HUMAN, not a census of the web. Its figures are requests, and one request is not equivalent to a user, session, purchase, or distinct agent. HUMAN combines behavioral signals, user-agent attribution, and publisher-level integration where available; the report does not quantify the accuracy or coverage of those identification methods for each product. Sector and route labels describe destination endpoints rather than an agent operator’s intent. The published shares therefore cannot establish how many people use each agent, how much commerce agents cause, or whether the observed mix represents sites outside HUMAN’s network.
Skill to build now
Add a weekly dashboard view that separates agent requests from human sessions and completed transactions. Break agent activity down by destination—product/search, account, authentication, and checkout—rather than relying on a single traffic total. If checkout requests rise without completed orders, inspect that path for bot controls, errors, and abandoned or blocked requests before treating the increase as commercial demand.
Reader takeaway
HUMAN’s data points to fast growth in agent requests, concentrated in Meta Muse and mostly directed at product discovery. For builders, the useful signal is not the headline multiplier alone: measure agent traffic on your own site, keep requests distinct from buyers, and watch whether activity reaches checkout and completes a transaction.