On October 29, Resemble AI launched Resemble Agent Detection, a behavioural model called RAD-S1 that labels each website session as a person or an AI agent. Instead of trusting a visitor’s user-agent string, it watches cursor movement, typing rhythm, scrolling, click timing and browser signals. The claim is simple: a spoofed label does not change the result.
The headline number, and what’s missing
In Resemble AI’s internal benchmark, RAD-S1 identified at least 98.8% of agents built to hide their automation. Six widely used bot tools tested on the same visits caught between 3.1% and 14.8%: BotD 14.8%, Akamai 8.7%, F5 6.8%, Cloudflare Turnstile 4.6%, Google reCAPTCHA 3.6% and Vercel BotID 3.1%.
But the test was run by the company selling the product, and the method leaves several questions open:
- No sample size or session count
- No list of agents tested or rival product versions
- No false-positive rate for human visitors
- No independent review
That third gap is the one marketers should push on hardest. A detector that calls every visitor an agent would score a perfect 100 on Resemble AI’s metric, but it would destroy checkout conversion and customer experience.
The result looks narrower up close
Resemble AI’s own model page shows the advantage is mostly about disguised agents. When automation left normal signs, rivals jumped: BotD scored at least 99.3%, Akamai 99.1%, and RAD-S1’s floor was 98.7%. In the session-takeover category, one unnamed tool caught up to 50.2%. The dramatic gap is real only for agents deliberately built to evade detection.
Why this matters for performance marketing
AI agents now browse product pages, add items and abandon carts. In analytics, an agent that loads a page and leaves can look like a shopper who lost interest. Those sessions feed conversion reports, retargeting audiences and the signals automated bidding learns from. Microsoft Clarity added a bot activity dashboard in January, and vendors have been pitching agent visibility to growth teams since at least April.
Resemble AI’s product does not block traffic by itself. It hands customers a verdict, and the site decides whether to allow the agent, route it to an API or MCP server, restrict sensitive pages or stop destructive actions.
“Detection is the first step, and the companies that can securely embrace authorized agent traffic will win the business those agents bring.” — Zohaib Ahmed, CEO and co-founder, Resemble AI
The practical read
Before you buy the 98.8% claim, ask how often the system mislabels humans. Then audit your own analytics for agent sessions and decide your routing strategy. The product is free for the first 1 million classified sessions, then $1 per 1,000 sessions, though one summary page says 10,000 visitors — so confirm the allowance before you build a business case.
Source: PPC Land
